When measuring_output_stream is used to calculate result's element size
it incorrectly takes into account not only serialized element size, but
a placeholder that ser::qr_partition__rows/qr_partition__static_row__cells
constructors puts in the beginning. Fix it by taking starting point in a
stream before element serialization and subtracting it afterwords.
Fixes#3755
Message-Id: <20180906153609.GJ2326@scylladb.com>
(cherry picked from commit d7674288a9)
The code uses incorrect output stream in case only digest is requested
and thus getting incorrect data size. Failing to correctly account
for static row size while calculating digest may cause digest mismatch
between digest and data query.
Fixes#3753.
Message-Id: <20180905131219.GD2326@scylladb.com>
(cherry picked from commit 98092353df)
When the list of values in the IN list of a single column contains
duplicates, multiple executors are activated since the assumption
is that each value in the IN list corresponds to a different partition.
this results in the same row appearing in the result number times
corresponding to the duplication of the partition value.
Added queries for the in restriction unitest and fixed with a bad result check.
Fixes#2837
Tests: Queries as in the usecase from the GitHub issue in both forms ,
prepared and plain (using python driver),Unitest.
Signed-off-by: Eliran Sinvani <eliransin@scylladb.com>
Message-Id: <ad88b7218fa55466be7bc4303dc50326a3d59733.1534322238.git.eliransin@scylladb.com>
(cherry picked from commit d734d316a6)
Fixes a regression introduced in
9e88b60ef5, which broke the lookup for
prefetched values of lists when a clustering key is specified.
This is the code that was removed from some list operations:
std::experimental::optional<clustering_key> row_key;
if (!column.is_static()) {
row_key = clustering_key::from_clustering_prefix(*params._schema, prefix);
}
...
auto&& existing_list = params.get_prefetched_list(m.key().view(), row_key, column);
Put it back, in the form of common code in the update_parameters class.
Fixes#3703
* https://github.com/duarten/scylla cql-list-fixes/v1:
tests/cql_query_test: Test multi-cell static list updates with ckeys
cql3/lists: Fix multi-cell static list updates in the presence of ckeys
keys: Add factory for an empty clustering_key_prefix_view
(cherry picked from commit 6937cc2d1c)
_value_views is the authoritative data structure for the
client-specified values. Indeed, the ctor called
transport::request::read_options() leaves _values completely empty.
In query_options::prepare() we were, however, using _values to
associated values to the client-specified column names, and not
_value_views. Fix this by using _value_views instead.
As for the reasons we didn't see this bug earlier, I assume it's
because very few drivers set the 0x04 query options flag, which means
column names are omitted. This is the right thing to do since most
drivers have enough information to correctly position the values.
Fixes#3688
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20180814234605.14775-1-duarte@scylladb.com>
(cherry picked from commit a4355fe7e7)
In previous versions of Fedora, the `crypt_r` function returned
`nullptr` when a requested hashing algorithm was not supported.
This is consistent with the documentation of the function in its man
page.
As of Fedora 28, the function's behavior changes so that the encrypted
text is not `nullptr` on error, but instead the string "*0".
The info pages for `crypt_r` clarify somewhat (and contradict the man
pages):
Some implementations return `NULL` on failure, and others return an
_invalid_ hashed passphrase, which will begin with a `*` and will
not be the same as SALT.
Because of this change of behavior, users running Scylla on a Fedora 28
machine which was upgraded from a previous release would not be able to
authenticate: an unsupported hashing algorithm would be selected,
producing encrypted text that did not match the entry in the table.
With this change, unsupported algorithms are correctly detected and
users should be able to continue to authenticate themselves.
Fixes#3637.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <bcd708f3ec195870fa2b0d147c8910fb63db7e0e.1533322594.git.jhaberku@scylladb.com>
(cherry picked from commit fce10f2c6e)
The calculation consists of several parts with preemption point between
them, so a table can be added while calculation is ongoing. Do not
assume that table exists in intermediate data structure.
Fixes#3636
Message-Id: <20180801093147.GD23569@scylladb.com>
(cherry picked from commit 44a6afad8c)
"
The problem happens under the following circumstances:
- we have a partially populated partition in cache, with a gap in the middle
- a read with no clustering restrictions trying to populate that gap
- eviction of the entry for the lower bound of the gap concurrent with population
The population may incorrectly mark the range before the gap as continuous.
This may result in temporary loss of writes in that clustering range. The
problem heals by clearing cache.
Caught by row_cache_test::test_concurrent_reads_and_eviction, which has been
failing sporadically.
The problem is in ensure_population_lower_bound(), which returns true if
current clustering range covers all rows, which means that the populator has a
right to set continuity flag to true on the row it inserts. This is correct
only if the current population range actually starts since before all
clustering rows. Otherwise, we're populating since _last_row and should
consult it.
Fixes#3608.
"
* 'tgrabiec/fix-violation-of-continuity-on-concurrent-read-and-eviction' of github.com:tgrabiec/scylla:
row_cache: Fix violation of continuity on concurrent eviction and population
position_in_partition: Introduce is_before_all_clustered_rows()
(cherry picked from commit 31151cadd4)
"
Changes made:
- switched the test to use do_with_cql_env_thread due to lack of SEASTAR_TEST_CASE_THREAD macro
- imported make_local_key() from master, needed for the database_test to pass
"
* tag 'tgrabiec/disable-min-max-sstable-filtering-v1-branch-2.1' of github.com:tgrabiec/scylla:
Merge "Disable sstable filtering based on min/max clustering key components" from Tomasz
tests: simple_schema: Generate local keys form make_pkeys()
tests: Import make_local_key() from master
"
With DateTiered and TimeWindow, there is a read optimization enabled
which excludes sstables based on overlap with recorded min/max values
of clustering key components. The problem is that it doesn't take into
account partition tombstones and static rows, which should still be
returned by the reader even if there is no overlap in the query's
clustering range. A read which returns no clustering rows can
mispopulate cache, which will appear as partition deletion or writes
to the static row being lost. Until node restart or eviction of the
partition entry.
There is also a bad interaction between cache population on read and
that optimization. When the clustering range of the query doesn't
overlap with any sstable, the reader will return no partition markers
for the read, which leads cache populator to assume there is no
partition in sstables and it will cache an empty partition. This will
cause later reads of that partition to miss prior writes to that
partition until it is evicted from cache or node is restarted.
Disable until a more elaborate fix is implemented.
Fixes#3552Fixes#3553
"
* tag 'tgrabiec/disable-min-max-sstable-filtering-v1' of github.com:tgrabiec/scylla:
tests: Add test for slicing a mutation source with date tiered compaction strategy
tests: Check that database conforms to mutation source
database: Disable sstable filtering based on min/max clustering key components
(cherry picked from commit e1efda8b0c)
On some build environment we may want to limit number of parallel jobs since
ninja-build runs ncpus jobs by default, it may too many since g++ eats very
huge memory.
So support --jobs <njobs> just like on rpm build script.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <20180425205439.30053-1-syuu@scylladb.com>
(cherry picked from commit 782ebcece4)
There is a bug in incremental_selector for partitioned_sstable_set, so
until it is found, stop using it.
This degrades scan performance of Leveled Compaction Strategy tables.
Fixes#3513. (as a workaround)
Introduced: 2.1
Message-Id: <20180613131547.19084-1-avi@scylladb.com>
(cherry picked from commit aeffbb6732)
ec2_snitch::gossiper_starting() calls for the base class (default) method
that sets _gossip_started to TRUE and thereby prevents to following
reconnectable_snitch_helper registration.
Fixes#3454
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1528208520-28046-1-git-send-email-vladz@scylladb.com>
(cherry picked from commit 2dde372ae6)
Fixes argument misquoting at $SRPM_OPTS expansion for the mock commands
and makes the --jobs argument work as supposed.
Signed-off-by: Mika Eloranta <mel@aiven.io>
Message-Id: <20180113212904.85907-1-mel@aiven.io>
(cherry picked from commit 7266446227)
We are currently moving the pointer we acquired to the segment inside
the lambda in which we'll handle the cycle.
The problem is, we also use that same pointer inside the exception
handler. If an exception happens we'll access it and we'll crash.
Probably #3440.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20180518125820.10726-1-glauber@scylladb.com>
(cherry picked from commit 596a525950)
This parameter is not available on recent Red Hat kernels or on
non-Red Hat kernels (it was removed on 3.10.0-772.el7,
RHBZ 1455932). The presence of the parameter on kernels that don't
support it cause the module load to fail, with the result that the
storage is not available.
Fix by removing the parameter. For someone running an older Red Hat
kernel the effect will be that discard is disabled, but they can fix
that by updating the kernel. For someone running a newer kernel, the
effect will be that they can access their data.
Fixes#3437.
Message-Id: <20180516134913.6540-1-avi@scylladb.com>
(cherry picked from commit 3b8118d4e5)
Dropping a user type requires that all tables using that type also be
dropped. However, a type may appear to be dropped at the same time as
a table, for instance due to the order in which a node receives schema
notifications, or when dropping a keyspace.
When dropping a table, if we build a schema in a shard through a
global_schema_pointer, then we'll check for the existence of any user
type the schema employs. We thus need to ensure types are only dropped
after tables, similarly to how it's done for keyspaces.
Fixes#3068
Tests: unit-tests (release)
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20180129114137.85149-1-duarte@scylladb.com>
(cherry picked from commit 1e3fae5bef)
When 'always_set_home' is specified on /etc/sudoers pbuilder won't read
.pbuilderrc from current user home directory, and we don't have a way to change
the behavor from sudo command parameter.
So let's use ~root/.pbuilderrc and switch to HOME=/root when sudo executed,
this can work both environment which does specified always_set_home and doesn't
specified.
Fixes#3366
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1523926024-3937-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit ace44784e8)
There is a race between cql connection closure and notifier
registration. If a connection is closed before notification registration
is complete stale pointer to the connection will remain in notification
list since attempt to unregister the connection will happen to early.
The fix is to move notifier unregisteration after connection's gate
is closed which will ensure that there is no outstanding registration
request. But this means that now a connection with closed gate can be in
notifier list, so with_gate() may throw and abort a notifier loop. Fix
that by replacing with_gate() by call to is_closed();
Fixes: #3355
Tests: unit(release)
Message-Id: <20180412134744.GB22593@scylladb.com>
(cherry picked from commit 1a9aaece3e)
Empty partition keys are not supported on normal tables - they cannot
be inserted or queried (surprisingly, the rules for composite
partition keys are different: all components are then allowed to be
empty). However, the (non-composite) partition key of a view could end
up being empty if that column is: a base table regular column, a
base table clustering key column, or a base table partition key column,
part of a composite key.
Fixes#3262
Refs CASSANDRA-14345
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20180403122244.10626-1-duarte@scylladb.com>
(cherry picked from commit ec8960df45)
start node 1 2 3
shutdown node2
shutdown node1 and node3
start node1 and node3
nodetool removenode node2
clean up all scylla data on node2
bootstrap node2 as a new node
I saw node2 could not bootstrap stuck at waiting for schema information to compelte for ever:
On node1, node3
[shard 0] gossip - received an invalid gossip generation for peer 127.0.0.2; local generation = 2, received generation = 1521779704
On node2
[shard 0] storage_service - JOINING: waiting for schema information to complete
This is becasue in nodetool removenode operation, the generation of node1 was increased from 0 to 2.
gossiper::advertise_removing () calls eps.get_heart_beat_state().force_newer_generation_unsafe();
gossiper::advertise_token_removed() calls eps.get_heart_beat_state().force_newer_generation_unsafe();
Each force_newer_generation_unsafe increases the generation by 1.
Here is an example,
Before nodetool removenode:
```
curl -X GET --header "Accept: application/json" "http://127.0.0.1:10000/failure_detector/endpoints/" | python -mjson.tool
{
"addrs": "127.0.0.2",
"generation": 0,
"is_alive": false,
"update_time": 1521778757334,
"version": 0
},
```
After nodetool revmoenode:
```
curl -X GET --header "Accept: application/json" "http://127.0.0.1:10000/failure_detector/endpoints/" | python -mjson.tool
{
"addrs": "127.0.0.2",
"application_state": [
{
"application_state": 0,
"value": "removed,146b52d5-dc94-4e35-b7d4-4f64be0d2672,1522038476246",
"version": 214
},
{
"application_state": 6,
"value": "REMOVER,14ecc9b0-4b88-4ff3-9c96-38505fb4968a",
"version": 153
}
],
"generation": 2,
"is_alive": false,
"update_time": 1521779276246,
"version": 0
},
```
In gossiper::apply_state_locally, we have this check:
```
if (local_generation != 0 && remote_generation > local_generation + MAX_GENERATION_DIFFERENCE) {
// assume some peer has corrupted memory and is broadcasting an unbelievable generation about another peer (or itself)
logger.warn("received an invalid gossip generation for peer {}; local generation = {}, received generation = {}",ep, local_generation, remote_generation);
}
```
to skip the gossip update.
To fix, we relax generation max difference check to allow the generation
of a removed node.
After this patch, the removed node bootstraps successfully.
Tests: dtest:update_cluster_layout_tests.py
Fixes#3331
Message-Id: <678fb60f6b370d3ca050c768f705a8f2fd4b1287.1522289822.git.asias@scylladb.com>
(cherry picked from commit f539e993d3)
The partition tombstone is not part of a mutation_fragment in the old
streamed_mutation, so it was not scattered correctly by fragment_scatterer.
This causes test failures if the mutations to be scattered have a partition
tombstone.
Fix by calling consume(tombstone) directly. This isn't nice, but the code
is dead anyway.
"
This fixes an abort in an sstable reader when querying a partition with no
clustering ranges (happens on counter table mutation with no live rows) which
also doesn't have any static columns. In such case, the
sstable_mutation_reader will setup the data_consume_context such that it only
covers the static row of the partition, knowing that there is no need to read
any clustered rows. See partition.cc::advance_to_upper_bound(). Later when
the reader is done with the range for the static row, it will try to skip to
the first clustering range (missing in this case). If clustering_ranges_walker
tells us to skip to after_all_clustering_rows(), we will hit an assert inside
continuous_data_consumer::fast_forward_to() due to attempt to skip past the
original data file range. If clustering_ranges_walker returns
before_all_clustering_rows() instead, all is fine because we're still at the
same data file position.
Fixes#3304.
"
* 'tgrabiec/fix-counter-read-no-static-columns' of github.com:scylladb/seastar-dev:
tests: mutation_source_test: Test reads with no clustering ranges and no static columns
tests: simple_schema: Allow creating schema with no static column
clustering_ranges_walker: Stop after static row in case no clustering ranges
(cherry picked from commit 054854839a)
This change is backported from 092f2e659c.
Previously, the sharded permissions cache was only accessible to the
implementation of `auth::service` in `auth/service.cc`. The intention
was that invoking `auth::service::get_permissions` on shard `k` would
query the cache on shard `k`, which would in turn depend on
`auth::service` on shard k to check for superuser status.
The problem is in `auth::service::start`.
`seastar::sharded<auth::permissions_cache>::start` is invoked with
`*this` of shard 0, causing all instances of the cache to reference the
same object.
I wasn't able to locally reproduce errors or crashes due to this bug
when I compiled a release build of Scylla. However, running a debug
build meant that the glorious `seastar::debug_shared_ptr_counter_type`
quickly saved the day with its checks that `seastar::shared_ptr` isn't
being misused.
To eliminate this problem, we move ownership of a single instance of
`auth::permissions_cache` to a single instance of `auth::service`. When
`auth::service` is sharded, so is the permissions cache.
I verified interactively that no assertions failed in debug mode with
this change.
Fixes#3296.
Tests: unit (debug, release)
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <280a889f551180db1c00d8a80eddf85b2ff0ac60.1521696176.git.jhaberku@scylladb.com>
In gossiper::handle_major_state_change() we set the endpoint_state for
a particular endpoint and replicate the changes to other cores.
This is totally unsynchronized with the execution of
gossiper::evict_from_membership(), which can happen concurrently, and
can remove the very same endpoint from the map (in all cores).
Replicating the changes to other cores in handle_major_state_change()
can interleave with replicating the changes to other cores in
evict_from_membership(), and result in an undefined final state.
Another issue happened in debug mode dtests, where a fiber executes
handle_major_state_change(), calls into the subscribers, of which
storage_service is one, and ultimately lands on
storage_service::update_peer_info(), which iterates over the
endpoint's application state with deferring points in between (to
update a system table). gossiper::evict_from_membership() was executed
concurrently by another fiber, which freed the state the first one is
iterating over.
Fixes#3299.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20180318123211.3366-1-duarte@scylladb.com>
(cherry picked from commit 810db425a5)
If there are a lot of ranges, e.g., num_tokens=2048, 10 ranges per
stream plan will cause tons of stream plan to be created to stream data,
each having very few data. This cause each stream plan has low transfer
bandwidth, so that the total time to complete the streaming increases.
It makes more sense to send a percentage of the total ranges per stream
plan than a fixed ranges.
Here is an example to stream a keyspace with 513 ranges in
total, 10 ranges v.s. 10% ranges:
Before:
[shard 0] range_streamer - Bootstrap with 127.0.0.1 for
keyspace=system_traces, 510 out of 513 ranges: ranges = 51
[shard 0] range_streamer - Bootstrap with ks for keyspace=127.0.0.1
succeeded, took 107 seconds
After:
[shard 0] range_streamer - Bootstrap with 127.0.0.1 for
keyspace=system_traces, 510 out of 513 ranges: ranges = 10
[shard 0] range_streamer - Bootstrap with ks for keyspace=127.0.0.1
succeeded, took 22 seconds
Message-Id: <a890b84fbac0f3c3cc4021e30dbf4cdf135b93ea.1520992228.git.asias@scylladb.com>
(cherry picked from commit 9b5585ebd5)
This reverts commit f792c78c96.
With the "Use range_streamer everywhere" (7217b7ab36) series,
all the user of streaming now do streaming with relative small ranges
and can retry streaming at higher level.
Reduce the time-to-recover from 5 hours to 10 minutes per stream session.
Even if the 10 minutes idle detection might cause higher false positive,
it is fine, since we can retry the "small" stream session anyway. In the
long term, we should replace the whole idle detection logic with
whenever the stream initiator goes away, the stream slave goes away.
Message-Id: <75f308baf25a520d42d884c7ef36f1aecb8a64b0.1520992219.git.asias@scylladb.com>
(cherry picked from commit ad7b132188)
"
Refs #2692Fixes#3246
The current restricting algorithm [1] restricts the active-reader queue
based on the memory consumption of the existing active readers. When
this memory consumption is above the limit new readers are not admitted.
The inactive reader queue on the other hand has a fixed length.
This caused performance regressions on two workloads:
* read-only: since the inactive-reader queue length is severly limited
(compared to the previous situation) reads will timeout at loads
comfortably handled before.
* mixed: since the memory consumption happens only at admission time
(already created active readers are not limited) memory consumption
growed significantly causing problems when compactions kicked in.
The solution is to reintroduce the old limit of 100 active concurrent
user-reads while still keeping the memory-based limit as well. For
workloads that don't consume a lot of memory or on large boxes with lots
of memory the count-based limit will be reached which is reverting to the
old well-known behaviour. For memory-hungry workloads or on small boxes
with little memory the memory based-limit will kick in sooner avoiding
memory overconsumption.
[1] introduced by bdbbfe9390
"
* 'restricted-reader-dual-limit/v3-backport-2.1' of https://github.com/denesb/scylla:
Modify unit tests so that they test the dual-limits
Use the reader_concurrency_semaphore to limit reader concurrency
Add reader_concurrency_semaphore
Add reader_resource_tracker param to mutation_source
mv reader_resource_tracker.hh -> reader_concurrency_semaphore.hh
This semaphore implements the new dual, count and memory based active
reader limiting. As purely memory-based limiting proved to cause
problems on big boxes admitting a large number of readers (more than any
disk could handle) the previous count-based limit is reintroduced in
addition to the existing memory-based limit.
When creating new readers first the count-based limit is checked. If
that clears the memory limit is checked before admitting the reader.
reader_conccurency_semaphore wraps the two semaphores that implement
these limits and enforces the correct order of limit checking.
This class also completely replaces the restricted_reader_config struct,
it encapsulates all data and related functinality of the latter, making
client code simpler.
Soon, reader_resource_tracker will only be constructible after the
reader has been admitted. This means that the resource tracker cannot be
preconstructed and just captured by the lambda stored in the mutation
source and instead has to be passed in along the other parameters.
In preparation to reader_concurrency_semaphore being added to the file.
The reader_resource_tracker is really only a helper class for
reader_concurrency_semaphore so the latter is better suited to provide
the name of the file.
This patch takes a modified version of the Ubuntu 14.04 housekeeping
service script and uses it in Docker to validate the current version.
To disable the version validation, pass the --disable-version-check flag
when running the container.
Message-Id: <20180220161231.1630-1-amnon@scylladb.com>
(cherry picked from commit edcfab3262)
Since we splited scylla-housekeeping service to two different services for systemd, we don't share same service name between systemd and upstart anymore.
So handle it independently for each distribution, try to install
/etc/init/scylla-housekeeping.conf on Ubuntu 14.04.
Fixes#3239
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1519852659-10688-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit 101e909483)
Debian and ubuntu list files come in two variations.
The housekeeping should support both.
This patch change the regexp that match the os in the repository file.
After the introduction of the second list variation, the os name can be in the middle of the path not only at the end.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20180227092543.19538-1-amnon@scylladb.com>
(cherry picked from commit 57d46c6959)
swap_tree() doesn't change the color of the header, and becasue header
was not initialized, it is undefined (can be both red or black). One
problem this causes is that algo::is_header() expects the header to be
always red. It is used by unlink(), which for trees which have a black
header would infinite-loop.
The fix is to initialize the header.
Fixes#3242.
Message-Id: <1519815091-13111-1-git-send-email-tgrabiec@scylladb.com>
(cherry picked from commit 30635510a2)
The test inserts some values with a TTL of 1 second and then
reads them back expecting them not to be expired yet. That may not
always be the case if the machine is slow and we are running in the
debug mode. Increasising the TTLs by x100 should help avoid these
false positives.
Message-Id: <20180219133816.17452-1-pdziepak@scylladb.com>
(cherry picked from commit d97eebe82d)
Commit 6ccd317 introduced a bug in partition_entry::evict() where a
partition entry may be partially evicted if there are non-evictable
snapshots in it. Partially evicting some of the versions may violate
consistency of a snapshot which includes evicted versions. For one,
continuity flags are interpreted realtive to the merged view, not
within a version, so evicting from some of the versions may mark
reanges as continuous when before they were discontinuous. Also, range
tombtsones of the snapshot are taken from all versions, so we can't
partially evict some of them without marking all affected ranges as
discontinuous.
The fix is to revert back to full eviciton, and avoid moving
non-evictable snapshots to cache. When moving whole partition entry to
cache, we first create a neutral empty partition entry and then merge
the memtable entry into it just like we would if the entry already
existed.
Fixes#3215.
Tests: unit (release)
Message-Id: <1518710592-21925-2-git-send-email-tgrabiec@scylladb.com>
(cherry picked from commit b0b57b8143)
The labels in database active_reads metrics where not define correctly.
Label should be created so it will be possible to select based on their
value.
The current implementation define a label "class" with three instances:
user, streaming, system.
Fixes: #2770
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20180123125206.23660-1-amnon@scylladb.com>
(cherry picked from commit a0a1961b6d)
"When moving whole partition entries from memtable to cache, we move
snapshots as well. It is incorrect to evict from such snapshots
though, because associated readers would miss data.
Solution is to record evictability of partition version references (snapshots)
and avoiding eviction from non-evictable snapshots.
Could affect scanning reads, if the reader uses partition entry from
memtable, and the partition is too large to fit in reader's buffer,
and that entry gets moved to cache (was absent in cache), and then
gets evicted (memory pressure). The reader will not see the remainder
of that entry. Found during code review.
Introduced in ca8e3c4, so affects 2.1+
Fixes#3186.
Tests: unit (release)"
* 'tgrabiec/do-not-evict-memtable-snapshots' of github.com:tgrabiec/scylla:
tests: mvcc: Add test for eviction with non-evictable snapshots
mutation_partition: Define + operator on tombstones
tests: mvcc: Check that partition is fully discontinuous after eviction
tests: row_cache: Add test for memtable readers surviving flush and eviction
memtable: Make printable
mvcc: Take partition_entry by const ref in operator<<()
mvcc: Do not evict from non-evictable snapshots
mvcc: Drop unnecessary assignment to partition_snapshot::_version
tests: Use partition_entry::make_evictable() where appropriate
mvcc: Encapsulate construction of evictable entries
(cherry picked from commit 6ccd317c38)
Race condition was introduced by commit 028c7a0888, which introduces chunk offset
compression, because a reading state is kept in the compress structure which is
supposed to be immutable and can be shared among shards owning the same sstable.
So it may happen that shard A updates state while shard B relies on information
previously set which leads to incorrect decompression, which in turn leads to
read misbehaving.
We could serialize access to at() which would only lead to contention issues for
shared sstables, but that can be avoided by moving state out of compress structure
which is expected to be immutable after sstable is loaded and feeded to shards that
own it. Sequential accessor (wraps state and reference to segmented_offset) is
added to prevent at() and push_back() interfaces from being polluted.
Tests: release mode.
Fixes#3148.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20180205192432.23405-1-raphaelsc@scylladb.com>
(cherry picked from commit 09f4ee808f)
These patches deal with the remaining exception safety issues in the
memtable partition range readers. That includes moving the assignment
to iterator_reader::_last outside of allocating section to avoid
problems caused by exception-unsafe assignment operator. Memory
accotuning code is also moved out of the retryable context to improve
the code robustness and avoid potential problems in the future.
Fixes#3172.
* https://github.com/pdziepak/scylla.git memtable-range-read-exception-safety-2.1/v1:
memtable: do not update iterator_reader::_last in alloc section
memtable: do not change accounting state in alloc section
tests/memtable: add more reader exception safety tests
These patches change the memtable reader implementation (in particular
partition_snapshot_reader) so that the existing exception safety
paroblems are fixed, but also in a way that, hopefully, would make it
easier to reason about the error handling and avoid future bugs in that
area.
The main difficulty related to exception safety is that when an
exception is thrown out of an allocating section that code is run again
with increased memory reserved. If the retryable code has side effects
it is very easy to get incorrect behaviour.
In addition to that, entering an allocating section is not exactly cheap
which encourages doing so rarely and having large sections.
The approach taken by this series is to, first, make entering allocating
sections cheaper and then reducing the amount of logic that runs inside
of them to a minimum.
This means that instead of entering a section once per a call to
flat_mutation_reader::fill_buffer() the allocation section is entered
once for each emitted row. The only state modified from within the
section are cached iterators to the current row, which are dropped on
retry. Hopefully, this would make the reader code easier to reason
about.
The optimisations to the allocating sections and managed_bytes
linearised context has successfully eliminated any penalty caused by
much more fine grained allocating sections.
Fixes#3123.
Fixes#3133.
Tests: unit-tests (release)
BEFORE
test iterations median mad min max
memtable.one_partition_one_row 1155362 869.139ns 0.282ns 868.465ns 873.253ns
memtable.one_partition_many_rows 127252 7.871us 15.252ns 7.851us 7.886us
memtable.many_partitions_one_row 58715 17.109us 2.765ns 17.013us 17.112us
memtable.many_partitions_many_rows 4839 206.717us 212.385ns 206.505us 207.448us
AFTER
test iterations median mad min max
memtable.one_partition_one_row 1194453 839.223ns 0.503ns 834.952ns 842.841ns
memtable.one_partition_many_rows 133785 7.477us 4.492ns 7.473us 7.507us
memtable.many_partitions_one_row 60267 16.680us 18.027ns 16.592us 16.700us
memtable.many_partitions_many_rows 4975 201.048us 144.929ns 200.822us 201.699us
./before_sq ./after_sq diff
read 337373.86 353694.24 4.8%
write 388759.99 394135.78 1.4%
* https://github.com/pdziepak/scylla.git memtable-exception-safety-2.1/v1:
flat_mutation_reader: add allocation point in push_mutation_fragment
linearization_context: remove non-trivial operations from fast path
lsa: split alloc section into reserving and reclamation-disabled parts
lsa: optimise disabling reclamation and invalidation counter
mutation_fragment: allow creating clustering row in place
paratition_snapshot_reader: minimise amount of retryable code
memtable: drop memtable_entry::read()
tests/memtable: add test for reader exception safety
Retryable code that has side effects is a recipe for bugs. This patch
reworkds the snapshot reader so that the amount of logic run with
reclamation disabled is minimal and has a very limited side effects.
Moving clustering_row is expensive due to amount of data stored
internally. Adding a mutation_fragment constructor that builds a
clustering_row in-place saves some of that moving.
Most of the lsa gory details are hidden in utils/logalloc.cc. That
includes the actual implementation of a lsa region: region_impl.
However, there is code in the hot path that often accesses the
_reclaiming_enabled member as well as its base class
allocation_strategy.
In order to optimise those accesses another class is introduced:
basic_region_impl that inherits from allocation_strategy and is a base
of region_impl. It is defined in utils/logalloc.hh so that it is
publicly visible and its member functions are inlineable from anywhere
in the code. This class is supposed to be as small as possible, but
contain all members and functions that are accessed from the fast path
and should be inlined.
Allocating sections reserves certain amount of memory, then disables
reclamation and attempts to perform given operation. If that fails due
to std::bad_alloc the reserve is increased and the operation is retried.
Reserving memory is expensive while just disabling reclamation isn't.
Moreover, the code that runs inside the section needs to be safely
retryable. This means that we want the amount of logic running with
reclamation disabled as small as possible, even if it means entering and
leaving the section multiple times.
In order to reduce the performance penalty of such solution the memory
reserving and reclamation disabling parts of the allocating sections are
separated.
Since linearization_context is thread_local every time it is accessed
the compiler needs to emit code that checks if it was already
constructed and does so if it wasn't. Moreover, upon leaving the context
from the outermost scope the map needs to be cleared.
All these operations impose some performance overhead and aren't really
necessary if no buffers were linearised (the expected case). This patch
rearranges the code so that lineatization_context is trivially
constructible and the map is cleared only if it was modified.
Exception safety tests inject a failure at every allocation and verify
whether the error is handled properly.
push_mutation_fragment() adds a mutation fragment to a circular_buffer,
in theory any call to that function can result in a memory allocation,
but in practice that depends on the implementation details. In order to
improve the effectiveness of the exception safety tests this patch adds
an explicit allocation point in push_mutation_fragment().
Fixes#3096
The credentials processing for transitional auth was broken
in ba6a41d, "auth: Switch to sharded service which effectively removed
the "virtualization" of underlying auth in the SASL challenge.
As a quick workaround, add the permissive exception handling to
sasl object as well.
Message-Id: <20180103102724.1083-1-calle@scylladb.com>
(cherry picked from commit 35b9ec868a)
* seastar 8d254a1...af1b789 (3):
> tls_test: Fix echo test not setting server trust store
> tls: Do not restrict re-handshake to client
> tls: Actually verify client certificate if requested
Fixes#3072
The reason sstable key estimation is inaccurate is that it doesn't account that
index sampling is now dynamic.
The estimation is done as follow:
uint64_t get_estimated_key_count() const {
return ((uint64_t)_components->summary.header.size_at_full_sampling + 1) *
_components->summary.header.min_index_interval;
}
The biggest problem is that _components->summary.header.min_index_interval isn't
actually the minimum interval, but instead the default interval value set in the
schema.
So the estimation gets worse the larger the average partition, because the larger
the average partition the lower the index sampling interval.
One of the problems is that estimation has a big influence on bloom filter size,
and so for large partitions we were generating bigger filters than we had to.
From now on, size at full sampling is calculated as if sampling were static
(which was the case until commit 8726ee937d which introduced size-based
sampling), using minimum index as a strict sampling interval.
Tests: units (release)
Fixes#3113.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20180122233612.11147-1-raphaelsc@scylladb.com>
(cherry picked from commit 2c181b69c9)
"_free_segments_in_zones is not adjusted by
segment_pool::reclaim_segments() for empty zones on reclaim under some
conditions. For instance when some zone becomes empty due to regular
free() and then reclaiming is called from the std allocator, and it is
satisfied from a zone after the one which is empty. This would result
in free memory in such zone to appear as being leaked due to corrupted
free segment count, which may cause a later reclaim to fail. This
could result in bad_allocs.
The fix is to always collect such zones.
Fixes#3129
Refs #3119
Refs #3120"
* 'tgrabiec/fix-free_segments_in_zones-leak' of github.com:scylladb/seastar-dev:
tests: lsa: Test _free_segments_in_zones is kept correct on reclaim
lsa: Expose max_zone_segments for tests
lsa: Expose tracker::non_lsa_used_space()
lsa: Fix memory leak on zone reclaim
(cherry picked from commit 4ad212dc01)
The call chain is:
storage_service::on_change() -> storage_service::handle_state_removing()
-> storage_service::restore_replica_count() -> streamer->stream_async()
Listeners run as part of gossip message processing, which is serialized.
This means we won't be processing any gossip messages until streaming
completes.
In fact, there is no need to wait for restore_replica_count to complete
which can take a long time, since when it completes, this node will send
notification to tell the removal_coordinator that the restore process is
finished on this node. This node will be removed from _replicating_nodes
on the removal_coordinator.
Tested with update_cluster_layout_tests.py
Fixes#2886
Message-Id: <8b4fe637dfea6c56167ddde3ca86fefb8438ce96.1516088237.git.asias@scylladb.com>
(cherry picked from commit 5107b6ad16)
Commit 2d5fb9d109 (gms/gossiper: Replicate changes incrementally to
other shards) changes the way we replicate _token_metadata and
endpoint_state_map. Before they are replicated at the same time, after
they are not any more. This causes a shard in NORMAL status can still be
with a empty _token_metadata.
We saw errors:
[shard 12] token_metadata - sorted_tokens is empty in first_token_index!
during CorruptThenRepairNemesis.
Fix by setting the gossip status to NORMAL after replication of
_token_metadata, so that once a node is in NORMAL, we can do repair. The
commit 69c81bcc87 (repair: Do not allow repair until node is in NORMAL
status) prevents the early repair operation by checking if a node is in
NORMAL status.
Fixes#3121
Message-Id: <af6a223733d2e11351f1fa35f59eacfa7d65dd30.1516065564.git.asias@scylladb.com>
(cherry picked from commit 3c8ed255ac)
Problem introduced in 375ed938b4
Also remove redefinition of schema in dummy incremental selector
which is supposed to use the one in base class instead.
Following tests are fixed:
./build/release/tests/mutation_reader_test
./build/release/tests/sstable_test -- -c1
./build/release/tests/row_cache_test
./build/release/tests/cache_flat_mutation_reader_test
./build/release/tests/row_cache_stress_test
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20180111153831.17462-1-raphaelsc@scylladb.com>
On Debian 9, 'pbuilder create' fails because of lack of GPG key for
3rdparty repo, so we need --allow-untrusted on 'pbuilder create' and
'pbuilder update'.
Also, apt-key adv --fetch-keys does not works correctly on it, but we can use
"curl <URL> | apt-key add -" as workaround.
Fixes#3088
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1513797714-18067-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit b68ee98310)
We provided "boost1.63" package for Debian 8 since we couldn't build
"scylla-boost163" package witch is available on Ubuntu14/16, but I fixed the
problem and now we have it for Debian 8 too, so switch to it.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1514220163-25985-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit 51013f561d)
The legacy mutation_reader/streamed_mutation design allowed very easily
to skip the partition merging logic if there was only one underlying
reader that has emitted it.
That optimisation was lost after conversion to flat mutation readers
which has impacted the performance. This patch mostly recovers it by
bypassing most of mutation_reader_merger logic if there is only a single
active reader for a given partition.
The performance regression was introduced in
8731c1bc66 "Flatten the implementation of
combined_mutation_reader".
perf_simple_query -c4 read results (medians of 60):
original regression
before 8731c1 after 8731c1 diff
read 326241.02 300244.09 -8.0%
this patch
before after diff
read 313882.59 325148.05 3.6%
Message-Id: <20180103121019.764-1-pdziepak@scylladb.com>
(cherry picked from commit b4a4c04bab)
The uninitialized session has no peer associated with it yet. There is
no point sending the failed message when abort the session. Sending the
failed message in this case will send to a peer with uninitialized
dst_cpu_id which will casue the receiver to pass a bogus shard id to
smp::submit_to which cases segfault.
In addition, to be safe, initialize the dst_cpu_id to zero. So that
uninitialized session will send message to shard zero instead of random
bogus shard id.
Fixes the segfault issue found by
repair_additional_test.py:RepairAdditionalTest.repair_abort_test
Fixes#3115
Message-Id: <9f0f7b44c7d6d8f5c60d6293ab2435dadc3496a9.1515380325.git.asias@scylladb.com>
(cherry picked from commit 774307b3a7)
Using Materialized Views, if the base table has static columns,
and the update in base table mutates static and non static rows,
the streamed_mutation is stopped before process non static row.
The patch avoids stopping the stream_mutation and adds a test case.
Message-Id: <20171220173434.25091-1-tavares.george@gmail.com>
(cherry picked from commit ceecd542cd)
After 611774b, we're blind again on which sstable caused a compaction
to fail, leaving us with cryptic message as follow:
compaction_manager - compaction failed: std::runtime_error (compressed
chunk failed checksum)
After this change, now both read failure in compaction or regular read
will report the guilty sstable, see:
compaction_manager - compaction failed: std::runtime_error (SSTable reader
found an exception when reading sstable ./data/.../keyspace1-standard1
ka-1-Data.db : std::runtime_error(compressed chunk failed checksum))
Fixes#3006.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20180102230752.14701-1-raphaelsc@scylladb.com>
(cherry picked from commit 4610e994e1)
In version 1.8.3 of JsonCpp shipped with Fedora 27, old FastWriter and
Reader classes from JsonCpp have been deprecated in favour of
newer/better ones: CharReaderBuilder/CharReader and
StreamWriterBuilder/StreamWriter.
This fix uses the new classes where available or resorts to old ones for
older versions of the library.
Fixes#2989
Signed-off-by: Vladimir Krivopalov <vladimir@scylladb.com>
(cherry picked from commit 76775ddf26)
'"The issue is triggered by compaction of sstables of level higher than 0.
The problem happens when interval map of partitioned sstable set stores
intervals such as follow:
[-9223362900961284625 : -3695961740249769322 ]
(-3695961740249769322 : -3695961103022958562 ]
When selector is called for first interval above, the exclusive lower
bound of the second interval is returned as next token, but the
inclusivess info is not returned.
So reader_selector was returning that there *were* new readers when
the current token was -3695961740249769322 because it was stored in
selector position field as inclusive, but it's actually exclusive.
This false positive was leading to infinite recursion in combined
reader because sstable set's incremental selector itself knew that
there were actually *no* new readers, and therefore *no* progress
could be made."
Fixes #2908.'
* 'high_level_compaction_infinite_recursion_fix_v4' of github.com:raphaelsc/scylla:
tests: test for infinite recursion bug when doing high-level compaction
Fix potential infinite recursion when combining mutations for leveled compaction
dht: make it easier to create ring_position_view from token
dht: introduce is_min/max for ring_position
(cherry picked from commit 375ed938b4)
Migrate all the places that used memtable::make_reader to use
memtable::make_flat_reader and remove memtable::make_reader.
* seastar-dev.git haaawk/remove_memtable_make_reader_v2_rebased:
Remove memtable::make_reader
Stop using memtable::make_reader in row_cache_stress_test
Stop using memtable::make_reader in row_cache_test
Stop using memtable::make_reader in mutation_test
Stop using memtable::make_reader in streamed_mutation_test
Stop using memtable::make_reader in memtable_snapshot_source.hh
Stop using memtable::make_reader in memtable::apply
Add consume_partitions(flat_mutation_reader& reader, Consumer consumer)
Add default parameter values in make_combined_reader
Migrate test_virtual_dirty_accounting_on_flush to flat reader
Migrate test_adding_a_column_during_reading_doesnt_affect_read_result
Simplify flat_reader_assertions& produces(const mutation& m)
Migrate test_partition_version_consistency_after_lsa_compaction_happens
flat_mutation_reader: Allow setting buffer capacity
Add next_mutation() to flat_mutation_reader_assertions
cf::for_all_partitions::iteration_state: don't store schema_ptr
read_mutation_from_flat_mutation_reader: don't take schema_ptr
Migrate test_fast_forward_to_after_memtable_is_flushed to flat reader
(cherry picked from commit b0a56a91c2)
'char' and int8_t ('unsigned char') are different types. 'bytes' base type
is int8_t - use the correct type for casting.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
(cherry picked from commit 22ca5d2596)
The following patches contain fixes for skipping to the next parititon
in multi_range_reader and completelty dissable support for fast
forwarding inside a single partition, which is not needed and would only
add unnecessary complexity.
* https://github.com/pdziepak/scylla.git fix-multi_range_reader/v1:
flat_multi_range_mutation_reader: disallow
streamed_mutation::forwarding
flat_multi_range_mutation_reader: clear buffer on next_partition()
tests/flat_multi_range_mutation_reader: test skipping to next
partition
(cherry picked from commit 71cc63dfa6)
In the case there are large number of column families, the sender will
send all the column families in parallel. We allow 20% of shard memory
for streaming on the receiver, so each column family will have 1/N, N is
the number of in-flight column families, memory for memtable. Large N
causes a lot of small sstables to be generated.
It is possible there are multiple senders to a single receiver, e.g.,
when a new node joins the cluster, the maximum in-flight column families
is number of peer node. The column families are sent in the order of
cf_id. It is not guaranteed that all peers has the same speed so they
are sending the same cf_id at the same time, though. We still have
chance some of the peers are sending the same cf_id.
Fixes#3065
Message-Id: <46961463c2a5e4f1faff232294dc485ac4f1a04e.1513159678.git.asias@scylladb.com>
(cherry picked from commit a9dab60b6c)
end_bound was not updated in one of the cases in which end and
end_kind was changed, as a result later merging decision using
end_bound were incorrect. end_bound was using the new key, but the old
end_kind.
Fixes#3083.
Message-Id: <1513772083-5257-1-git-send-email-tgrabiec@scylladb.com>
(cherry picked from commit dfe48bbbc7)
"4b9a34a85425d1279b471b2ff0b0f2462328929c "Merge sstable_data_source
into sstable_mutation_reader" has introduced unintentional changes, some
of them causing excessive read amplification during empty range reads.
The following patches restore the previous behaviour."
* tag 'fix-read-amplification/v1' of https://github.com/pdziepak/scylla:
sstables: set _read_enabled to false if possible
sstables: set _single_partition_read for single parititon reads
(cherry picked from commit 772d1f47d7)
Don't enforce the outgoing connections from the 'listen_address'
interface only.
If 'local_address' is given to connect() it will enforce it to use a
particular interface to connect from, even if the destination address
should be accessed from a different interface. If we don't specify the
'local_address' the source interface will be chosen according to the
routing configuration.
Fixes#3066
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1513372688-21595-1-git-send-email-vladz@scylladb.com>
(cherry picked from commit be6f8be9cb)
"Fixes #3057."
* 'summary_recreation_fixes_v2' of github.com:raphaelsc/scylla:
tests: sstable summary recreation sanity test
sstables: make loading of sstable without summary to work again
sstables: fix summary generation with dynamic index sampling
(cherry picked from commit 11de20fc33)
Currently scylla-housekeeping-daily.service/-restart.service hardcoded
"--repo-files '/etc/yum.repos.d/scylla*.repo'" to specify CentOS .repo file,
but we use same .service for Ubuntu/Debian.
It doesn't work correctly, we need to specify .list file for Debian variants.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1513385159-15736-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit c2e87f4677)
There are some users used original default cluster_name 'Test Cluster',
they will fail to start the node for cluster_name change if they use
new scylla.yaml.
'ScyllaDB Cluster' isn't more beautiful than 'Test Cluster', reset back
to original old to avoid problem for users.
Fixes#3060
Signed-off-by: Amos Kong <amos@scylladb.com>
Message-Id: <8c9dab8a64d0f4ab3a5d6910b87af696c60e5076.1513072453.git.amos@scylladb.com>
(cherry picked from commit b07de93636)
"While aa8c2cbc16 'Merge "Migrate sstables
to flat_mutation_reader" from Piotr' has converted the low-level sstable
reader to the new flat_mutation_reader interface there were still
multiple readers related to sstables that required converting,
including:
- restricted reader
- filtering reader
- single partition sstable reader
This series completes their conversion to the flat stream interface."
* tag 'flat_mutation_reader-sstable-readers/v2' of https://github.com/pdziepak/scylla:
db: convert single_key_sstalbe_reader to flat streams
db: fully convert incremental_reader_selector to flat readers
db: make make_range_sstable_reader() return flat reader
db: make column_family::make_reader() return flat reader
db: make column_family::make_sstable_reader() return a flat reader
filtering_reader: switch to flat mutation fragment streams
filtering_reader: pass a const dht::decorated_key& to the callback
mutation_reader: drop make_restricted_reader()
db: use make_restricted_flat_reader
mutation_reader: convert restricted reader to flat streams
(cherry picked from commit 6cb3b29168)
We have had an issue recently where failed SSTable writes left the
generated SSTables dangling in a potentially invalid state. If the write
had, for instance, started and generated tmp TOCs but not finished,
those files would be left for dead.
We had fixed this in commit b7e1575ad4,
but streaming memtables still have the same isse.
Note that we can't fix this in the common function
write_memtable_to_sstable because different flushers have different
retry policies.
Fixes#3062
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20171213011741.8156-1-glauber@scylladb.com>
(cherry picked from commit 1aabbc75ab)
"Fixes cache reader to not skip over data in some cases involving overlapping
range tombstones in different partition versions and discontinuous cache.
Introduced in 2.0
Fixes #3053."
* tag 'tgrabiec/fix-range-tombstone-slicing-v2' of github.com:scylladb/seastar-dev:
tests: row_cache: Add reproducer for issue #3053
tests: mvcc: Add test for partition_snapshot::range_tombstones()
mvcc: Optimize partition_snapshot::range_tombstones() for single version case
mvcc: Fix partition_snapshot::range_tombstones()
tests: random_mutation_generator: Do not emit dummy entries at clustering row positions
(cherry picked from commit 051cbbc9af)
"Didn't affect any release. Regression introduced in 301358e.
Fixes#3041"
* 'resharding_fix_v4' of github.com:raphaelsc/scylla:
tests: add sstable resharding test to test.py
tests: fix sstable resharding test
sstables: Fix resharding by not filtering out mutation that belongs to other shard
db: introduce make_range_sstable_reader
rename make_range_sstable_reader to make_local_shard_sstable_reader
db: extract sstable reader creation from incremental_reader_selector
db: reuse make_range_sstable_reader in make_sstable_reader
(cherry picked from commit d934ca55a7)
When fast forwarding is enabled and all readers positioned inside the
current partition return EOS, return EOS from the combined-reader
too. Instead of skipping to the next partition if there are idle readers
(positioned at some later partition) available. This will cause rows to
be skipped in some cases.
The fix is to distinguish EOS'd readers that are only halted (waiting
for a fast-forward) from thoose really out of data. To achieve this we
track the last fragment-kind the reader emitted. If that was a
partition-end then the reader is out of data, otherwise it might emit
more fragments after a fast-forward. Without this additional information
it is impossible to determine why a reader reached EOS and the code
later may make the wrong decision about whether the combined-reader as
a whole is at EOS or not.
Also when fast-forwarding between partition-ranges or calling
next_partition() we set the last fragment-kind of forwarded readers
because they should emit a partition-start, otherwise they are out of
data.
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <6f0b21b1ec62e1197de6b46510d5508cdb4a6977.1512569218.git.bdenes@scylladb.com>
(cherry picked from commit 9661769313)
"Convert combined_mutation_reader into a flat_mutation_reader impl. For
now - in the name of incremental progress - all consumers are updated to
use the combined reader through the
mutation_reader_from_flat_mutation_reader adaptor. The combined reader also
uses all it's sub mutation_readers through the
flat_mutation_reader_from_mutation_reader adaptor."
* 'bdenes/flatten-combined-reader-v8' of https://github.com/denesb/scylla:
Add unit tests for the combined reader - selector interactions
Add flat_mutation_reader overload of make_combined_reader
Flatten the implementation of combined_mutation_reader
Add mutation_fragment_merger
mutation_fragment::apply(): handle partition start and end too
Add non-const overload of partition_start::partition_tombstone()
Make combined_mutation_reader a flat_mutation_reader
Move the mutation merging logic to combined_mutation_reader
Remove the unnecessary indirection of mutation_reader_merger::next()
Move the implementation of combined_mutation_reader into mutation_reader_merger
Remove unused mutation_and_reader::less_compare and operator<
(cherry picked from commit 046991b0b7)
"This series makes sstable tests use flat stream interface. The main
motivation is to allow eventual removal of mutation_reader and
streamed_mutation and ensuring that the conversion between the
interfaces doesn't hide any bugs that would be otherwise found."
* tag 'flat_mutation_reader-sstable-tests/v1' of https://github.com/pdziepak/scylla:
sstables: drop read_range_rows()
tests/mutation_reader: stop using read_range_rows()
incremental_reader_selector: do not use read_range_rows()
tests/sstable: stop using read_range_rows()
sstables: drop read_row()
tests/sstables: use read_row_flat() instead of read_row()
database: use read_row_flat() instead of read_row()
tests/sstable_mutation_test: get flat_mutation_readers from mutation sources
tests/sstables: make sstable_reader return flat_mutation_reader
sstable: drop read_row() overload accepting sstable::key
tests/sstable: stop using read_row() with sstable::key
tests/flat_mutation_reader_assertions: add has_monotonic_positions()
tests/flat_mutation_reader_assertions: add produces(Range)
tests/flat_mutation_reader_assertions: add produces(mutation)
tests/flat_mutation_reader_assertions: add produces(dht::decorated_key)
tests/flat_mutation_reader_assertions: add produces(mutation_fragment::kind)
tests/flat_mutation_reader_assertions: fix fast forwarding
(cherry picked from commit 601a03dda7)
Since pbuilder chroot environment does not install CA certificates by default,
accessing https://download.opensuse.org will cause certificate verification
error.
So we need to install it before installing 3rdparty repo GPG key.
Also, checking existance of gpgkeys_curl is not needed, since it's always
not installed since we are running the script in clean chroot environment.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1512517001-27524-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit 8f02967a3b)
We are getting package build error on dh_auto_install which is invoked by
pybuild.
But since we handle all installation on debian/scylla-server.install, we can
simply skip running dh_auto_install.
Fixes#3036
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1512065117-15708-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit 25bc18b8ff)
The universal reference was introduced so we could bind an rvalue to
the argument, but it would have sufficed to make the argument a const
reference. This is also more consistent with the function's other
overload.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171129132758.19654-1-duarte@scylladb.com>
(cherry picked from commit cda3ddd146)
Fix two issues with serializing non-compound range tombstones as
compound: convert a non-compound clustering element to compound and
actually advertise the issue to other nodes.
* git@github.com:duarten/scylla.git rt-compact-fixes/v1:
compound_compact: Allow rvalues in size()
sstables/sstables: Convert non-compound clustering element to compound
tests/sstable_mutation_test: Verify we can write/read non-correct RTs
service/storage_service: Export non-compound RT feature
(cherry picked from commit e9cce59b85)
after 7f8b62bc0b, its move operator and ctor broke. That potentially
leads to error because data_consume_context dtor moves sstable ref
to continuation when waiting for in-flight reads from input stream.
Otherwise, sstable can be destroyed meanwhile and file descriptor
would be invalid, leading to EBADF.
Fixes#3020.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20171129014917.11841-1-raphaelsc@scylladb.com>
(cherry picked from commit f699cf17ae)
This series converts memtable flush reader to the new flat mutation
readers. Just like the scanning reader, flush reader concatenates
multiple partition snapshot readers in order to provide a stream
of all partitions in the memtable.
* https://github.com/pdziepak/scylla.git flat_mutation_reader-memtable-flush/v1
tests/flat_mutation_reader_assertion: add produces_partition()
memtable: make make_flush_reader() return flat_mutation_reader
flat_mutation_reader: add optimised flat_mutation_reader_opt
memtable: switch flush reader implementation to flat streams
tests/memtable: add test for flush reader
(cherry picked from commit 04106b4c96)
These tests now require having the storage service initialize, which
is needed to decide whether correct non-compound range tombstones
should be emitted or not.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171126152921.5199-1-duarte@scylladb.com>
(cherry picked from commit 922f095f22)
The following patches convert sstable writers to use flat mutation
readers instead of the legacy mutation_reader interface.
Writers were already using flat consumer interface and used
consume_flattened_in_thread(), so most of the work was limited to
providing an appropriate equivalent for flat mutation readers.
* https://github.com/pdziepak/scylla.git flat_mutation_reader-sstable-write/v1:
flat_mutation_reader: move consumer_adapter out of consume()
flat_mutation_reader: introduce consume_in_thread()
tests/flat_mutation_reader: test consume_in_thread()
sstables: switch write_components() to flat_mutation_reader
streamed_mutation: drop streamed_mutation_returning()
sstables: convert compaction to flat_mutation_reader
mutation_reader: drop consume_flattened_in_thread()
(cherry picked from commit 596ebaed1f)
This series mainly fixes issues with the serialization of promoted
index entries for non-compound schemas and with the serialization of
range tombstones, also for non-compound schemas.
We lift the correct cell name writing code into its own function,
and direct all users to it. We also ensure backward compatibility with
incorrectly generated promoted indexes and range tombstones.
Fixes#2995Fixes#2986Fixes#2979Fixes#2992Fixes#2993
* git@github.com:duarten/scylla.git promoted-index-serialization/v3:
sstables/sstables: Unify column name writers
sstables/sstables: Don't write index entry for a missing row maker
sstables/sstables: Reuse write_range_tombstone() for row tombstones
sstables/sstables: Lift index writing for row tombstones
sstables/sstables: Leverage index code upon range tombstone consume
sstables/sstables: Move out tombstone check in write_range_tombstone()
sstables/sstables: A schema with static columns is always compound
sstables/sstables: Lift column name writing logic
sstables/sstables: Use schema-aware write_column_name() for
collections
sstables/sstables: Use schema-aware write_column_name() for row marker
sstables/sstables: Use schema-aware write_column_name() for static row
sstables/sstables: Writing promoted index entry leverages
column_name_writer
sstables/sstables: Add supported feature list to sstables
sstables/sstables: Don't use incorrectly serialized promoted index
cql3/single_column_primary_key_restrictions: Implement is_inclusive()
cql3/delete_statement: Constrain range deletions for non-compound
schemas
tests/cql_query_test: Verify range deletion constraints
sstables/sstables: Correctly deserialize range tombstones
service/storage_service: Add feature for correct non-compound RTs
tests/sstable_*: Start the storage service for some cases
sstables/sstable_writer: Prepare to control range tombstone
serialization
sstables/sstables: Correctly serialize range tombstones
tests/sstable_assertions: Fix monotonicity check for promoted indexes
tests/sstable_assertions: Assert a promoted index is empty
tests/sstable_mutation_test: Verify promoted index serializes
correctly
tests/sstable_mutation_test: Verify promoted index repeats tombstones
tests/sstable_mutation_test: Ensure range tombstone serializes
correctly
tests/sstable_datafile_test: Add test for incorrect promoted index
tests/sstable_datafile_test: Verify reading of incorrect range
tombstones
sstables/sstable: Rename schema-oblivious write_column_name() function
sstables/sstables: No promoted index without clustering keys
tests/sstable_mutation_test: Verify promoted index is not generated
sstables/sstables: Optimize column name writing and indexing
compound_compat: Don't assume compoundness
(cherry picked from commit bd1efbc25c)
TTL of 1 second may cause the cell to expire right after we write it,
if the second component of current time changes right after it. Use
larger ttl to avoid spurious faliures due to this.
Message-Id: <1511463392-1451-1-git-send-email-tgrabiec@scylladb.com>
(cherry picked from commit 35e404b1a2)
Prometheus histograms have 3 embedded metrics: count, buckets, and sum.
Currently we fill up count and buckets but sum is left at 0. This is
particularly bad, since according to the prometheus documentation, the
best way to calculate histogram averages is to write:
rate(metric_sum[5m]) / rate(metric_count[5m])
One way of keeping track of the sum is adding the value we sampled,
every time we sample. However, the interface for the estimated histogram
has a method that allows to add a metric while allowing to adjust the
count for missing metrics (add_nano())
That makes acumulating a sum inaccurate--as we will have no values for
the points that were added. To overcome that, when we call add_nano(),
we pretend we are introducing new_count - _count metrics, all with the
same value.
Long term, doing away with sampling may help us provide more accurate
results.
After this patch, we are able to correctly calculate latency averages
through the data exported in prometheus.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20171122144558.7575-1-glauber@scylladb.com>
(cherry picked from commit 6c4e8049a0)
* seastar-dev.git haaawk/flat_reader_remove_read_rows:
sstable_mutation_test: use read_rows_flat instead of read_rows
perf_sstable: use read_rows_flat instead of read_rows
Remove sstable::read_rows
(cherry picked from commit e9ffe36d65)
Introduce sstable::read_row_flat and sstable::read_range_rows_flat methods
and use them in sstable::as_mutation_source.
* https://github.com/scylladb/seastar-dev/tree/haaawk/flat_reader_sstables_v3:
Introduce conversion from flat_mutation_reader to streamed_mutation
Add sstables::read_rows_flat and sstables::read_range_rows_flat
Turn sstable_mutation_reader into a flat_mutation_reader
sstable: add getter for filter_tracker
Move mp_row_consumer methods implementations to the bottom
Remove unused sstable_mutation_reader constructor
Replace "sm" with "partition" in get_next_sm and on_sm_finished
Move advance_to_upper_bound above sstable_mutation_reader
Store sstable_mutation_reader pointer in mp_row_consumer
Stop using streamed_mutation in consumer and reader
Stop using streamed_mutation in sstable_data_source
Delete sstable_streamed_mutation
Introduce sstable::read_row_flat
Migrate sstable::as_mutation_source to flat_mutation_reader
Remove single_partition_reader_adaptor
Merge data_consume_context::impl into data_consume_context
Create data_consume_context_opt.
Merge on_partition_finished into mark_partition_finished
Check _partition_finished instead of _current_partition_key
Merge sstable_data_source into sstable_mutation_reader
Remove sstable_data_source
Remove get_next_partition and partition_header
(cherry picked from commit aa8c2cbc16)
Since we want to support cross building, we shouldn't hardcode GPG file path,
even these files provided on recent version of mock.
This fixes build error on some older build environment such as CentOS-7.2.
Fixes#3002
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1511277722-22917-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit c1b97d11ea)
We get following warning from antlr3 header when we compile Scylla with gcc-7.2:
/opt/scylladb/include/antlr3bitset.inl: In member function 'antlr3::BitsetList<AllocatorType>::BitsetType* antlr3::BitsetList<AllocatorType>::bitsetLoad() [with ImplTraits = antlr3::TraitsBase<antlr3::CustomTraitsBase>]':
/opt/scylladb/include/antlr3bitset.inl:54:2: error: nonnull argument 'this' compared to NULL [-Werror=nonnull-compare]
To make it compilable we need to specify '-Wno-nonnull-compare' on cflags.
Message-Id: <1510952411-20722-2-git-send-email-syuu@scylladb.com>
(cherry picked from commit f26cde582f)
Switch Debian 3rdparty packages to our OBS repo
(https://build.opensuse.org/project/subprojects/home:scylladb).
We don't use 3rdparty packages on dist/debian/dep, so dropped them.
Also we switch Debian to gcc-7.2/boost-1.63 on same time.
Due to packaging issues following packages doesn't renamed our 3rdparty
package naming rule for now:
- gcc-7: renamed as 'xxx-scylla72', instead of scylla-xxx-72.
- boost1.63: doesn't renamed, also doesn't changed prefix to /opt/scylladb
Message-Id: <1510952411-20722-1-git-send-email-syuu@scylladb.com>
(cherry picked from commit ab9d7cdc65)
The exception handling code inspects server state, which could be
destroyed before the handle_exception() task runs since it runs after
exiting the gate. Move the exception handling inside the gate and
avoid scheduling another accept if the server has been stopped.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171116122921.21273-1-duarte@scylladb.com>
(cherry picked from commit 34a0b85982)
These patches convert queries (data, mutation and counter) to flat
mutation readers. All of them already use consume_flattened() to
consume a flat stream of data, so the only major missing thing
was adding support for reversed partitions to
flat_mutation_reader::consume().
* pdziepak flat_mutation_reader-queries/v3-rebased:
flat_mutation_reader: keep reference to decorated key valid
flat_muation_reader: support consuming reversed partitions
tests/flat_mutation_reader: add test for
flat_mutation_reader::consume()
mutation_partition: convert queries to flat_mutation_readers
tests/row_cache_stress_test: do not use consume_flattened()
mutation_reader: drop consume_flattened()
streamed_mutation: drop reverse_streamed_mutation()
(cherry picked from commit 6969a235f3)
For a time range tombstone that was already removed from a tree
is owned by a raw pointer. This doesn't end well if creation of
a mutation fragment or a call to push_mutation_fragment() throw.
Message-Id: <20171121105749.16559-1-pdziepak@scylladb.com>
(cherry picked from commit 1b936876b7)
This series reworks handling of range tombstones in reversed queries
so that they are applied to correct rows. Additionally, the concept
of flipped range tombstones is removed, since it only made it harder
to reason about the code.
Fixes#2982.
* https://github.com/pdziepak/scylla fix-reverse-query-range-tombstone/v2:
streamed_mutation: fix reversing range tombstones
range_tombstone: drop flip()
tests/cql_query_test: test range tombstones and reverse queries
tests/range_tombstone_list: add test for range_tombstone_accumulator
(cherry picked from commit cec5b0a5b8)
Currently flat_mutation_reader_from_mutation_reader()'s
converting_reader will throw std::runtime_error if fast_forward_to() is
called when its internal streamed_mutation_opt is disengaged. This can
create problems if this reader is a sub-reader of a combined reader as the
latter has no way to determine the source of a sub-reader EOS. A reader
can be in EOS either because it reached the end of the current
position_range or because it doesn't have any more data.
To avoid this, instead of throwing we just silently ignore the fact that
the streamed_mutation_opt is disengaged and set _end_of_stream to true
which is still correct.
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <83d309b225950bdbbd931f1c5e7fb91c9929ba1c.1511180262.git.bdenes@scylladb.com>
(cherry picked from commit 8065dca4a1)
Don't std::move() the "query" string inside the parallel_for_each() lambda.
parallel_for_each is going to invoke the given callback object for each element of the range
and as a result the first call of lambda that std::move()s the "query" is going to destroy it for
all other calls.
Fixes#2998
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1511225744-1159-1-git-send-email-vladz@scylladb.com>
(cherry picked from commit 941aa20252)
_current_pi_idx was not reset from advance_to_next_partition(), which
is used when we skip to the next partition before fully consuming
it. As a result, if we try to skip to a clustering position which is
before the index block used by the last skip in the previous
partition, we would not skip assuming that the new position is in the
current block. This may result in more data being read from the
sstable than necessary.
Fixes#2984
Message-Id: <1510915793-20159-1-git-send-email-tgrabiec@scylladb.com>
* seastar 11ad0b1...78cd87f (3):
> Merge "http: Use output stream for files" from Amnon
> tutorial: a section about when_all() and when_all_succeed()
> Merge "Power8 related changes (what's left of them)" from Vlad
As can be seen in one of the traces in #2958, the copy constructor
of index_entry is called in response to std::vector<index_entry>::push_back(index_entry&&).
This is wasteful. Fix by providing the full suite of constructors/assignment operators.
Message-Id: <20171116121608.5580-1-avi@scylladb.com>
"This patch series addresses #2929. The objective is to eliminate global
state from the implementation and use of all access-control functionlity.
I've made every effort to make these patches logically independent and
incremental, but the final patch is big: this was necessary because
eliminating the global instances themselves is an atomic change."
* 'jhk/non_global_auth/v2' of https://github.com/hakuch/scylla:
auth: Switch to sharded service
tracing/trace_keyspace_helper: Use internal `client_state`
auth: Make the QP an explicit dependency
auth: Unify Java class name attributes
auth: Make life-time control more consistent
auth: Move metadata constants
auth: Don't expose internal constant
auth: Extract `permissions_cache`
utils/loading_cache: Include necessary dependency
auth: Fix static constant initialization
auth: Extract `delayed_tasks` from `auth.cc`
This change appears quite large, but is logically fairly simple.
Previously, the `auth` module was structured around global state in a
number of ways:
- There existed global instances for the authenticator and the
authorizer, which were accessed pervasively throughout the system
through `auth::authenticator::get()` and `auth::authorizer::get()`,
respectively. These instances needed to be initialized before they
could be used with `auth::authenticator::setup(sstring type_name)`
and `auth::authorizer::setup(sstring type_name)`.
- The implementation of the `auth::auth` functions and the authenticator
and authorizer depended on resources accessed globally through
`cql3::get_local_query_processor()` and
`service::get_local_migration_manager()`.
- CQL statements would check for access and manage users through static
functions in `auth::auth`. These functions would access the global
authenticator and authorizer instances and depended on the necessary
systems being started before they were used.
This change eliminates global state from all of these.
The specific changes are:
- Move out `allow_all_authenticator` and `allow_all_authorizer` into
their own files so that they're constructed like any other
authenticator or authorizer.
- Delete `auth.hh` and `auth.cc`. Constants and helper functions useful
for implementing functionality in the `auth` module have moved to
`common.hh`.
- Remove silent global dependency in
`auth::authenticated_user::is_super()` on the auth* service in favour
of a new function `auth::is_super_user()` with an explicit auth*
service argument.
- Remove global authenticator and authorizer instances, as well as the
`setup()` functions.
- Expose dependency on the auth* service in
`auth::authorizer::authorize()` and `auth::authorizer::list()`, which
is necessary to check for superuser status.
- Add an explicit `service::migration_manager` argument to the
authenticators and authorizers so they can announce metadata tables.
- The permissions cache now requires an auth* service reference instead
of just an authorizer since authorizing also requires this.
- The permissions cache configuration can now easily be created from the
DB configuration.
- Move the static functions in `auth::auth` to the new `auth::service`.
Where possible, previously static resources like the `delayed_tasks`
are now members.
- Validating `cql3::user_options` requires an authenticator, which was
previously accessed globally.
- Instances of the auth* service are accessed through `external`
instances of `client_state` instead of globally. This includes several
CQL statements including `alter_user_statement`,
`create_user_statement`, `drop_user_statement`, `grant_statement`,
`list_permissions_statement`, `permissions_altering_statement`, and
`revoke_statement`. For `internal` `client_state`, this is `nullptr`.
- Since the `cql_server` is responsible for instantiating connections
and each connection gets a new `client_state`, the `cql_server` is
instantiated with a reference to the auth* service.
- Similarly, the Thrift server is now also instantiated with a reference
to the auth* service.
- Since the storage service is responsible for instantiating and
starting the sharded servers, it is instantiated with the sharded
auth* service which it threads through. All relevant factory functions
have been updated.
- The storage service is still responsible for starting the auth*
service it has been provided, and shutting it down.
- The `cql_test_env` is now instantiated with an instance of the auth*
service, and can be accessed through a member function.
- All unit tests have been updated and pass.
Fixes#2929.
Rather than have all uses of the QP in auth reference global variables,
we supply a QP reference to both the authenticator and authorizer on
construction.
The caller still references a global variable when constructing the
instances, but fixing this problem is a much larger task that is out of
scope of this change.
This change is motivated partly be aesthetics, but more significantly
due to the future work to refactor `auth` into a sharded service. Since
doing so will require writing `auth::auth` from scratch, these
constants (and other common functionality) need a new home.
Using "Meyer's singletons" eliminate the problem of static constant
initialization order because static variables inside functions are
initialized only the first time control flow passes over their
declaration.
Fixes#2966.
This simple task scheduler is used by the auth module to delay metadata
creation until the system is settled.
Extracting it out allows the `auth` module to be refactored into a
sharded service and for other components of `auth` to make use of it.
Fixes#2965.
This patchset prepares sstables read path for flat_mutation_reader.
It cuts some dependencies between classes and replaces
sstables::mutation_reader with ::mutation_reader. This will make it
possible to gradually convert the code to flat_mutation_reader because
we have converters between flat_mutation_reader and ::mutation_reader.
* seastar-dev.git haaawk/flat_reader_prepare_sstables_rebased
Reduce dependencies from mp_row_consumer to sstable_streamed_mutation
Replace sstables::mutation_reader with ::mutation_reader
Remove range_reader_adaptor
Remove sstable_range_wrapping_reader
The wrapper is no longer needed because
read_range_rows returns ::mutation_reader instead of
sstables::mutation_reader and the reader returned from
it keeps the pointer to shared_sstable that was used to
create the reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
The wrapper is no longer needed because
read_range_rows returns ::mutation_reader instead of
sstables::mutation_reader and the reader returned from
it keeps the pointer to shared_sstable that was used to
create the reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This will make migration to flat_mutation_reader much
easier and sstables::mutation_reader is going away with
this migration anyway.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Before this patch mp_row_consumer was using sstable_streamed_mutation
in two ways:
1. Populate sstable_streamed_mutation's buffer with mutation_fragments
2. Advance sstable_streamed_mutation's sstable_data_source to new position.
We can easily reduce those dependencies only to the first one.
This will reduce the coupling between those classes and simplify
the flow of execution.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
"Otherwise, such strategies couldn't behave as expected when it needs to do STCS."
* 'respecting_stcs_options_v2' of github.com:raphaelsc/scylla:
tests: enable twcs test that relied on size-tiered properties
twcs: respect stcs options by forwarding them to stcs method
lcs: forward stcs options to respect them
stcs: make most_interesting_bucket respect size-tiered options
stcs: make most_interesting_bucket respect thresholds
compaction: make size_tiered_most_interesting_bucket static method of stcs class
stcs: introduce new ctor
stcs: make header self contained
stcs: inline function definition so as not to break one definition rule
Broken in f570e41d18.
Not replicating this may cause coordinator to treat a node which is
down as alive, or vice verse.
Fixes regression in dtest:
consistency_test.py:TestAvailability.test_simple_strategy
which was expected to get "unavailable" exception but it was getting a
timeout.
Message-Id: <1510666967-1288-1-git-send-email-tgrabiec@scylladb.com>
Make sure loading_cache::stop() is always called where appropriate:
regardless whether the test failed or there was an exception during the test.
Otherwise a false-alarm use-after-free error may occur.
Fixes#2955
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1510625736-3109-1-git-send-email-vladz@scylladb.com>
"Fixes #2944."
* tag 'tgrabiec/cache-exception-safety-fixes-v2' of github.com:scylladb/seastar-dev:
tests: row_cache: Add test for exception safety of multi-partition scans
tests: row_cache: Add test for exception safety of single-partition reads
tests: mutation_source_tests: Always print the seed
tests: Disable alloc failure injection in test assertions
tests: Avoid needless copies
row_cache: Fix exception safety of cache_entry::read()
row_cache: scanning_and_populating_reader: Fix exception unsafety causing read to skip data
row_cache: partition_range_cursor: Extract valid() and advance_to() from refresh()
cache_streamed_mutation: Add trace-level logging to cache_streamed_mutation
mvcc: Lift noexcept off partition_snapshot_row_weakref assignment/constructors
cache_streamed_mutation: Make advancing to the next range exception-safe
cache_streamed_mutation: Make add_clustering_row_to_buffer() exception-safe
cache_streamed_mutation: Make drain_tombstones() exception-safe
cache_streamed_mutation: Return void from start_reading_from_underlying()
cache_streamed_mutation: Document invariants related to exception-safety
streamed_mutation: Add reserve_one()
lsa: Guarantee invalidated references on allocating section retry
mvcc: partition_snapshot_row_cursor: Mark allocation points
BOOST_TEST_MESSAGE() is not logged by default, and for some tests we
don't want to enable that because it's too noisy. But we need to know
the seed to reproduce a failure, so we better to always print it.
If assignment to _lower_bound in the "_secondary_in_progress = true;"
case in do_read_from_primary() throws due to allocation failure, the
update section will be retried and we will take the not_moved path,
skipping the range which was discontinuous and was supposed to be read
from underlying.
Fix by redoing lookup using _lower_bound in case the section is
retried. When we retry, _primary.valid() will be false. We need to
ensure now that _lower_bound is always valid.
Fixes#2944.
Changing _ck_ranges_curr and _lower_bound should be atomic, either
both fail or both succeed. Currently it could happen that if
position_in_partition::for_range_start() fails, _ck_ranges_curr would
be advanced but _lower_bound not.
We need to maintain the following invariants:
(1) no fragment with position >= _lower_bound was pushed yet
(2) If _lower_bound > mf.position(), mf was emitted
Before this patch (1) could be violated if drain_tombstones() failed
in the middle. (2) could be violated if push_mutation_fragment()
failed.
There is existing code (e.g. use of partition_snapshot_row_cursor in
cache_streamed_mutation) which assumes that references will be
invalidated when bad_alloc is thrown from allocating_section. That is
currently the case because on retry we will attempt memory reclamation
which will invalidate references either through compaction or
eviction. Make this guarantee explicit.
thrift/server.cc:237:6: required from here
thrift/server.cc:236:9: error: cannot call member function ‘void thrift_server::maybe_retry_accept(int, bool, std::__exception_ptr::exception_ptr)’ without object
maybe_retry_accept(which, keepalive, std::move(ex));
gcc version: gcc (GCC) 6.3.1 20161221 (Red Hat 6.3.1-1)
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20171113184537.10472-1-raphaelsc@scylladb.com>
"The following patches convert streaming and repair code to the new
flat mutation reader interface. In particular this involves changing::
- fragment_and_freeze() -- a consumer that fragments and freezes mutations
- checksum computation for repair which until now was using two-level
mutation_reader/streamed_mutation interface
- multi_range_reader -- a mutation reader that automatically fast
forwards to between given partiton ranges"
* tag 'flat_mutation_reader-streaming/v2' of https://github.com/pdziepak/scylla: (24 commits)
mutation_reader: drop multi_range_reader
db: convert make_streaming_reader() to flat_mutation_reader
tests/flat_mutation_reader: add test for multi range reader
tests/flat_mutation_reader_assertions: add fast_forward_to()
tests/simple_schema: add to_ring_positions() helper
flat_mutation_reader: convert flat_multi_range_mutation_reader
flat_mutation_reader: add partition_range_forwarding
flat_mutation_reader: make pop_mutation_fragment() public
flat_mutation_reader: copy multi_range_mutation_reader
streamed_mutation: drop mutation_hasher
tests/flat_mutation_reader: add test for partition checksum
repair: convert partition_checksum::compute_streamed() to flat streams
repair: make partition_hasher consume flat mutation streams
mutation_hasher: copy mutation_hasher to repair.cc
partition_start: make partition_tombstone() const
partition_checksum: introduce compute() for flat_mutation_reader
db: drop single-range make_streaming_reader()
fragment_and_freeze: drop streamed_mutation overload
stream_transfer_task: switch to flat_mutation_reader
tests/flat_mutation_reader: add test for fragment_and_freeze
...
flat_mutation_reader::partition_range_forwarding and
mutation_reader::forwarding are aliases of the same type. The change was
necessary in order to make mutation_reader::forwarding available in
flat_mutation_reader.hh even though it is included by mutation_reader.hh
flat_mutation_reader public interface already exposes low leve
is_buffer_empty() and is_buffer_full() adding pop_mutation_fragment()
will make implementation of intermediate readers more straightforward.
There is a user of fragment_and_freeze() (streaming) that will need
to be able to break the loop Right now, it does that between
streamed_mutation, but that won't be possible after we switch to flat
readers.
partition_entry::read() calls open_version() under standard allocator,
but it may allocate a new partition version if a snapshot already
exists which was created in an earlier phase. Versions are supposed to
be allocated using region's allocator, they will be freed using
region's allocator. LSA will delegate free() to the standard allocator
correctly in this case, but it will subtract from its
_non_lsa_occupancy, assuming the allocation was done through it. This
will corrupt occupancy() for cache region.
Fixes#2948.
Message-Id: <1510229584-14398-1-git-send-email-tgrabiec@scylladb.com>
"Ensure stop() waits for the accept loop to complete to avoid crashes
during shutdown."
* 'thrift-server-stop/v4' of https://github.com/duarten/scylla:
thrift/server: Restore code format
thrift/server: Stopping the server waits for connection shutdown
thrift/server: Abort listeners on stop()
thrift/server: Avoid manual memory management
thrift/server: Add move ctor for connection
thrift/server: Extract retry logic
thrift/server: Retry with backoff for some error types
thrift/server: Retry accept in case of error
This patch ensures the future returned from stop() resolves only when
all connections and listeners are no longer in use.
Fixes#2657Fixes#2942
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
In case of errors like ECONNABORTED, we want to retry accepting
connections. Right now we immediately retry the accept, but in
subsequent patches we introduce a backoff for other types of errors.
We also consider fatal errors like EBADFD, which should not trigger a
retry.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Currently, we use type type of the column as the accumulator when we
average it. This can easily overflow, e.g. (2^31-1)+(3) = overflow.
Fix by using __int128 for the accumulator. It's not standard, but
it's way more efficient and simpler than the alternatives.
Inspired by CASSANDRA-12417, but much simpler due to the availability
of __int128.
Message-Id: <20171112173529.30764-1-avi@scylladb.com>
"This patch adds support for varint and decimal to aggregate functions.
Some other types (like byte or smallint) weren't supported and they are
supported by C*. So their aggregate functions were added as well.
To allow aggregate functions for big_decimal, following methods were added
to big_decimal type:
* Division by int64_t that preservers number of decimal digits.
* Operator += .
* Comparison operators.
Fixes #2842."
* 'danfiala/scylla-2842-send-002' of https://github.com/hagrid-the-developer/scylla:
tests: Add tests for aggregate functions.
tests: Add tests for big_decimal type.
cql3/functions: Add aggregate functions for big_decimal.
utils/big_decimal: Added necessary operators and methods for aggregate functions.
cql3/functions: Add aggregate functions for types for which it is trivial.
Since we started accounting virtual dirty memory we no longer have a cap
on real dirty memory. In most situations that is not needed, since real
dirty will just be at most twice as much as virtual dirty (current
flushing memtable plus new memtable).
However, due to things like cache updates and component flushing we can
end up having a lot of memtables that are virtually freed but not yet
fully released, leading real dirty memory to explode using all the box'
memory.
This patch adds a cap on real dirty memory as well. Because of the
hierarchical nature of region_group, if the parent blocks due to memory
depletion, so will the child (virtual dirty region group).
After that is done, we need to make sure that dirty memory is not seen
as freed until the cache update is done. Until a particular partition is
moved to the cache it is not evictable. As a result we can OOM the
system if we have a lot of pending cache updates as the writes will not
be throttled and memory won't be made available.
This patch pins the memory used by the region as real dirty before the
cache update starts, and unpins it when it is over. In the mean time it
gradually releases memory of the partitions that are being moved to
cache.
I have verified in a couple of workloads that the amount of memory
accounted through this is the same amount of memory accounted through
the memtable flush procedure.
Fixes#1942
* git@github.com:glommer/scylla.git glommer/update-cache-v4:
row_cache: modernize use of seastar threads
mutation_partition: estimate size of partition
memtable: factor out calculation of memtable_entry memory size
memtable: add a method to export memtable's dirty memory manager
dirty_memory_manager: block if we hit the real dirty limit
dirty_memory_manager: add functions to manipulate real dirty
partition: add method to calculate memory size of a partition
row cache: pin real dirty during cache updates.
This patch fixes 'DROP INDEX' CQL statement to also drop the underlying
index view automatically so that we don't leave unused materialized
views behind.
Message-Id: <1510303421-15945-1-git-send-email-penberg@scylladb.com>
Right now, once a region is moved to the cache is no longer visible to
the dirty memory system. Not as real dirty nor virtual dirty.
The problem is that until a particular partition is moved to the cache
it is not evictable. As a result we can OOM the system if we have a lot
of pending cache updates as the writes will not be throttled and memory
won't be made available.
This patch pins the memory used by the region as real dirty before the
cache update starts, and unpins it when it is over. In the mean time it
gradually releases memory of the partitions that are being moved to
cache.
I have verified in a couple of workloads that the amount of memory
accounted through this is the same amount of memory accounted through
the memtable flush procedure.
Fixes#1942
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Once that is added, also add a method to a memtable entry to calculate
the entire size of a memtable entry. Right now we only have one method
to calculate the size minus rows.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
There are times in which we want to add and remove real dirty memory
without impacting virtual dirty. One such example is the cache update
process, where real dirty is the limiting factor.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Since we started accounting virtual dirty memory we no longer have a cap
on real dirty memory. In most situations that is not needed, since real
dirty will just be at most twice as much as virtual dirty (current
flushing memtable plus new memtable).
However, due to things like cache updates and component flushing we can
end up having a lot of memtables that are virtually freed but not yet
fully released, leading real dirty memory to explode using all the box'
memory.
This patch adds a cap on real dirty memory as well. Because of the
hierarchical nature of region_group, if the parent blocks due to memory
depletion, so will the child (virtual dirty region group).
A next step is to add a controller that will increase the priority of
the tasks involving in releasing real dirty memory if we get dangerously
close to the threshold.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
The total size is the sum of two components. Add a method that
does that sum so this code gets easier to reuse.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
In the memtable flusher, we account for the size of a partition as we
read them. However, there are other points in the architecture where we
would like to calculate the size of a partition in a point in which we
are not reading it. One such example is the cache update process.
This patch enhances the mutation_partition adding a method that returns
the total size for this partition.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
For a while now we have an async() function, that simplifies the code by not
needing to issue an explicit join. This patch converts the row cache to use
async() as well, which most of our code already does. Doing so will make
it easier to make changes to update_cache.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
"Implement flat_mutation_reader::consume and add tests for it.
For that implement flat_mutation_reader_from_mutations and
read_mutation_from_flat_mutation_reader."
* 'haaawk/flat_reader_consume_v3' of github.com:scylladb/seastar-dev:
Add tests for flat_mutation_reader::consume
Add tests for flat_mutation_reader utils
Introduce read_mutation_from_flat_mutation_reader
Make mutation_rebuilder streamed_mutation independent
flat_mutation_reader_from_mutation: support multiple mutations
Introduce flat_mutation_reader::consume
Move FlattenedConsumer concept to flat_mutation_reader.hh
"This patchset introduces memtable::make_flat_reader that returns
flat_mutation_reader and converts internal memtable readers into
flat_mutation_readers.
It also introduces some utility methods like make_forwardable and
make_partition_snapshot_flat_reader."
* 'haaawk/flat_reader_memtable_v4' of github.com:scylladb/seastar-dev:
Turn scanning_reader into flat_mutation_reader
Change memtable_entry::read to return flat_mutation_reader
Make iterator_reader independent from mutation_reader
Introduce make_partition_snapshot_flat_reader
Prepare partition_snapshot_flat_reader
Introduce flat_mutation_reader_from_mutation
Prepare flat_mutation_reader_from_mutation
Introduce make_forwardable
Prepare make_forwardable for flat_mutation_reader
Introduce empty_flat_reader
memtable: Introduce make_flat_reader
mutation_rebuilder will be used not only with streamed_mutations
but also with flat_mutation_readers so it's better for it to be
independent from streamed_mutation.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Rename flat_mutation_reader_from_mutation to
flat_mutation_reader_from_mutations.
Make it work with std::vector<mutation> instead of a single
mutation.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This concept will be used both in flat_mutation_reader.hh
and mutation_reader.hh. mutation_reader.hh includes
flat_mutation_reader.hh so we have to move the concept to
make it accessible in both files.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Make it possible for a mutation_source to be created both for sources
that use old mutation_reader and new flat_mutation_reader.
Add tests for flat_mutation_reader::next_partition to run_mutation_source_tests.
* seastar-dev.git 'dev/haaawk/flat_reader_mutation_source_v3':
Remove mutation_reader.hh dependency from flat_mutation_reader.hh
Prepare mutation_source for more than one implementation
Add flat reader mutation source implementation
Add mutation_source::make_flat_mutation_reader
Use mutation_source::make_flat_mutation_reader in tests
Add flat_mutation_reader_assertions
Add test for flat_mutation_reader::next_partition
This commit creates a copy of partition_snapshot_reader
and names it partition_snapshot_flat_reader.
This new class will be turned into a flat_mutation_reader
in the next commit.
The purpose of this commit is to make it easier to review the next
commit.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This is a utility method that will be handy in conversion
from mutation_reader to flat_mutation_reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This commit copies streamed_mutation_from_mutation
from streamed_mutation to flat_mutation_reader
and renames it to streamed_mutation_from_mutation_copy.
This copy will be used as a base for
flat_mutation_reader_from_mutation.
The purpose of this commit is to make it easier to review the next
commit.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
It will add the ability to fast_forward_to on position_range
to flat_mutation_reader that does not have this ability.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This commit copies make_forwardable from streamed_mutation
to flat_mutation_reader and renames it to make_forwardable_copy.
This copy will be used as a base for make_forwardable implementation
for flat_mutation_reader.
The purpose of this commit is to make it easier to review the next
commit.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This method creates a flat_mutation_reader
instead of mutation_reader. All users will be gradually
converted to the new interface. make_reader is implemented
using make_flat_reader and will be removed once all users
are migrated.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This will be used as an intermediate state of migration
from mutation_reader to flat_mutation_reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
There will be a second implementation that will be used by
sources that are converted to flat_mutation_reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
It's not needed and causes cyclic dependency when we need
flat_mutation_reader in mutation_source.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
We don't support non-PK restrictions correctly as explained in commit
3c90607 ("tests/cql_query_test: Fix view creation in
test_duration_restrictions()") and Apache Cassandra doesn't support them
for MVs either. Change some test cases to not rely on them.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171107165138.3176-1-duarte@scylladb.com>
f44131226a introduced a regression where for some verbs we would
return partitions in their natural sort order, but since thrift
partition ranges can wrap-around, what we need to preserve is query
order.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171103201118.18175-1-duarte@scylladb.com>
Fixes#2938.
* 'tgrabiec/fix-range-tombstone-list-exception-safety-v1' of github.com:scylladb/seastar-dev:
tests: range_tombstone_list: Add test for exception safety of apply()
tests: Introduce range_tombstone_list assertions
cache: Make range tombstone merging exception-safe
range_tombstone_list: Introduce apply_monotonically()
range_tombstone_list: Make reverter::erase() exception-safe
range_tombstone_list: Fix memory leaks in case of bad_alloc
mutation_partition: Fix abort in case range tombstone copying fails
managed_bytes: Declare copy constructor as allocation point
Integrate with allocation failure injection framework
We don't support non-PK restrictions correctly as explained in commit
3c90607 ("tests/cql_query_test: Fix view creation in
test_duration_restrictions()") and Apache Cassandra doesn't support them
for MVs either. Disable the tests, but don't remove them because they
will be resurrected once CASSANDRA-13832 is fixed.
Message-Id: <1510052422-3478-1-git-send-email-penberg@scylladb.com>
range_tombstone_list::apply() has no exception safety guarantees about
the logical state. The target mutation_partition in cache should be
assumed to be left in unspecified state. In particular, some of the
preexisting overlapping tombstones may be removed and not reinserted,
so the cache would be missing some of the range tombstone information
in case the whole allocating section fails.
Use apply_monotonically() which provides the needed guarantees.
Fixes#2938.
erase_undo_op() constructor takes ownership of *it, and destroys it
when it goes out of scope. If emplace_back() fails, *it would be
destroyed before being removed from its container (_dst._tombstones).
Fix by making sure _ops.emplace_back() won't fail.
If exception is thrown from _row_tombstones.apply(), _rows will be
left uncleared. This will trigger assertion in bi::set_member_hook
destructor, which assrts that the hook is not linked.
Always clear _rows.
Because of the small size optimization, not all copies will call the
allocator, so allocation failure injection may miss this site if the
value is not large enough. Make the testing more effective by marking
this place explicitly as an allocation point.
* seastar d71922c...8040cab (4):
> util: Introduce support for allocation failure injection
> Adding dpdk-port-index as a command line option with default value of 0
> core/sharded: Introduce invoke_on_others()
> noncopyable_function: improve support for capturing mutable lambdas
"Fixes #2933
Fixes regressions introduced by config restructuring.
Allows "base" config to handle errors by warning, while
other uses can opt otherwise."
[pdziepak: resolved merge conflict]
* 'calle/cfgfix' of github.com:scylladb/seastar-dev:
config_test: Use error handler (ignore errors) + add error test
config: Resurrect command line aliases that where lost
main: Use error handler for config parse
config_file: Add optional "error_handler" to yaml parse functions
"This patch series adds support for secondary index queries using the
backing index view that's created when CREATE INDEX statement is
executed.
Example:
-- Create keyspace and table:
CREATE KEYSPACE ks WITH REPLICATION = {'class' : 'SimpleStrategy', 'replication_factor' : 1};
CREATE TABLE ks.users (
userid uuid,
name text,
email text,
country text,
PRIMARY KEY (userid)
);
-- Create secondary indexes:
CREATE INDEX ON ks.users (email);
CREATE INDEX ON ks.users (country);
-- Insert some data:
INSERT INTO ks.users (userid, name, email, country) VALUES (uuid(), 'Bondie Easseby', 'beassebyv@house.gov', 'France');
INSERT INTO ks.users (userid, name, email, country) VALUES (uuid(), 'Demetri Curror', 'dcurrorw@techcrunch.com', 'France');
INSERT INTO ks.users (userid, name, email, country) VALUES (uuid(), 'Langston Paulisch', 'lpaulischm@reverbnation.com', 'United States');
INSERT INTO ks.users (userid, name, email, country) VALUES (uuid(), 'Channa Devote', 'cdevote14@marriott.com', 'Denmark');
-- Query on the secondary-index backed non-primary keys:
SELECT * FROM ks.users WHERE email = 'beassebyv@house.gov';
userid | country | email | name
--------+---------+-------+------
022238c8-5213-44b5-959e-4e3e1b032f85 | France | beassebyv@house.gov | Bondie Easseby
(1 rows)
SELECT * FROM ks.users WHERE country = 'France';
userid | country | email | name
--------------------------------------+---------+-------------------------+----------------
2152d85a-61f6-4eab-af4d-e7e7d0872319 | France | beassebyv@house.gov | Bondie Easseby
59fddb6d-bfc9-4636-a9a0-85383fd815ee | France | dcurrorw@techcrunch.com | Demetri Curror
Known imitations:
- Only regular column indexes return results. Indexing primary key
components like clustering keys return empty result set because of
index view query partition key serialization issues that will be fixed
in subsequent patches.
- Secondary index queries are not paginated, which can cause problems
for queries that return a large number of rows.
- Multiple restrictions don't work correctly if one of them is backed by
a secondary-index.
- Only one secondary-indexed restriction per query is supported -- other
restrictions are ignored.
- Compound partition keys are not supported.
- ALLOW FILTERING on non-primary key columns does not work correctly
without secondary index (see issue #2200)."
* 'penberg/cql-2i-queries/v2' of github.com:penberg/scylla:
tests/cql_query_test: Add test case for secondary index queries
cql3: Secondary-index backed select statements
index: Fix index view schema when primary key component is indexed
tests/cql_query_test: Fix view creation in test_duration_restrictions()
cql3/restrictions: Add statement_restrictions::index_restrictions() helper
index: Implement index::supports_expression() for EQ operator
cql3: Make operator_type class non-copyable
index: Fix index::supports_expression() operator parameter type
cql3: Implement statement_restriction index validation
Before the patch we appended and queried at the front. Insert at the
front instead, so that writes and reads overlap. Stresses eviction and
population more.
Message-Id: <1506369562-14892-1-git-send-email-tgrabiec@scylladb.com>
"We currently can't insert row entries at any position_in_partition,
but only at full keys and after all keys. If a query range has bounds
such that we have to insert a dummy entry at non-representable position
then information about range continuity will not be fully populated.
In particular, single-row queries of a row which is not present in sstables
will miss when repeated again.
The series fixes the problem by marking the whole query range as continuous
by inserting dummy entries at boundaries when necessary.
Refs #2579."
* tag 'tgrabiec/cache-range-continuity-v2' of github.com:scylladb/seastar-dev:
tests: row_cache: Add test for population of single rows
tests: Add test for population of continuity
tests: mutation_reader_assertions: Introduce produces_compacted()
mutation: Introduce apply(mutation_fragment)
cache: Document invariants of cache_streamed_mutation::_lower_bound
cache_streamed_mutation: Special-case population for singular ranges
query: Introduce is_single_row()
cache_streamed_mutation: Increment mispopulation counter when can't populate due to eviction
cache_streamed_mutation: Override continuity of older versions when populating
cache_streamed_mutation: Mark whole query range as continuous
tests: cache_streamed_mutation: Allow creating expected_row at any position_in_partition
cache_streamed_mutation: Populate continuity when range adjacent to non-latest version rows
cache_streamed_mutation: Avoid lookup in maybe_add_to_cache() in more cases
row_cache: Make read_context::key() valid before reading from underlying starts
mutation_partition: Allow creating rows_entry at any clustered position_in_partition
position_in_partition: Do not use -2 and +2 weights
clustering_ranges_walker: Make contains() drop range tombstones adjacent to query range
mutation_partition: Remove delegating_compare()
mvcc: Print iterators in operator<< for partition_snapshot_row_cursor
mvcc: Introduce partition_snapshot_row_weakref
mvcc: Make the null state of partition_snapshot::change_mark explicit
mvcc: Add partition_snapshot::region() getter
mvcc: Add partition_snapshot::schema() getter
position_in_partition: Introduce before_key()
position_in_partition: Introduce min()
position_in_partition: Introduce for_static_row()
While not being a real unit tests memory_footprint can be a quite useful
tool and running it among other tests will ensure that we will notice
when it gets broken.
Message-Id: <20171102160233.6756-2-pdziepak@scylladb.com>
When created cache registers several metrics, since attempts to create
an already existing metrics result in an exception being thrown it is no
longer possible to have two cache instances at the same time. This is
exactly what happens in memory_footprint: one (useless) cache object is
created through a call to do_with_cql_env() and, then, memory_footprint
explicitly creates another one (not a useless one).
The tests itself doesn't really need a full cql environment and the only
reason it was added is so that storage_service is initialised and various
code paths can query for the available cluster features. This can be
done in a much lightweight way using storage_service_for_tests.
Fixes memory_footprint failure (until next time we decide there is
nothing wrong with globals).
Message-Id: <20171102160233.6756-1-pdziepak@scylladb.com>
$basearch isn't parsed as expected, the finaly baseurl is wrong.
We only have x86_64 arch in external 3rdparty repository, and
the conf file is only for x86_64, so it's fine to use hardcode
x86_64.
The problem was introduced by commit
b5e83ebd94 ("dist/redhat: switch 3rdparty
packages to external build service").
Fixes#2930
Signed-off-by: Amos Kong <amos@scylladb.com>
Message-Id: <708f46a7c36623e86fee278462c80db1eff3b820.1509700430.git.amos@scylladb.com>
This patch adds support for secondary-index backed select statements.
Current select_statement class is split into two separate classes:
primary_key_select_statement that retains regular query behavior and
indexed_table_select_statement that introduces the new secondary-index
backed query logic. One of the two behaviors is selected at query
preparation time to minimize overhead for non-indexed queries.
This fixes index view schema to exclude indexed column when a primary
key component like clustering key is indexed. This fixes a server crash
when CREATE INDEX statement is executed on a clustering key column.
The materialized view created in test_duration_restriction() restricts
on a non-PK column. Since Scylla's ALLOW FILTERING and secondary index
validation path is broken, once we start to do secondary index queries,
query processor thinks there's a secondary index backing that non-PK
column and fails because it's unable to find such column.
Fix up the view to only trigger the duration type validation error we're
interested in here.
Fixes the case of continuity not being populated when the row which is
the upper bound of the population range belongs to a non-latest
version. In such case we wouldn't mark the range as continuous,
because we can't modify rows of non-latest versions. To fix this,
create an empty entry in latest version which will just override the
continuity flag of the old entry.
Before this patch only ranges between returned row fragments were
marked as continuous. In the extreme case, there could be no such
fragments, in which case next read would miss as well. To avoid this,
mark whole query range as continuous by inserting dummy entries when
necessary.
Refs #2579.
Current code will not mark the range as continuous if the previous
entry does not come in the latest version. Fix that by switching
to partition_snapshot_row_pointer, which is capable of checking
in older versions as necessary.
Also, we avoid the key comparison if we know that the iterator
is still valid.
So that we can call cache_streamed_mutation::can_populate() before
we start reading from underlying. Will be needed in upcoming changes
which insert dummy entries when falling back to underlying.
::weight() is using those values for excl_end and excl_start in order
to be able to represent non-overlapping ranges. In their model the end
bound is inclusive. We don't need this, since position_range has end
bound exclusive.
This change makes that:
position_in_partition::after_key(y)
== position_in_partition::for_range_end(clutering_range::make({x}, {y})
This patch changes the way the multiget_{slice, count} verbs return
their results, by ensuring a queried key that produced no results is
still present in the returned map, associated with an empty list.
This is not required by the thrift interface, and it is a performance
step back, but matches the behavior of Apache Cassandra.
Said behavior is relied upon by projects like JanusGraph, whose
integration with Scylla motivated this patch.
Fixes#2900
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171019161104.22797-2-duarte@scylladb.com>
Most common operations, like multiget_count and multiget_slice, return
maps. So, instead of keeping a vector internally in column_visitor
that we later transform into a map, keep a map that we transform into
a vector for the uncommon operations.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20171019161104.22797-1-duarte@scylladb.com>
"This series cleans-up the query processor header and source file,
including deleting dead Java code.
There are no functional or interface changes.
I've run all unit tests and observed no failures."
* 'jhk/clean_up_qp/v2' of github.com:hakuch/scylla:
cql3/query_processor: Fix formatting
cql3/query_processor: Organize headers
cql3/query_processor.hh: Consolidate `public` and `private` sections
cql3/query_processor: Remove dead Java code
* seastar 8babd1f...d71922c (11):
> configure.py: add -Wno-sign-compare to compile Boost.Test with gcc-7
> log: Print nested exceptions
> reactor: do not account non idle activity for total idle time calculation
> execution_stage: defer execution less aggressively
> Fix -Wreturn-type warnings
> cpu scheduler: make _reciprocal_shares_times_2_32 wider to avoid overflow problems
> noncopyable_function add bool operator
> execution_stage: make make_execution_stage return a named type
> memory: support overriding the default allocator page size
> memory: fix crash during startup with large page_size
> core: io_destroy is missing when destructing reactor, which causes io_context leak
We are failing to build .deb package on pbuilder due to lack of build time
dependencies so we need add those packages on Build-Depends, also we need to
follow Debian packaging style for the package contains python scripts.
Fixes#2918
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1508457215-11552-1-git-send-email-syuu@scylladb.com>
Switch to g++-7/boost-1.63 for Ubuntu 14.04/16.04 that newly provided via
our 3rdparty PPA.
To make Scylla compilable with boost-1.63/g++-7, we need to disable following
warnings:
- misleading-indentation
- overflow
- noexcept-type
Compile error message:
https://gist.github.com/syuu1228/96acc640c56c3316df5ce6911d60beea
Seastar also has similar problem, it needs to disable 'sign-compare', detail
is in a patch for Seastar.
This update also fixes current Ubuntu 14.04/16.04 compilation error problem,
since errors were come from too old g++/boost.
Fixes#2902Fixes#2903
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1508457509-13122-1-git-send-email-syuu@scylladb.com>
"When reading a single row it is possible that the read will be satisfied
by just reading from one of the data source candidates. To exploit this
an optimization is employed which sorts data source candidates by their
timestamp and reads mutations from the most recent to the oldest. When
all needed cells are present and their earliest timestamp is still
later than the latest one of the remaining data source the read can be
terminated early.
However this optimization also has the possibility to backfire as the
data sources are read sequentially, so if all of them has to be read
eventually then we will end up worse then without it.
Thus the optimization can be disabled up-front or enabled to only run
until its efficiency degrades below a certain threshold.
Also counters are added to column-families to make it possible to
observe how well it performs.
Benchmarking
Benchmarking was done with disabled cache and at a constant op rate of
4k (1/3 of the max op rate on my box), against 3 sstables containing the
same 10000 rows.
1) Optimization turned off (all sstables read paralelly)
latency mean : 1.3 [simple:1.3]
latency median : 1.0 [simple:1.0]
latency 95th percentile : 2.4 [simple:2.4]
latency 99th percentile : 2.9 [simple:2.9]
latency 99.9th percentile : 8.0 [simple:8.0]
latency max : 13.5 [simple:13.5]
2) Optimization turned on, best case (1 of 3 sstables read)
latency mean : 0.6 [simple:0.6]
latency median : 0.6 [simple:0.6]
latency 95th percentile : 1.0 [simple:1.0]
latency 99th percentile : 1.2 [simple:1.2]
latency 99.9th percentile : 4.4 [simple:4.4]
latency max : 13.4 [simple:13.4]
3) Optimization turned on, best case, IN query (1 of 3 sstables read)
latency mean : 0.7 [simple_in:0.7]
latency median : 0.6 [simple_in:0.6]
latency 95th percentile : 1.1 [simple_in:1.1]
latency 99th percentile : 1.4 [simple_in:1.4]
latency 99.9th percentile : 5.4 [simple_in:5.4]
latency max : 16.8 [simple_in:16.8]
4) Optimization turned on, worst case (3 of 3 sstables read sequentally)
latency mean : 2.8 [simple:2.8]
latency median : 2.3 [simple:2.3]
latency 95th percentile : 5.4 [simple:5.4]
latency 99th percentile : 6.5 [simple:6.5]
latency 99.9th percentile : 13.5 [simple:13.5]
latency max : 19.2 [simple:19.2]
5) Optimization turned on, mid case (2 of 3 sstables read sequentally)
latency mean : 1.4 [simple:1.4]
latency median : 1.1 [simple:1.1]
latency 95th percentile : 2.7 [simple:2.7]
latency 99th percentile : 3.2 [simple:3.2]
latency 99.9th percentile : 7.7 [simple:7.7]
latency max : 15.1 [simple:15.1]"
Ref #324
* 'bdenes/optimize_single_row_read_v6' of github.com:denesb/scylla:
Add unit tests for single_key_sstable_reader
Add counters for the single-key reader optimization
Add single_key_parallel_scan_threshold option
single_key_sstable_reader: optimize single-row queries
single_key_sstable_reader: move reading code into it's own method
Add selects_only_full_rows() and selects_only_full_rows_with_atomic_columns()
Add two counters, one to determine how many of the reads fall into the
optimization, and a second one to determine it's effectiveness.
The first one is single_key_reader_optimization_hit_rate. It contains
the rate of reads that the optimization applies to out of all the reads
that go into the single_key_sstable_reader.
The second one, single_key_reader_optimization_extra_read_proportion is
a histogram of the efficiency of the optimization. It contains the
proportion of extra data-sources read. It's a number between 0 and 1,
where 0 is the best case (only one data-source was read) and 1 is the
worst case (all data-sources were read eventually). This is the same
number that is used for the threshold option (see previous patch).
Each of the histogram's buckets cover a chunk of 0.1 from the [0, 1]
range.
Note that single_key_parallel_scan_threshold effectively provides an
upper bound for the proportion as the optimization is turned off as soon
as it goes above that number.
The counters are disabled if single_key_parallel_scan_threshold is set
to 0 disabling the optimization entirely.
This option regulates when exactly the single-key optimization is
considered ineffective and turned off.
The threshold is the proportion of the extra data source candidates that
can be read before the optimization is considered ineffective and
disabled. The proportion is calculated as follows:
(read_data_sources - 1) / (total_data_sources - 1)
We substract 1 from the read_data_sources and total_data_sources to
effectively measure the rate of *extra* data sources we read. This
makes sure that the proportion is meaningful even if e.g. we have only
have a total of 2 data-sources and we read only 1 (best case).
Whenever this number goes above the threshold the optimization is
disabled. The threshold is number between 0 and 1, 0 forces the
optimization off and 1 forces it on. Increase the treshold to favor
throughput over latency for single-row reads, decrease the treshold to
improve latency at the expense of throughput.
If the threshold is > 0 (it's not force disabled) and the optimization
is disabled due to a read crossing the threshold, we will issue
"probing" reads (every 100th read) to determine if the optimization is
worth re-enabling. Probing reads are allowed to run through the
optimization path and if they go below the threshold the optimization is
re-enabled.
For single-row queries that only query atomic cells one can put a lower
bound on the timestamps which may affect the query results and thus rule
out entire data sources. This allows the query to read only those
sstables that actually contribute to the result.
To do this we incrementally move through the sstables overlapping with
the query range, checking after each read mutation whether we already
have a value for all required cells and whether the lower-bound of their
timestamps is higher than the upper-bound of the timestamps of all the
remaining data-sources. When this condition is met we terminate the
read.
"The main problem fixed is slow processing of application state changes.
This may lead to a bootstrapping node not having up to date view on the
ring, and serve incorrect data.
Fixes #2855."
* tag 'tgrabiec/gossip-performance-v3' of github.com:scylladb/seastar-dev:
gms/gossiper: Remove periodic replication of endpoint state map
gossiper: Check for features in the change listener
gms/gossiper: Replicate changes incrementally to other shards
gms/gossiper: Document validity of endpoint_state properties
storage_service: Update token_metadata after changing endpoint_state
gms/gossiper: Process endpoints in parallel
gms/gossiper: Serialize state changes and notifications for given node
utils/loading_shared_values: Allow Loader to return non-future result
gms/gossiper: Encapsulate lookup of endpoint_state
storage_service: Batch token metadata and endpoint state replication
utils/serialized_action: Introduce trigger_later()
gossiper: Add and improve logging
gms/gossiper: Don't fire change listeners when there is no change
gms/gossiper: Allow parallel apply_state_locally()
gms/gossiper: Avoid copies in endpoint_state::add_application_state()
gms/failure_detector: Ignore short update intervals
"Extracts the yaml/boost-po aspects of the "self-describing" db::config
into an abstract type.
db::config is then reimplemented in said type, removing some of the
slightly cumbersome entanglement with seastar opts (log).
Adds a main hook for additional configuration files (options + file)"
* 'calle/config' of github.com:scylladb/seastar-dev:
main/init: Add registerable configuration objects
db::config: Re-implement on utils/config_file.
utils::config_file: Abstract out config file to external type
storage_service depends on endpoint states to be replicated to all
shards before token metadata is replicated. Currently this is taken
care of by storage_service::replicate_to_all_cores(), invoked from
storage_service's change listener. It copies whole endpoint state map,
which is expensive in large clusters. It's more efficient to replicate
only incremental changes, and only once, rather than for each
application state.
There is a requirement that whatever is present in token_metadata,
should also be present in endpoint_state. Because of that, we should update
endpoint_state first (set_gossip_tokens).
Apache Cassandra switched to this order as well in commit
b39d984f7bd682c7638415d65dcc4ac9bcb74e5f.
Makes state application faster due to increased parallelism.
Refs #2855.
Bootrap of 11th node, ignoring apply_state_locally() which complete instantly:
Before:
DEBUG 2017-10-06 15:24:04,213 [shard 0] gossip - apply_state_locally() took 1230 ms
DEBUG 2017-10-06 15:24:04,223 [shard 0] gossip - apply_state_locally() took 1421 ms
DEBUG 2017-10-06 15:24:04,225 [shard 0] gossip - apply_state_locally() took 607 ms
DEBUG 2017-10-06 15:24:04,288 [shard 0] gossip - apply_state_locally() took 488 ms
DEBUG 2017-10-06 15:24:04,408 [shard 0] gossip - apply_state_locally() took 1425 ms
After:
DEBUG 2017-10-06 16:24:13,130 [shard 0] gossip - apply_state_locally() took 814 ms
It's possible that a change listener for a later state will run before
change listener for the previous state completes, in which case
node's state can be corruped. For example, the previous change listener
may override sysytem.peers with an old value.
This patch fixes the problem by serializing state changes and
listeners for each node.
The implementation uses loading_shared_values so that the lock remains
alive as long as there is anyone holding it. Using endpoint_state_map
for that doesn't seem appropraite, because entries can be removed from
it while listeners are still running. There is code in the gossiper
which anticipates that entry may be gone across deferring points in
some places.
apply_new_states() always fires change listeners for received values,
even if we already processed the state earlier. Some change listeners
are heavy-weight, e.g. storage_service::handle_state_normal(). We
should avoid calling them more than necessary.
Make sure that we always run the change listeners by putting them in a
defer() block. Otherwise, if exception is thrown in the middle of state
application, change listeners would not be run. Later we would not
detect the change for states which were already applied, and not run
the change listers.
Fixes#2867
It is serialized since e428d06f40. This causes regression in
performance of application state propagation due to reduced
parallelism.
Processing states for each node has high latency due to memtable
flushes triggered by update_tokens() and commitlog syncs done by
system.peers updates, if commitlog sync mode is set to "batch". We
have high internal concurrency for these, so increasing parallelism
significantly reduces time to process all states.
Fixes#2855.
Failure detector decides that a node is down if it hasn't received a change of
its heartbeat for longer than ~11 times the average of past intervals between
updates.
If there are multiple incoming ACKs containing information about the
same node, we may detect and report a change for each of them. This
will cause failure_detector to establish that the average report
period is in milliseconds. After the update storm is over, it will
claim the node failure very soon, because report period will now be a
large multiple of the average.
Fix by not counting short updates into the calculation of average
arrival time.
Fixes#2861.
Handling all the boost::commandline + YAML stuff.
This patch only provides an external version of these functions,
it does not modify the db::config object. That is for a follow-up
patch.
"Histograms are a native prometheus type, and there are many functions
available that operate on them. There is extensive documentation about
them at https://prometheus.io/docs/practices/histograms/
One example is the function histogram_quantile(), that can extract
useful quantiles from the histograms. Currently, those functions don't
work well.
The reasons are twofold:
1) We are only exporting 16 metrics, starting from 1usec. That means
that the highest latency we can differentiate is 4ms. After that,
everything falls into the same bin.
2) The format that prometheus expects is that each bin will contain
the total number of points seen *up until that bin*, while we
currently export the total number of points that falls between bins.
IOW, it is a cummulative histogram.
About point two, granted it is a bit hidden in their website, but it is
there. The following phrase about a caveat make it clear:
"Note that we divide the sum of both buckets. The reason is that the
histogram buckets are cumulative. The le="0.3" bucket is also contained
in the le="1.2" bucket; dividing it by 2 corrects for that."
It is also not needed to accumulate things that fall over the last bin:
the _count component of the histogram will already account for that."
Acked-by: Amnon Heiman <amnon@scylladb.com>
Acked-by: Gleb Natapov <gleb@scylladb.com>
* 'prometheus-histograms' of github.com:glommer/scylla:
storage_proxy: change reporting of estimated histograms
estimated_histogram: bring histogram closer to what prometheus expects.
We moved to new dependency package names like
antlr3-c++-dev to scylla-antlr35-c++-dev when we moved to ppa on Ubuntu, but
Debian still uses old dist/debian/dep packages.
So keep using old style package names.
Fixes#2831
Message-Id: <1508245175-2184-1-git-send-email-syuu@scylladb.com>
query::full_slice doesn't select any regular or static columns, which
is at odds with the expectations of its users. This patch replaces it
with the schema::full_slice() version.
Refs #2885
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1507732800-9448-2-git-send-email-duarte@scylladb.com>
This patch introduces schema::full_slice(), which returns a
partition_slice selecting the full clustering range, as well as all
static and regular columns. No options aside from the default are
set in that partition_slice.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1507732800-9448-1-git-send-email-duarte@scylladb.com>
Fixes: 2898
Typo error in gensalt(). Only returned selected hash method, not
the random salt bytes. Does not prevent the hash function from
operating, but strength is ever so reduced.
Message-Id: <20171016130505.25593-2-calle@scylladb.com>
"This changeset is the first step to flatten mutation_reader.
Then it introduces new mutation_fragment types for partition header and end of partition.
Using those a new flat_mutation_reader is defined.
Finally it introduces converters between new flat_mutation_reader and
old mutation_reader."
* 'haaawk/flattened_mutation_reader_v12' of github.com:scylladb/seastar-dev:
Add tests for flat_mutation_reader
Introduce conversion from flat_mutation_reader to mutation_reader
Introduce conversion from mutation_reader to flat_mutation_reader
Introduce flat_mutation_reader
Extract FlattenedConsumer concept using GCC6_CONCEPT
Introduce partition_end mutation_fragment
Introduce a position for end of partition
Introduce partition_start mutation_fragment
Introduce FragmentConsumer
Introduce a position for partition start
streamed_mutation: Extract concepts using GCC6_CONCEPT macro
The command scans random set of objects in a small pool (or, optionally
only objects of a certain size) for vptrs and builds a histogram, so that
most often used vptrs can be easily found. The command is useful to find
"memory leaks" caused by creating of too many tasks of a certain type
which is usually a result of unlimited parallelism somewhere.
Message-Id: <20171015081634.GB21092@scylladb.com>
Those tests run mutation source test for all sources
using conversion to and from flat_mutation_reader.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This reader operates on mutation_fragments instead of
streamed_mutations.
Each partition starts with a partition_header fragment
and ends with end_of_partition fragment.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This fixes a regression introduced in 27a3b4bca9 (master only).
partition_range_cursor assumes that as long as references are valid,
_end is valid as well. But if new entries were inserted before _end,
it may not, if the new entries fall after the query range. This may
result in reads returning partitions from outside the query range.
Message-Id: <1507815478-20269-1-git-send-email-tgrabiec@scylladb.com>
"Based on the functions get_endpoint_state_for_endpoint_ptr(),
get_application_state_ptr() and
endpoint_state::get_application_state_ptr(), this series
cleanups miscelaneous functions related to the gossiper.
It not only removes duplicated code, but also omits many copies.
All pointer usages have been audited for safety."
Acked-by: Asias He <asias@scylladb.com>
Acked-by: Tomasz Grabiec <tgrabiec@scylladb.com>
* 'gossiper-cleanup/v2' of github.com:duarten/scylla: (27 commits)
gms/endpoint_state: Remove get_application_state()
service/storage_service: Avoid copies in prepare_replacement_info()
service/storage_service: Cleanup get_application_state_value()
service/storage_service: Cleanup handle_state_removing()
service/storage_service: Cleanup get_rpc_address()
locator/reconnectable_snitch_helper: Avoid versioned_value copies
locator/production_snitch_base: Cleanup get_endpoint_info()
service/migration_manager: Avoid copies in is_ready_for_bootstrap()
service/migration_manager: Cleanup has_compatible_schema_tables_version()
service/migration_manager: Fix usages of get_application_state()
cache_hit_rate: Avoid copies in get_hit_rate()
gms/endpoint_state: Avoid copies in is_shutdown()
service/load_broadcaster: Avoid copy in on_join()
gms/gossiper: Cleanup get_supported_features()
gms/gossiper: Cleanup get_gossip_status()
gms/gossiper: Cleanup seen_any_seed()
gms/gossiper: Cleanup get_host_id()
gms/gossiper: Removed dead uses_vnodes() function
gms/gossiper: Cleanup uses_host_id()
gms/gossiper: Add get_application_state_ptr()
...
We were taking a reference to a temporary value in different places.
Fix them by using get_application_state_ptr(), which also avoids a copy.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the get_application_state_ptr() function, which
allows access to a versioned_value of a particular endpoint.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Now that we have get_endpoint_state_for_endpoint_ptr(), which does not
return a copy and allows mutating the actual state, we can use it
instead of repeating the lookup code.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Make it use get_endpoint_state_for_endpoint_ptr(), check if gossiper is
enabled, mark it as const, and have some callers use it instead of open
coding the logic.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Have convict() use get_endpoint_state_for_endpoint_ptr(), simplify
logging, and also protect expensive operations by checking the log
level.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
For every finished compaction, we were calculating shards for all
existing tables. With ignore_msb set to 0, it's probably not a big
deal, but if ignore_msb is like 12 and LCS is used (meaning thousands
of tables possibly), the operation may stall the reactor for a
considerable amount of time. That's fixed by caching shards.
Fixes#2875.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20171011053424.22308-1-raphaelsc@scylladb.com>
"gossiper::get_endpoint_state_for_endpoint() returns a copy of
endpoint_state, which we've seen can be very expensive. This
series introduces a function that returns a pointer and avoids
the copy.
Fixes#764"
* 'endpoint-state/v2' of https://github.com/duarten/scylla:
gossiper: Avoid endpoint_state copies
endpoint_state: const-qualify functions
storage_service: Remove duplicate endpoint state check
This type of mutation_fragment will be used in new mutation_reader
to signal the end of the current partition.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This position will be used for mutation fragment
that represents the end of partition.
This position sorts after all other mutation fragments.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This type of mutation_fragment will be used in new mutation_reader
to signal the beginning of the next partition.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
gossiper::get_endpoint_state_for_endpoint() returns a copy of
endpoint_state, which we've seen can be very expensive.
This patch adds a similar function which returns a pointer instead,
and changes the call sites where using the pointer-returning variant
is deemed safe (the pointer neither escapes the function, nor crosses
any defer point).
Fixes#764
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Seastar small pools can now fall back to smaller spans. Adjust
the 'scylla memory' command accordingly.
Message-Id: <20171005123935.13503-1-avi@scylladb.com>
* seastar c62bbf9...8babd1f (9):
> Enhanced support for Travis CI: build with and without DPDK support, use varioius compilers (GCC 5/6/7)
> backtrace: Allow whitespace after the backtrace addresses
> test.py: fix typo in noncopyable_function_test
> utils: introduce noncopyable_function
> Revert "utils: introduce noncopyable_function"
> utils: introduce noncopyable_function
> Add seastar-addr2line helper script to decode backtraces
> execution_stage: pass scheduling_group to constructor
> reactor: preempt tasks when a signal is received
"This series implements CAST AS functions in scylla.
It allows to use expressions of the form CAST(x AS type) in select statements.
Primary motivation for this functions came from aggregate functions, because
function avg(.) gives rounded results for interger columns. Now it is possible
to convert such column to float/double and obtain floating point results:
SELECT ... avg(cast(x as double)), ...
Fixes #2280."
* 'danfiala/2280-patch-series-v2' of https://github.com/hagrid-the-developer/scylla:
tests: Add test for CAST AS functions.
cql3: Add support for CAST AS functions to ANTLR grammar.
cql3/selectable: Add selectable::with_cast for CAST AS functions.
cql3/functions: Add support for CAST AS functions.
types:: Add support for CAST AS functions.
types: Moved code that implements conversion of types' values to string.
"This patch series adds backing materialized view for secondary indices.
When a new index is created with the 'CREATE INDEX' statement, a backing
materialized view is created automatically.
For example, assuming the following table:
CREATE TABLE ks1.users (
userid uuid,
email text,
PRIMARY KEY (userid)
);
When the following index is created:
CREATE INDEX user_email ON ks1.users (email);
The following materialized view is also created:
cqlsh> DESCRIBE ks1.users;
<snip>
CREATE MATERIALIZED VIEW ks1.user_email_index AS
SELECT email, userid
FROM ks1.users
WHERE email IS NOT NULL
PRIMARY KEY (email, userid)
WITH CLUSTERING ORDER BY (userid ASC)
AND bloom_filter_fp_chance = 0.01
AND caching = {'keys': 'ALL', 'rows_per_partition': 'ALL'}
AND comment = ''
AND compaction = {'class': 'SizeTieredCompactionStrategy'}
AND compression = {'sstable_compression': 'org.apache.cassandra.io.compress.LZ4Compressor'}
AND crc_check_chance = 1.0
AND dclocal_read_repair_chance = 0.1
AND default_time_to_live = 0
AND gc_grace_seconds = 864000
AND max_index_interval = 2048
AND memtable_flush_period_in_ms = 0
AND min_index_interval = 128
AND read_repair_chance = 0.0
AND speculative_retry = '99.0PERCENTILE';
CQL queries will use the backing materialized view as part of queries on
indexed columns to fetch the primary keys."
* 'penberg/cql-2i-backing-view/v3' of github.com:scylladb/seastar-dev:
schema_tables: Create backing view for indices
database: Kill obsolete secondary index manager stub
cql3: Wire up secondary index manager
cql3/restrictions: Add term_slice::is_supported_by() function
index: Add secondary_index_manager::create_view_for_index()
index: Add target_parser::parse() helper
cql3/statements: Add index_target::from_sstring() helper
index: Add secondary_index_manager::get_dependent_indices()
index: Add secondary_index_manager::reload()
index: Add secondary_index_manager::list_indexes()
index: Add index class
index: Pass column_family to secondary_index_manager constructor
database: Make secondary index manager per-column family
This patch wires calls to secondary index manager reload() in
merge_tables_and_views() and changes make_update_indices_mutations() to
also create mutations for the backing materialized view. After this
patch, "CREATE INDEX" CQL statement also creates a materialized view.
We are currently collapsing the histograms in 16 points, exponentially
increasing in value, starting from 1.
While reducing the number of points is a worthy goal, the current
configuration caps us at 4ms. Our latencies tend to be higher than this.
Starting from 1 is also a bit of an exhaggeration: rarely are our
latencies in that range. This patch changes reporting so that we
report 20 points starting from 32.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Histograms are a native prometheus type, and there are many functions
available that operate on them. There is extensive documentation about
them at https://prometheus.io/docs/practices/histograms/
One example is the function histogram_quantile(), that can extract
useful quantiles from the histograms. Currently, those functions don't
work well.
The reasons are twofold:
1) We are only exporting 16 metrics, starting from 1usec. That means
that the highest latency we can differentiate is 4ms. After that,
everything falls into the same bin.
2) The format that prometheus expects is that each bin will contain
the total number of points seen *up until that bin*, while we
currently export the total number of points that falls between bins.
IOW, it is a cummulative histogram.
About point two, granted it is a bit hidden in their website, but it is
there. The following phrase about a caveat make it clear:
"Note that we divide the sum of both buckets. The reason is that the
histogram buckets are cumulative. The le="0.3" bucket is also contained
in the le="1.2" bucket; dividing it by 2 corrects for that."
It is also not needed to accumulate things that fall over the last bin:
the _count component of the histogram will already account for that.
This patch changes the histogram format to be more in line with what
prometheus expect.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
The call to std::ref() is not namespace-qualified, and so can conflict
with seastar::ref().
Fix by naming std::ref() explicitly.
Message-Id: <20171004155250.4960-1-avi@scylladb.com>
"Makes authorizer/authenticator actually pluggable (by class name)
and adds a "Transitional" type for both, conforming to the DSE
definition of the types.
The idea is to allow a rolling upgrade of a cluster to
authentication op by first making all clients provide credentials
(ignored by non-auth), then node by node enable auth with
transitional handlers, then ensure user DB is populated and
distributed, and finally rollingly enable strict auth for
each node. Pfew."
Fixes#2836
* auth: Transitional auth wrappers
auth: Make authenticator/authorizer use actual name based lookup
Similar to DSE objects with similar name. Basically ignores
all authentication/authorization except "superuser" login. All others
sessions are treated as anonymous.
Note: like DSE counterparts, a client session must still _use_
authentication to be able to connect, even though the actual content of
the auth is mostly ignored.
Allows registry to give back, for example, shared_ptr etc instead of
solely unique_ptr. If a registry is defined with seastar/std
shared/lw_shared/unique_ptr as "BaseType", the type will assume
this is the intended result type.
"Currently restricting_mutation_reader restricts mutation_readears on a
count basis. This is inaccurate on multiple levels. The reader might be
a combined_mutation_reader, which might be composed of multiple
individual readers, whose number might change during the lifetime of the
reader. The memory consumption of the readers can vary and may change
during the lifetime of the reader as well.
To remedy this, make the restriction memory-consumption based. The
restricting semaphore is now configured with the amound of memory
(bytes) that its readers are allowed to consume in total. New readers
consume 128k units up-front to account for read-ahead buffers, and then
consume additional units for any buffer (returned
from input_stream<>::read()) they keep around.
Like before, readers already allowed to read will not be blocked,
instead new readers will be blocked on their first read if all the units
all consumed.
Fixes #2692."
* 'bdenes/restricting_mutation_reader-v5' of https://github.com/denesb/scylla:
Update reader restriction related metrics
Add restricted_reader_test unit test
restricted_mutation_reader: restrict based-on memory consumption
mutation_reader.hh: Move restricted_reader related code
"
The original motivation for the "utils: introduce a loading_shared_values" series was a hinted handoff work where
I needed an on-demand asynchronously loading key-value container (a replica address to a commitlog instance map).
It turned out that we already have the classes that do almost what I needed:
- utils::loading_cache
- sstables::shared_index_lists
Therefore it made sense to find a common ground, unify this functionality and reuse the code both in the classes above and in the
new hinted handoff code.
This series introduces the utils::loading_shared_values that generalizes the sstables::shared_index_lists
API on top of bi::unordered_set with the rehashing logic from the utils::loading_cache triggered by an addition
of an entry to the set (PATCH1).
Then it reworks the sstables::shared_index_lists and utils::loading_cache on top of the new class (PATCH2 and PATCH3).
PATCH4 optimizes the loading_cache for the long timer period use case.
But then we have discovered that we have another "customer" for the loading_cache. Apparently our prepared statements cache
had a birth flaw - it was unlimited in size - unless the corresponding keyspace and/or table are modified/dropped the entries
are never evicted. We clearly need to limit its size and it would also make sense to evict the cache entries that haven't been
used long enough.
This seems like a perfect match for a utils::loading_cache except for prepared statements don't need to be reloaded after
they are created.
Patches starting from PATCH5 are dealing with adding the utils::loading_cache the missing functionality (like making the "reloading"
conditional and adding the synchronous methods like find(key)) and then transitioning the CQL and Thrift prepared statements
caches to utils::loading_cache.
This also fixes #2474."
* 'evict_unused_prepared-v5' of https://github.com/vladzcloudius/scylla:
tests: loading_cache_test: initial commit
cql3::query_processor: implement CQL and Thrift prepared statements caches using cql3::prepared_statements_cache
cql3: prepared statements cache on top of loading_cache
utils::loading_cache: make the size limitation more strict
utils::loading_cache: added static_asserts for checking the callbacks signatures
utils::loading_cache: add a bunch of standard synchronous methods
utils::loading_cache: add the ability to create a cache that would not reload the values
utils::loading_cache: add the ability to work with not-copy-constructable values
utils::loading_cache: add EntrySize template parameter
utils::loading_cache: rework on top of utils::loading_shared_values
sstables::shared_index_list: use utils::loading_shared_values
utils: introduce loading_shared_values
Update description of existing reader count metrics, add memory
consumption metrics. Use labels to distinguish between system, user and
streaming reads related metrics.
Restrict readers based on their memory consumption, instead of the count
of the top-level readers. To do this an interposer is installed at the
input_stream level which tracks buffers emmited by the stream. This way
we can have an accurate picture of the readers' actual memory
consumption.
New readers will consume 16k units from the semaphore up-front. This is
to account their own memory-consumption, apart from the buffers they
will allocate. Creating the reader will be deferred to when there are
enough resources to create it. As before only new readers will be
blocked on an exhausted semaphore, existing readers can continue to
work.
"Currently restricting_mutation_reader restricts mutation_readears on a
count basis. This is inaccurate on multiple levels. The reader might be
a combined_mutation_reader, which might be composed of multiple
individual readers, whose number might change during the lifetime of the
reader. The memory consumption of the readers can vary and may change
during the lifetime of the reader as well.
To remedy this, make the restriction memory-consumption based. The
restricting semaphore is now configured with the amound of memory
(bytes) that its readers are allowed to consume in total. New readers
consume 128k units up-front to account for read-ahead buffers, and then
consume additional units for any buffer (returned
from input_stream<>::read()) they keep around.
Like before, readers already allowed to read will not be blocked,
instead new readers will be blocked on their first read if all the units
all consumed."
Fixes#2692.
* 'bdenes/restricting_mutation_reader-v4' of https://github.com/denesb/scylla:
Update reader restriction related metrics
Add restricted_reader_test unit test
restricted_mutation_reader: restrict based-on memory consumption
mutation_reader.hh: Move restricted_reader related code
* seastar 899fc4e...c62bbf9 (6):
> Merge "CPU Scheduler for seastar" from Avi
> reactor: set SCHED_FIFO policy for timer thread
> future: mark future::wait() as noexcept
> shared_promise: Make get_shared_future() const-qualified
> Remove pessimizing and redundant std::move()-s reported by Clang-tidy utility
> Work around GCC 5 bug: scylladb/seastar#338, scylladb/seastar#339
Fixes#2770.
Fixes#2819.
* seastar 92fdce2...899fc4e (14):
> scollectd: increment the metadata iterator with the values
> Enable Travis CI builds for Seastar.
> tests: Fix httpd test compilation error caused by unconditionally explicit tuple constructor in GCC5: scylladb/seastar#326
> core::shared_future: add available() and failed() methods
> rpc: make sure that _write_buf stream is always properly closed
> log: Fail on attempt to register logger with the same name twice
> Merge "Make backtraces useful on ASLR-enabled machines as well" from Botond
> reactor: add option to bypass fsync
> future-util: modernize do_until() implementation
> future-util: fix do_until() API to not have forwarding references
> input_stream: add rvalue variant of input_stream::consume()
> logger: remove extra spaces after timestamp
> tutorial: lifetime management
> Fix broken link for fsqual failure message
We don't pull schema during rolling upgrade, that is until
schema_tables_v3 feature is enabled on all nodes.
Because features are enabled from gossiper timer, there is a race
between feature enablement and processing of endpoint states which may
trigger schema pull. It can happen that we first try to pull, but
only later enable the feature. In that case the schema pull will not
happen until the next schema change.
The fix is to ensure that pulls abandoned due to feature not being enabled
will be retried when it is enabled.
Fixes sporadic failure in dtest:
repair_additional_test.py:RepairAdditionalTest.repair_schema_test
Message-Id: <1506428715-8182-2-git-send-email-tgrabiec@scylladb.com>
Dirty memory manager for non-system column families was being used
when applying mutations to system cfs.
That previously lead to deadlock when updating history. Basically,
write disable waits on compaction, and compaction waits on a write
that would release dirty memory for updating compaction history.
Only using the correct dirty manager wouldn't solve this problem
if write is disabled for system cf, but the problem is completely
solved in addition to previous change which updates history
outside the sstable lock.
Refs #2769.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170918215238.9810-3-raphaelsc@scylladb.com>
The reason to do that is because compaction can deadlock if refresh
disables write which waits for compaction, and compaction in turn
waits for dirty memory[1] that would be released by memtable write.
Dirty memory manager for non-system cfs was being used for system cfs,
which was useful for exposing this problem.
[1]: when updating compaction history.
Fixes#2769.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170918215238.9810-2-raphaelsc@scylladb.com>
"Fixes the problem of concurrent populations of clustering row ranges
leading to some readers skipping over some of the rows.
Spotted during code review.
Fixes #2834."
* tag 'tgrabiec/fix-cache-reader-skipping-rows-v2' of github.com:scylladb/seastar-dev:
tests: mvcc: Add test for partition_snapshot_row_cursor
tests: row_cache: Add test for concurrent population
tests: row_cache: Make populate_range() accept partition_range
tests: Add simple_schema::make_ckey_range()
cache_streamed_mutation: Add missing _next_row.maybe_refresh() call
mvcc: partition_snapshot_row_cursor: Fix cursor skipping over rows added after its position
mvcc: partition_snapshot_row_cursor: Rename up_to_date() to iterators_valid()
mvcc: Keep track of all iterators in partition_snapshot_row_cursor
mvcc: Make partition_snapshot_row_cursor printable
Make get_natural_endpoints return local address iff token metadata
is not yet setup (since that is the one address we already know of).
If a request has a consistency level requiring more endpoints, it
will still fail, but for calls with, for example, CL=ONE, at startup
we will succeed, and more or less act like local strategy. Yet,
further down the line, have data distributed as desired.
Acked-by: Gleb Natapov <gleb@scylladb.com>
Message-Id: <20170926113512.15707-1-calle@scylladb.com>
We were checking if the cursor is up_to_date(), but this is not enough
to guarantee that the cursor is valid, merely that its iterators are
valid. The cursor may be invalidated even if its iterators are valid
if there was an insertion after cursor's position.
Fixes#2834.
The cursor maintains a heap of iterators in all versions. If rows were
inserted before the latest version's iterator, cursor would not see
them. Fix by redoing the lookup for iterators not in the current row
in maybe_refresh().
Refs #2834.
Will be needed when updating the iterator for latest version. Before
this change, such iterator could be neither in _current_row nor in
_heap.
Besides that, this will allow user to always access the iterator of
latest version, which enables some optimizations in the future of
avoiding unnecessary lookups. get_iterator_in_latest_version() is now
always valid.
Since commit 8378fe190, we disable schema sync in a mixed cluster.
The detection is done using gossiper features. We need to make sure
the features are registerred, and thus can be enabled, before the
bootstrapping of a non-seed node happens. Otherwise the bootstrap will
hang waiting on schema sync which will not happen.
Message-Id: <1505893837-27876-2-git-send-email-tgrabiec@scylladb.com>
This position will be used for mutation fragment
that represents the start of a partition.
This position sorts before static row.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
It makes it easier to actually use those concepts.
Lambdas passed to mutation_fragment::visit have to declare
return type otherwise compiler fails with:
internal compiler error: Segmentation fault
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
The gossiper checks if features should be enabled from its timer
callback when it detects that endpoint_state_map changed, that is
different than shadow_endpoint_state_map.
shadow_endpoint_state_map is also assigned from endpoint_state_map in
storage_service::replicate_tm_and_ep_map(), called from
storage_service::on_change()
Call gossiper:maybe_enable_features() in replicate_tm_and_ep_map so
that we won't miss gossip feature update.
Fixes#2824
* git@github.com:scylladb/seastar-dev asias/gossip_miss_feature_update_v1:
gossip: Move the _features_condvar signal code to
maybe_enable_features
gossip: Make maybe_enable_features public
storage_service: Check gossip feature update in
replicate_tm_and_ep_map
This is another place we can update endpoint_state_map in addition to
gossiper::run().
Call the gossiper:maybe_enable_features() so that we won't miss gossip
feature update.
Restrict readers based on their memory consumption, instead of the count
of the top-level readers. To do this an interposer is installed at the
input_stream level which tracks buffers emmited by the stream. This way
we can have an accurate picture of the readers' actual memory
consumption.
New readers will consume 16k units from the semaphore up-front. This is
to account their own memory-consumption, apart from the buffers they
will allocate. Creating the reader will be deferred to when there are
enough resources to create it. As before only new readers will be
blocked on an exhausted semaphore, existing readers can continue to
work.
It skipped one sub-range in each of the 10 range batch, and
tried to access the range vector using end() iterator.
Fixes sporadic failures of
update_cluster_layout_tests.py:TestUpdateClusterLayout.simple_add_node_1_test.
Message-Id: <1505848902-16734-1-git-send-email-tgrabiec@scylladb.com>
"On bad_alloc the section is retried. If the exception happened inside
fast_forward_to() on the underlying reader, that call will be
retried. However, the reader should not be used after exception is
thrown, since it is in unspecified state. Also, calling
fast_forward_to() with cache region locked increases the chances of it
failing to allocate.
We shouldn't call fast_forward_to() with the cache region locked.
Fixes #2791."
* 'tgrabiec/dont-ffwd-in-alloc-section' of github.com:scylladb/seastar-dev:
cache_streamed_mutation: De-futurize cursor movement
cache_streamed_mutation: Call fast_forward_to() outside allocating section
cache_streamed_mutation: Switch from flags to explicit state machine
If we read a partition from a single sstable (a fairly common case),
we can bypass mutation_merger and just return the input.
Message-Id: <20170918181418.14021-1-avi@scylladb.com>
gossiper::apply_state_locally() calls handle_major_state_change() for
each endpoint, in a seastar thread, which calls mark_alive() for new
nodes, which calls ms().send_gossip_echo(id).get(). So it synchronously
waits for each node to respond before it moves on to the next entry. As
a result it may take a while before whole state is processed.
Apache (tm) Cassandra (tm) sends echos in the background.
In a large cluster, we see at the time the joining node starts
streaming, it hasn't managed to apply all the endpoint_state for peer
nodes, so the joining node does not know some of the nodes yet, which
results in the joining node ingores to stream from some of the existing
nodes.
Fixes#2787Fixes#2797
Message-Id: <3760da2bef1a83f1b6a27702a67ca4170e74b92c.1505719669.git.asias@scylladb.com>
When incremental_reader_selector is used for compaction, it will
first call incremental selector of partitioned sstable set with
minimum token that will result in first interval being skipped,
which means not everything being compacted. The interval is
skipped because iterator is incorrectly advanced when token
lies before it.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170918021446.15920-1-raphaelsc@scylladb.com>
The current sequential scan can take a long time on a small or empty table
with a large (nr_nodes * nr_vnodes) count, and can time out. Switching to
exponential scan reduces the time.
Fixes#1230.
Message-Id: <20170912173803.8277-1-avi@scylladb.com>
Forward-declare untyped_result_set and untyped_result_set_row, and remove
the include from query_processor.hh.
Message-Id: <20170916170859.27612-3-avi@scylladb.com>
"sstables.hh is already too big, and it is soon to become bigger with the
inclusion of the read_monitor, to pair it with the write_monitor.
It's a good opportunity for us to reduce sstables.hh dependencies by
moving the write monitor to its own reader. One obvious caller is
already changed so we don't need to include sstables.hh anymore."
* 'progress-monitor' of https://github.com/glommer/scylla:
sstables: do not include sstables.hh from memtable glue
sstables: move write_monitor to its own header
log_histogram is not really a histogram, it is a heap-like container.
Rename to log_heap in case we do want a log_histogram one day.
Message-Id: <20170916172137.30941-1-avi@scylladb.com>
This patch change the implementation of storage_service
repair_async_status to throw an exception, this way a 400 return code
will be returned.
Fixes#2794
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170917080533.6612-1-amnon@scylladb.com>
- Transition the prepared statements caches for both CQL and Trhift to the cql3::prepared_statements_cache class.
- Add the corresponding metrics to the query_processor:
- Evictions count.
- Current entries count.
- Current memory footprint.
Fixes#2474
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This is a template class that implements caching of prepared statements for a given ID type:
- Each cache instance is given 1/256 of the total shard memory. If the new entry is going to overflow
this memory limit - the less recently used entries are going to be evicted so that the new entry could
be added.
- The memory consumption of a single prepared statement is defined by a cql3::prepared_cache_entry_size
functor class that returns a number of bytes for a given prepared statement (currently returns 10000
bytes for any statement).
- The cache entry is going to be evicted if not used for 60 minutes or more.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Ensure that the size of the cache is never bigger than the "max_size".
Before this patch the size of the cache could have been indefinitely bigger than
the requested value during the refresh time period which is clearly an undesirable
behaviour.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Sometimes we don't want the cached values to be periodically reloaded.
This patch adds the ability to control this using a ReloadEnabled template parameter.
In case the reloading is not needed the "loading" function is not given to the constructor
but rather to the get_ptr(key, loader) method (currently it's the only method that is used, we may add
the corresponding get(key, loader) method in the future when needed).
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Current get(...) interface restricts the cache to work only with copy-constructable
values (it returns future<Tp>).
To make it able to work with non-copyable value we need to introduce an interface that would
return something like a reference to the cached value (like regular containers do).
We can't return future<Tp&> since the caller would have to ensure somehow that the underlying
value is still alive. The much more safe and easy-to-use way would be to return a shared_ptr-like
pointer to that value.
"Luckily" to us we value we actually store in a cache is already wrapped into the lw_shared_ptr
and we may simply return an object that impersonates itself as a smart_pointer<Tp> value while
it keeps a "reference" to an object stored in the cache.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Allow a variable entry size parameter.
Provide an EntrySize functor that would return a size for a
specific entry.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Get rid of the "proprietary" solution for asynchronous values on-demand loading.
Use utils::loading_shared_values instead.
We would still need to maintain intrusive set and list for efficient shrink and invalidate
operations but their entry is not going to contain the actual key and value anymore
but rather a loading_shared_values::entry_ptr which is essentially a shared pointer to a key-value
pair value.
In general, we added another level of dereferencing in order to get the key value but since
we use the bi::store_hash<true> in the hook and the bi::compare_hash<true> in the bi::unordered_set
this should not translate into an additional set lookup latency.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Since utils::loading_shared_values API is based on the original shared_index_list
this change is mostly a drop-in replacement of the corresponding parts.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This class implements an key-value container that is populated
using the provided asynchronous callback.
The value is loaded when there are active references to the value for the given key.
Container ensures that only one entry is loaded per key at any given time.
The returned value is a lw_shared_ptr to the actual value.
The value for a specific key is immediately evicted when there are no
more references to it.
The container is based on the boost::intrusive::unordered_set and is rehashed (grown) if needed
every time a new value is added (asynchronously loaded).
The container has a rehash() method that would grow or shrink the container as needed
in order to get the load factor into the [0.25, 0.75] range.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
There is no need to include the whole sstables.hh file in
memtable-sstable.hh anymore. All we need is the shared_sstable
definition and the progress monitor.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Soon I am about to introduce a read monitor, and pairing infrastructure
to manage it. Having it all living in sstables.hh force to include it
everytime, even in places that don't really need it.
Move to its own header.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
On bad_alloc the section is retried. If the exception happened inside
fast_forward_to() on the underlying reader, that call will be
retried. However, the reader should not be used after exception is
thrown, since it is in unspecified state. Also, calling
fast_forward_to() with cache region locked increases the chances of it
failing to allocate.
We shouldn't call fast_forward_to() with the cache region locked.
Fixes#2791.
We're in one state at a time, so it's better to express it as a single
variable rather than N independent flags.
In preparation before adding more states.
Right now we pass a permit to the memtable writer and that permit is
used insite write_memtable_to_sstable to compose a write_monitor.
We would like to extend the write_monitor to include other things, that
right now are not available as parameters to write_memtable_to_sstable -
and which are possibly too specialized to be.
The solution for that is to pass the write_monitor instead of the permit
to the writer. Conceptually, that also makes sense because the
write_monitor is something the sstable writer is aware of. Permits, on
the other hand, are a database concept that is alien to the sstable
writer.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170915032836.21154-1-glauber@scylladb.com>
"When there are at least 2 nodes upgraded to 2.0, and the two exchanged schema
for some reason, reads or writes which involve both 1.7 and 2.0 nodes may
start to fail with the following error logged:
storage_proxy - Exception when communicating with 127.0.0.3: Failed to load schema version 58fc9b89-74ab-37ca-8640-8b38a1204f8d
The situation should heal after whole cluster is upgraded.
Table schema versions are calculated by 2.0 nodes differently than 1.7 nodes
due to change in the schema tables format. Mismatch is meant to be avoided by
having 2.0 nodes calculate the old digest on schema migration during upgrade,
and use that version until next time the table is altered. It is thus not
allowed to alter tables during the rolling upgrade.
Two 2.0 nodes may exchange schema, if they detect through gossip that their
schema versions don't match. They may not match temporarily during boot, until
the upgraded node completes the bootstrap and propagates its new schema
through gossip. One source of such temporary mismatch is construction of new
tracing tables, which didn't exist on 1.7. Such schema pull will result in a
schema merge, which cause all tables to be altered and their schema version to
be recalculated. The new schema will not match the one used by 1.7 nodes,
causing reads and writes to fail, because schema requesting won't work during
rolling upgrade from 1.7 to 2.0.
The main fix employed here is to hold schema pulls, even among 2.0 nodes,
until rolling upgrade is complete."
* 'tgrabiec/fix-schema-mismatch' of github.com:scylladb/seastar-dev:
tests: schema_change_test: Add test_merging_does_not_alter_tables_which_didnt_change test case
tests: cql_test_env: Enable all features in tests
schema_tables: Make make_scylla_tables_mutation() visible
migration_manager: Disable pulls during rolling upgrade from 1.7
storage_service: Introduce SCHEMA_TABLES_V3 feature
schema_tables: Don't alter tables which differ only in version
schema_mutations: Use mutation_opt instead of stdx::optional<mutation>
If there is a schema pull during rolling upgrade among a two 2.0 nodes,
then schema merge will delete the persisted schema version. When the node
loads that table again, e.g. on restart, it will generate a version
which is different than the one which 1.7 nodes use. This will
cause reads and writes to fail.
To avoid this, disable pulls until all nodes are upgraded.
Fixes#2802.
We apply deletion of scylla_tables.version to the incoming schema
mutations so that table schema version is recalculated after merge.
The mutations which we read from local schema tables may not have it
deleted in which case all tables would be considered as differing on
the presence of the version field. Avoid this by deleting the field
from old mutations as well.
After the service started, a state of the service may become
"failed", "active" or "activating".
But our script does not accept scylla-ami-setup.service become "failed" state,
in result the script shows up wrong message.
So we handle these three types of state correctly.
Fixes#2759
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1504589079-1986-1-git-send-email-syuu@scylladb.com>
evict() doesn't guarantee that the whole partition is discontinuous.
In particular, partition tombstone cannot be marked as discontinuous.
The parts which are still continuous must be updated.
Broken after c78047fa5b.
Message-Id: <1505375684-28574-1-git-send-email-tgrabiec@scylladb.com>
Collect coordinator side read statistic per CF and use them in percentile
speculative read executor. Getting percentile from estimated_histogram
object is rather expensive, so cache it and recalculate only once per
second (or if requested percentile changes).
Fixes#2757
Message-Id: <20170911131752.27369-3-gleb@scylladb.com>
Currently overflow values are stored in incorrect bucket (last one
instead of special "overflow" one) and percentile() function throws
if there is overflow value. The patch fixes the code to store overflow
value in corespondent bucket and makes percentile() to take it into
account instead of throwing.
Message-Id: <20170911131752.27369-2-gleb@scylladb.com>
"This series fixes the problem of active reads causing OOM due to the fact that
partition snapshots they hold are not evictable. In particular, a single scan
of a partition larger than memory will bad_alloc due to itself.
After this, when partition entry is evicted from cache, data in all the snapshots
is also evicted. We still don't have row-level eviction, but this series lays some
grounds for it by making cache readers prepared for the possibility of rows
being evicted.
Fixes#2775.
Fixes #2730."
* tag 'tgrabiec/snapshot-evicition-in-cache-v1' of github.com:scylladb/seastar-dev:
tests: Add test for partition_entry::evict()
mutation_partition: Introduce range continuity checking methods
mutation_partition: Enable rows_entry::compare() on position_in_partition_views
tests: Extract mvcc tests to separate file
tests: row_cache: Add evicition tests
tests: simple_schema: Add new_tombstone() helper
tests: streamed_mutation_assertions: Introduce produces(mutation&)
streamed_mutation: Allow setting buffer capacity
row_cache: Evict partition snapshots
mvcc: Introduce partition_entry::evict()
row_cache: Handle eviction in partition reader
tests: row_cache_test: Don't assume mvcc snapshots are not evictable
row_cache: Reuse allocation_strategy::invalidate_references()
row_cache: Don't invalidate references on insertion
lsa: Move reclaim counter concept to allocation_strategy level
mvcc: Ensure partition_snapshot always destroys versions using proper allocator
mvcc: Encapsulate reference stability check in partition_snapshot
mvcc: Store LSA region reference in partition_snapshot
If snapshots are not evicted, they may pin unbouned amount of memory
for a long time in cache, which may lead to OOM. Evict snapshots
together with the entry.
Fixes#2775.
Fixes#2730.
The test was not updating the underlying mutation source but still
expecting to get the right data after calling invalidate(). If
snapshots are evictable, that's not guaranteed. Apply to underlying as
well, so data is read from underlying if necessary.
modification_count is currently only used to detect invalidation of
references, intended to be incremented on erasure.
Insertion into intrusive set doesn't invalidate references, so no
need to increment the counter.
partition_snapshot is managed by lw_shared_ptr. Currently it is
assumed that before it dies, maybe_merge_versions() is called on it,
which destroyes it in the right allocator context. It's not very
safe. This patch improves safety by using the right allocator in
snapshot's destructor.
This fixes a failure in view_schema_test, which starts many instances
of single_node_cql_env. cancel_atomic_deletions() causes later
deletions to fail, which causes some of the test cases to fail.
Message-Id: <1505311250-3118-2-git-send-email-tgrabiec@scylladb.com>
Fixes a hang on shutdown with --smp 2 in perf_fast_forward. The hang
is in sstables::await_background_jobs_on_all_shards(), which is
waiting on sstable deletions. Not all shards agree to delete certain
sstables, because e.g. not all shards decide to compact them
yet. Cancel those deletes after database is stopped on all shards,
like we do in main.cc
Fixes#2796.
Message-Id: <1505292239-26032-1-git-send-email-tgrabiec@scylladb.com>
Finds live objects on seastar heap of current shard which contain
given value. Prints results in 'scylla ptr' format.
Example:
(gdb) scylla find 0x600005321900
thread 1, small (size <= 512), live (0x6000000f3800 +48)
thread 1, small (size <= 56), live (0x6000008a1230 +32)
Message-Id: <1505284614-19577-1-git-send-email-tgrabiec@scylladb.com>
compaction_strategy.hh throws an exception, but it doesn't add the
exception header. It is working in-tree because of inclusion order,
but it broke one of my yet-out-of-tree changes.
In any case, it is best to add the headers we will need to the files,
and that is what this patch does.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170912233326.26114-1-glauber@scylladb.com>
filter.cc has just two smallish functions, which are part of the sstable
class. Move them to sstables.cc where the rest of the class members are defined.
Message-Id: <20170912080541.7836-1-avi@scylladb.com>
The test used apply() variant which assumed that it was invoked in a
seastar thread, which is no longer the case after commit d22fdf4. Fix
by copying outisde cache update, and use non-deferring apply() variant
for cache update.
Message-Id: <1505200142-3650-1-git-send-email-tgrabiec@scylladb.com>
This patchset reduces includes of sstables.hh, reducing compile time
by both reducing the amount of code compiled, and the amount of
needless recompiles caused by false dependencies. It does so by
replacing lw_shared_ptr<sstable>, which requires a complete class,
with a new custom type shared_sstable, which allows an incomplete
sstable class definition.
* https://github.com/avikivity/scylla deps2/v2.1
database: change truncate() to flush while compaction is disabled
database: make run_with_compaction_disabled() a non-template
database: add indirection to compaction_manager instance
database: remove dependency on compaction.hh and compaction_manager.hh
size_estimates_virtual_reader.hh: add missing include
system_keyspace: add missing include
main: add missing include
storage_service: add missing include
repair: add missing include
compaction.hh: add missig include and forward declaration
compaction_manager: add missing include
shared_index_lists.hh: add missing include
perf_fast_forward: add missing include
sstable_mutation_test: add missing include
sstables: extract version and format enum into a separate header file
database.hh: add missing forward declaration for
foreign_sstable_open_info
cql_test_env: add forward declaration
database: make column_family::disable_sstable_write() out-of-line
sstables: introduce make_sstable() as a shortcut for
make_lw_shared<sstable>
treewide: use shared_sstable, make_sstable in place of
lw_shared_ptr<sstable>
sstables: use support for lw_shared_ptr with incomplete type for
shared_sstable
sstables: reduce dependencies
streaming: remove unneeded includes
When table is created, it doesn't contain any data, so we can mark the whole
data range as continuous in cache. This way reads will immediately hit, and
flushes will populate. If sstables are later attached, the attaching process
is supposed to invalidate affected ranges (and it does).
Fixes#2536.
Message-Id: <1505200269-4031-1-git-send-email-tgrabiec@scylladb.com>
Use the lw_shared_ptr deleter support to define shared_sstable without
pulling the definition of class sstable, reducing compile time and
dependencies if only shared_sstable is needed.
The parser generator somehow confuses the use-after-scope sanitizer, causing it
to use large amounts of stack space. Disable that sanitizer on that file.
Message-Id: <20170905110628.18047-1-avi@scylladb.com>
In preparation to make run_with_compaction_disabled() a non-template,
we want to remove any non-copyable captures (so the function can be
an std::function, which requires copyability). Move the flush within
the compaction disabled region. This changes the behavior, but it shouldn't
matter.
"These patches make Scylla refuse to load counter sstables that may
contain unsupported counter shards. They are recognised by the lack of
the Scylla component.
Fixes #2766."
* tag 'reject-non-scylla-counter-sstables/v1' of https://github.com/pdziepak/scylla:
db: reject non-Scylla counter sstables in flush_upload_dir
db: disallow loading non-Scylla counter sstables
sstable: add has_scylla_component()
A keyspace can be deleted while write is ongoing, so the object cannot
be used after defer point. The keyspace reference is only used to check
how many replies a write operation should wait for and this can be
precalculated during write handler creation.
Fixes#2777
Message-Id: <20170911084436.GG24167@scylladb.com>
"This series tries to improve the bootstrap of a node in a large cluster by
improving how gossip applies the gossip node state. In #2404, the joining node
failed to bootstrap, because it did not see the seed node when
storage_service::bootstrap ran. After this series, we apply the whole gossip
state contained in the gossip ack/ack2 message before applying the next one,
and we apply the state of the seed node earlier than non-seed node so we can
have the seed node's state faster. We also add some randomness to the order of
applying gossip node state to prevent some of the nodes' state are always
applied earlier than the others.
This series improves apply_state_locally for large cluster:
- Tune the order of applying endpoint_state
- Serialize apply_state_locally
- Avoid copying of the gossip state map
Fixes#2404"
* tag 'asias/gossip_issue_2404_v2' of github.com:scylladb/seastar-dev:
gossip: Avoid copying with apply_state_locally
gossip: Serialize apply_state_locally
gossip: Tune the order of applying endpoint_state in apply_state_locally
gossip: Introduce is_seed helper
gossip: Pass const endpoint_state& in notify_failure_detector
gossip: Pass reference in notify_failure_detector
Move the std::map<inet_address, endpoint_state> map from the gossip
ack/ack2 message directly and move it around in apply_state_locally to
avoid copying the map.
apply_state_locally will be called when gossip ack/ack2
message is received. It will use the std::map<inet_address,
endpoint_state>& map to update the endpoint state.
However, we can receive multiple such gossip ack/ack2 messages from
multiple peer nodes in parallel. Currently, we process them in parallel.
It is better to apply all the states from one node then move to apply
all the states from another node than interleaving. Because it is more
important to have the state of the whole cluster than to have a bit
newer state from another peer (if it is newer), especially when the node
boots up and runs its first round of gossip exchange.
After this patch, we apply the whole gossip state contained in the
gossip ack/ack2 message before applying the next one.
We currently always apply the endpoint_state in the order of the
endpoint ip address. This is not good because some of the endpoint's
state is always applied earlier than the others.
In large cluster, the number of endpoints can be large, it takes time to
apply all of them. To make it more fair, we apply the endpoint_state
randomly.
Apply the seed node's state earlier because in bootstrap, we will check
if we have seen the seed node in storage_service::bootstrap. In #2404,
the bootstrap failed because, the joining node hasn't apply the seed
node's state when storage_service::bootstrap runs.
* seastar 85ca12d...31b925d (19):
> net/byteorder: fix 64 bit ntohq and htonq on big endian machines
> core, util: fix compilation on non-x86 processors
> core/memory: Fix SIGSEGV in small_pool::add_more_objects()
> log: remove debug leftovers
> Merge "TLS state machine fixes" from Calle
> logger: allow adjusting the timestamp style for stdout logs
> thread: make thread_context::s_main portable
> core: add seastar::cache_line_size constant
> Add detach() to input_stream and output_stream
> Install dependencies for Arch Linux.
> tls: Guard non-established sockets in sesrefs + more explicit close + states
> tls: Make vec_push fully exception safe
> basic_sstring: resize uses sstring
> Merge "Add and correct unit tests" from Jesse
> tcp: enforce 1-byte maximum segment invariant with zero window
> tcp: verify 1-byte maximum segment invariant during send with zero window
> memory: reduce small_pool vulnerability to fragmentation further
> Prometheus: avoid merging all metrics family
> net: Fix possible NULL pointer dereference.
validate_request_failure() assumed that the future returned by execute_cql()
is always ready, which doesn't have to be the case, and caused aborts
in debug mode build.
Message-Id: <1504701342-13300-1-git-send-email-tgrabiec@scylladb.com>
Scylla already refuses to load counter sstables that do not have Scylla
component. However, if this happens because of 'nodetool refresh'
command the existing protection will trigger after sstables have been
moved to the data directory. This is too later, so an additional check
is added when the upload directory is scanned.
has_scylla_component() is going to be used to verify that an sstable has
been generated by a recent version of Scylla. This would make it
possible to reject sstables that may be unsafe to load (e.g. sstables
containing legacy counter shards).
"This series implements the missing API to terminate all repairs.
For example:
$ curl -X POST --header "Accept: application/json"
"http://127.0.0.1:10000/storage_service/force_terminate_repair"
With the new stream_plan::abort() api we can now abort the stream
session assocaited with the repair as well.
On top of this, we can support termination of single repair job instead all
jobs.
Fixes#2105"
* tag 'asisas/repair_abort_v4' of github.com:scylladb/seastar-dev:
repair: Support termination of repair jobs
repair: Track repair_info
repair: Intorduce repair id to repair_info map
api: Add force_terminate_repair API
streaming: Add abort to stream_plan
streaming: Add abort_all_stream_sessions for stream_coordinator
streaming: Introduce streaming::abort()
streaming: Make stream_manager and coordinator message debug level
streaming: Check if _stream_result is valid
streaming: Log peer address in on_error
streaming: Introduce received_failed_complete_message
"Scylla 1.7.4 and older use incorrect ordering of counter shards, this
was fixed in 0d87f3dd7d ("utils::UUID:
operator< should behave as comparison of hex strings/bytes"). However,
that patch was not backported to 1.7 branch until very recently. This
means that versions 1.7.4 and older emit counter shards in an incorrect
order and expect them to be so. This is particularly bad when dealing
with imported correct sstables in which case some shards may become
duplicated.
The solution implemented in this patch is to allow any order of counter
shards and automaticly merge all duplicates. The code is written in a
way so that the correct ordering is expected in the fast path in order
not to excessively punish unaffected deployments.
A new feature flag CORRECT_COUNTER_ORDER is introduced to allow seamless
upgrade from 1.7.4 to later Scylla versions. If that feature is not
available Scylla still writes sstables and sends on-wire counters using
the old ordering so that it can be correctly understood by 1.7.4, once
the flag becomes available Scylla switches to the correct order.
Fixes #2752."
* tag 'fix-upgrade-with-counters/v2' of https://github.com/pdziepak/scylla:
tests/counter: verify counter_id ordering
counter: check that utils::UUID uses int64_t
mutation_partition_serializer: use old counter ordering if necessary
mutation_partition_view: do not expect counter shards to be sorted
sstables: write counter shards in the order expected by the cluster
tests/sstables: add storage_service_for_tests to counter write test
tests/sstables: add test for reading wrong-order counter cells
sstables: do not expect counter shards to be sorted
storage_service: introduce CORRECT_COUNTER_ORDER feature
tests/counter: test 1.7.4 compatible shard ordering
counters: add helper for retrieving shards in 1.7.4 order
tests/counter: add tests for 1.7.4 counter shard order
counters: add counter id comparator compatible with Scylla 1.7.4
tests/counter: verify order of counter shards
tests/counter: add test for sorting and deduplicating shards
counters: add function for sorting and deduplicating counter cells
counters: add counter_id::operator>
Until the cluster is fully upgraded from a version that uses the
incorrect counter shard ordering it is essential to keep using it lest
the old nodes corrupt the data upon receiving mutations with a counter
shard ordering they do not expect.
If the feature signaling that we have switched to the correct ordering
of counter shards is not enabled it means that the user still can do a
rollback to a version that expects wrong ordering. In order to avoid any
disasters when that happens write sstables using the 1.7.4 order until
we know for sure that it is no longer needed.
Scylla 1.7.4 used incorrect ordering of counter shards. In order to fix
this problem a new feature is introduced that will be used to determine
when nodes with that bug fixed can start sending counter shard in the
correct order.
Due to a bug in an implementation of UUID less compare some Scylla
versions sort counter shards in an incorrect order. Moreover, when
dealing with imported correct data the inconsistencies in ordering
caused some counter shards to become duplicated.
* 'tgrabiec/make-row-cache-update-exception-safe' of github.com:scylladb/seastar-dev:
row_cache: Improve safety of cache updates
row_cache: Extract invalidate_sync()
memtable: Mark mark_flushed() as noexcept
database: Add non-throwing try_trigger_compaction()
database: Make add_sstable() have strong exception guarantees
row_cache: Don't require presence checker to be supplied externally
database: Supply presence checker in sstable snapshots
mutation_source: Introduce mutation_source::make_partition_presence_checker()
mutation_reader: Move definitions up in the header
mutation_reader: Use constructor delegation to reduce code duplication
row_cache: Make populate() preserve continuity
row_cache: Allow marking as fully continuous on construction
database: Add missing serialization of sstable set udpate and cache invalidation
Cache imposes requirements on how updates to the on-disk mutation source
are made:
1) each change to the on-disk muation source must be followed
by cache synchronization reflecting that change
2) The two must be serialized with other synchronizations
3) must have strong failure guarantees (atomicity)
Because of that, sstable list update and cache synchronization must be
done under a lock, and cache synchronization cannot fail to synchronize.
Normally cache synchronization achieves no-failure thing by wiping the
cache (which is noexcept) in case failure is detect. There are some
setup steps hoever which cannot be skipped, e.g. taking a lock
followed by switching cache to use the new snapshot. That truly cannot
fail. The lock inside cache synchronizers is redundant, since the
user needs to take it anyway around the combined operation.
In order to make ensuring strong exception guarantees easier, and
making the cache interface easier to use correctly, this patch moves
the control of the combined update into the cache. This is done by
having cache::update() et al accept a callback (external_updater)
which is supposed to perform modiciation of the underlying mutation
source when invoked.
This is in-line with the layering. Cache is layered on top of the
on-disk mutation source (it wraps it) and reading has to go through
cache. After the patch, modification also goes through cache. This way
more of cache's requirements can be confined to its implementation.
The failure semantics of update() and other synchronizers needed to
change due to strong exception guaratnees. Now if it fails, it means
the update was not performed, neither to the cache nor to the
underlying mutation source.
The database::_cache_update_sem goes away, serialization is done
internally by the cache.
The external_updater needs to have strong exception guarantees. This
requirement is not new. It is however currently violated in some
places. This patch marks those callbacks as noexcept and leaves a
FIXME. Those should be fixed, but that's not in the scope of this
patch. Aborting is still better than corrupting the state.
Fixes#2754.
Also fixes the following test failure:
tests/row_cache_test.cc(949): fatal error: in "test_update_failure": critical check it->second.equal(*s, mopt->partition()) has failed
which started to trigger after commit 318423d50b. Thread stack
allocation may fail, in which case we did not do the necessary
invalidation.
Every mutation source can have a presence checker. By default all
answer "maybe contains".
Having this on mutation_source level will be useful for simplifying
cache update flow. The cache can ask the right snapshot for a presence
checker rather than relying on database to know when and how to make
the right one which preserves all invariants.
This will be especially useful once all updates of the underlying
mutation source of cache (e.g. sstable list) will have to go through
cache for safety reasons.
Commit e3ad676433 missed a few places.
It is required to serialize sstable list update and cache synchronization
in order to preserve partition update isolation.
Fixes#2746.
* dist/ami/files/scylla-ami b41e5eb...5ffa449 (3):
> amzn-main.repo: stick to Amazon Linux 2017.03 kernel (4.9.x)
> Prevent dependency error on 'yum update'
> scylla_create_devices: don't raise error when no disks found
This was part of "add gate for generic async operations to column family" but
somehow didn't make it into the final patch.
Add the missing piece.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170830164205.4497-1-glauber@scylladb.com>
"optional interposer that will check integrity of writes to sstable components.
The option name is enable_sstable_data_integrity_check, it's disabled by default
and can be enabled via config file.
It will provide enough details that will help to find the root of the issue.
if disk failed for example, we would've something like the following reported:
ERROR 2017-08-17 09:18:11,577 [shard 0] sstable - integrity check failed for ./data/data/system_schema/aggregates-924c55872e3a345bb10c12f37c1ba895/system_schema-aggregates-ka-111-Scylla.db, stage: read after write verification, write: 4096 bytes to offset 0, reason: data read from underlying storage isn't the same as written, mismatch at byte 0:
data written sample: 10000001000000010000001a00000001
date read sample: 00000000000000000000000000000000"
* 'integrity_check_interposer_v3' of github.com:raphaelsc/scylla:
sstables: optionally check integrity of sstable component writes
sstables: remove unneeded new_sstable_component_file variant
db/config: add sstable_data_integrity_check option
sstables: introduce file interposer for integrity check
If config file's sstable_data_integrity_check option is enabled,
new integrity check interposer will be used in addition to the
existing one. Performance is expected to drop because of all
the integrity checks for every write. This new interposer will
provide detailed info when integrity check fails.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
optional interposer that will check integrity when writing to
sstable components. It will provide enough details that will
help to find the root of the issue, which may come from lower
level layers.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
This patch implements the missing API to terminate all repairs.
For example:
$ curl -X POST --header "Accept: application/json"
"http://127.0.0.2:10000/storage_service/force_terminate_repair"
With the new stream_plan::abort() api we can now abort the stream
session assocaited with the repair as well.
Fixes#2105
The maps are stored in a vector. The vector has smp::count elements, each
element will be accessed by only one shard.
The add_repair_info, remove_repair_info and get_repair_info helpers
are added.
The api /storage_service/force_terminate is supposed to be
/storage_service/force_terminate_repair.
scylla-jmx uses /storage_service/force_terminate api.
So instead of renaming it, it is better to add a new name for it.
When we abort a session, it is possible that:
node 1 abort the session by user request
node 1 send the complete_message to node 2
node 2 abort the session upon receive of the complete_message
node 1 sends one more stream message to node 2 and the stream_manager
for the session can not be found.
It is fine for node 2 to not able to find the stream_manager, make the
log on node 2 less verbose to confuse user less.
It is the handler for the failed complete message. Add a flag to
remember if we received a such message from peer, if so, do not send
back the failed complete message back to the peer when running
close_session with failed status.
This reverts commit b56ba02335.
After commit 8fa35d6ddf (messaging_service: Get rid of timeout and retry
logic for streaming verb), streaming verb in rpc does not check if a
node is in gossip memebership since all the retry logic is removed.
Remove the extra wait before removing the joining node from gossip
membership.
Message-Id: <a416a735bb8aad533bbee190e3324e6b16799415.1504063598.git.asias@scylladb.com>
Thread stack allocation may fail, in which case we did not do the
necessary invalidation. Fix by hoisting the scope of the cleanup function.
Also fixes the following test failure:
tests/row_cache_test.cc(949): fatal error: in "test_update_failure": critical check it->second.equal(*s, mopt->partition()) has failed
which started to trigger after commit 318423d50b.
Message-Id: <1504023113-30374-2-git-send-email-tgrabiec@scylladb.com>
With the "Use range_streamer everywhere" (7217b7ab36) seires, all
the user of streaming now do streaming with relative small ranges and
can retry streaming at higher level.
There are problems with timeout and retry at RPC verb level in streaming:
1) Timeout can be false negative.
2) We can not cancel the send operations which are already called. When
user aborts the streaming, the retry logic keeps running for a long
time.
This patch removes all the timeout and retry logic for streaming verbs.
After this, the timeout is the job of TCP, the retry is the job of the
upper layer.
Message-Id: <df20303c1fa728dcfdf06430417cf2bd7a843b00.1503994267.git.asias@scylladb.com>
- Removed text from Report's "PURPOSE" section, which was referring to the "MANUAL CHECK LIST" (not needed anymore).
- Removed curl command (no longer using the api_address), instead using scylla --version
- Added -v flag in iptables command, for more verbosity
- Added support to for OEL (Oracle Enterprise Linux) - minor fix
- Some text changes - minor
- OEL support indentation fix + collecting all files under /etc/scylla
- Added line seperation under cp output message
Signed-off-by: Tomer Sandler <tomer@scylladb.com>
Message-Id: <20170828131429.4212-1-tomer@scylladb.com>
Large deques require contiguous storage, which may not be available (or may
be expensive to obtain). Switch to new custom container instead, which allocates
less contiguous storage.
Allocation problems were observed with the summary and compression info. While
there is work to reduce compression info contiguous space use, this solves
all std::deque problems (and should not conflict with that work).
Fixes#2708
* tag '2708/v6' of https://github.com/avikivity/scylla:
sstables: switch std::deque to chunked_vector
tests: add test for chunked_vector
utils: add a new container type chunked_vector
"Fixes #2734."
* 'tgrabiec/make-sstable-reader-work-with-empty-range-set' of github.com:scylladb/seastar-dev:
tests: Introduce clustering_ranges_walker_test
tests: simple_schema: Add missing include
sstables: reader: Make clustering_ranges_walker work with empty range set
clustering_ranges_walker: Make adjacency more accurate
Such queries can be issued by counter updates which involve only
static row.
Causes failure in test_query_only_static_row invoked from
sstable_mutation_test. See commit 6572f38, which fixed the problem in
cache reader.
Fixes#2734.
Current check considered some adjacent range tombstones as overlapping
with the ranges. Making this more accurate will become more important
after we will rely on putting p_i_p::after_all_clustered_rows() in
_current_start in out-of-range state.
Previously _current_bucket_segment_index was used differently depending on
whether update_position_trackers() is used in a random or sequential
access. In the former case was used as the absolute index of the segment
(independent of the buckets) and in the latter as the relative index of
the segment within its bucket. This caused problems when there was a
switch between random and sequential access, meaning one could get different
results for an at() call depending on what was the previous at() call.
Fix this by consistently using _current_bucket_segment_index as - like its
name suggest - the bucket relative segment index.
Ref #1946.
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <7f68ac1d32c80e8dea6dfa11be02acaa961bce2a.1503924927.git.bdenes@scylladb.com>
* 'deps1/v1' of https://github.com/avikivity/scylla:
types.hh: extract marshal_exception from types.hh into a new file
utils: remove dependency on types.hh
locator: add missing include "log.hh"
supervisor: remove dependency on init.hh
tracing: add missing include "log.hh"
gms: remove unneeded #include "types.hh"
* 'tgrabiec/fix-fast-forwarding' of github.com:scylladb/seastar-dev:
tests: mutation_source_test: Add more tests for fast forwarding across partitions
sstables: Fix abort in mutation reader for certain skip pattern
sstables: Fix reader returning partition past the query range in some cases
sstables: Introduce data_consume_context::eof()
The problem happens for the following sequence of events:
1) reader stops in the middle of some partition before it
skips to another partition range
2) reader is fast forwarded to a partition range which has no data in
the sstable. There are some partitions between the previous
partition range and the one we skip to
3) the reader is asked for next partition
The problem was that mutation_reader::fast_forward_to() was putting
the reader in _read_enabled == false state in step 2, but
data_consume_context was not fast forwarded to the range. When in step
3 we were asked for the next partition, we attempted to skip using
index (because of 1). The result of the skip was some position which
is outside of the current range of data_consume_context, which causes
it to abort. To fix, add a check for _read_enabled before we try to
skip.
If index was used to skip to the next partition (because the current
partition wasn't consumed in full) and reader's partition range ends
before the data file ends, we did not detect that we're out of range
before returning a streamed_mutation. Fix by checking _context.eof()
before doing that.
Refs #2733.
For better or worse, marshal_exception is used from utils/, and it's not good
to have utils/ depend on types.hh. Extract marshal_exception to make it possible
to remove the dependency.
* seastar 0083ee8...85ca12d (1):
> Merge "Run-time logging configuration" from Jesse
Includes patch from Jesse:
"Switch to Seastar for logging option handling
In addition to updating the abstraction layer for Seastar logging in `log.hh`,
the configuration system (`db/config.{hh,cc}`) has been updated in two ways:
- The string-map type for Boost.program_options is now defined in Seastar.
- A configuration value can be marked as `UsedFromSeastar`. This is like `Used`,
except the option is expected to be defined in the Boost.Program_options
description for Seastar. If the option is not defined in Seastar, or it is
defined with a different type, then a run-time exception is thrown early in
Scylla's initialization. This is necessary because logging options which are
now defined in Seastar were previously defined in Scylla and support for these
options in the YAML file cannot be dropped. In order to be able to verify that
options marked `UsedFromSeastar` are actually defined in Seastar, the
interface for adding options to `db::config` has changed from taking a
`boost::program_options::options_description_easy_init` (which is handle into
a `boost::program_options::options_description` which only allows adding
options) to taking a `boost::program_options::options_description`
directly (which also allows querying existing options).
Scylla also fully defers to Seastar's support for run-time logging
configuration."
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <ef26cffb91bef1ae95d508187a6dd861a6c4fc84.1503344007.git.jhaberku@scylladb.com>
"We have recently found two problems with the drop_column_family code
that needs addressing. The first is that exceptions in truncate() may
lead to stop() being skipped, which can cause Scylla to crash.
The other is that a truncate() issued before drop_column_family may get
the chance to execute only after the column family is already dropped
and also crash (That is issue 2726).
The second problem is the classic problem of asynchronous execution on
an object that may terminate, which we have been traditionally solving
with a gate. We add a gate to the column family that will be closed
during CF stop(), and we will require all asychronous operations to
enter it.
The immediate fix is for truncate(), where we have seen a real, concrete
problem. But it would be good to audit other code paths to make sure
that they are sane.
The most obvious ones, flush, compaction and sstable deletion are
already sane, since they are waited on explicitly during stop()."
Fixes#2726.
* 'issue-2726-v2-master' of github.com:glommer/scylla:
database: add gate for generic async operations to column family
database: make sure that column family is always stopped when dropped
We currently use std::deque<> for when we need large random-access containers,
but deque<> requires nr_items * sizeof(T) / 64 bytes of contiguous memory, which can
exceed our 256k fragmentation unit with large sstables. The new
container, which is a cross between deque and vector, has much lower
limitations.
Like deque, we allocate chunks of contiguous items, but they are
128k in size instead of 512. The last chunk can be smaller to avoid
allocating 128k for a really small vector.
run_with_compaction_disabled(), which is called by truncate, has a
pretty large defer point in remove(). When the code gets to finally
execute, we can't guarantee that the column family will still be alive.
That is true in particular if we issued a drop table command following
truncate: by the time truncate gets to resume, the CF will be gone.
Before the column family is dropped, it will always call its stop()
method, which means we have an opportunity to do some waiting there. We
already wait for flushes and current compactions to end.
Traditionally, we have been solving similar problems by adding a gate
that will catch asynchronous operations and making sure that potentially
asynchronous operations will enter the gate before executing. Let's do
the same thing here. We will close() the gate during stop().
Fixes#2726
Signed-off-by: Glauber Costa <glauber@scylladb.com>
truncate can throw exceptions. If it does, cf->stop() will never be
called because it is contained in a .then clause instead of finally.
One of the things that truncate does - in a finally block of its own -
is initiate a final compaction. If it returns an exception nobody will
wait for that compaction to finish (since cf->stop() is the one doing
that) and we'll crash.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
"Preserve the networking configuration mode during the upgrade by generating the /etc/scylla.d/perftune.yaml
file and using it."
Fixes#2725.
* 'dist_respect_cpuset_conf-v3' of https://github.com/vladzcloudius/scylla:
scylla_prepare: respect the cpuset.conf when configuring the networking
scylla_cpuset_setup: rm perftune.yaml
scylla_cpuset_setup: add a missing "include" of scylla_lib.sh
Choose the networking configuration mode according to the current contents of /etc/scylla.d/cpuset.conf.
If it doesn't exist - use the default mode.
If it exists - use the mode that has been used for generation of the CPU set.
Store the configuration into the /etc/scylla.d/perftune.yaml
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
scylla_setup resets our configuration and perftune.yaml is a part of it.
perftune.yaml is generated based on the contents of cpuset.conf therefore we should reset
these together.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
The scylla_cpuset_setup uses a verify_args() function that is defined in the scylla_lib.sh.
Fixes#2716
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
* seastar e96881a...b9f1eb7 (9):
> httpd: indentation patch
> httpd: handle exception when shutting down
> stall-detector: Allow backtrace throttling to be configured
> stall-detector: Fix messages about suppresssion not appearing
> scripts: posix_net_conf.sh: allow passing a perftune.py configuration file as a parameter
> scripts: perftune.py: add the possibility to pass the parameters in a configuration file and print the YAML file with the current configuration
> scripts: perftune.py: actually use the number of Rx queues when comparing to the number of CPU threads
> core: make current_backtrace() noexcept
> memory: add large allocation detector stubs for default allocator
Fixes#2723.
* tag 'asias/repair_issue_2723_v1' of github.com:cloudius-systems/seastar-dev:
repair: Do not allow repair until node is in NORMAL status
gossip: Add is_normal helper
The following backtrace was reported by user when running repair and keeping restarting the node at the same time.
#0 0x00007eff077281d7 in raise () from /lib64/libc.so.6
#1 0x00007eff07729a08 in abort () from /lib64/libc.so.6
#2 0x00007eff07721146 in __assert_fail_base () from /lib64/libc.so.6
#3 0x00007eff077211f2 in __assert_fail () from /lib64/libc.so.6
#4 0x00000000010ef2c2 in locator::token_metadata::first_token_index (this=0x641000214e98, start=...) at locator/token_metadata.cc:133
#5 0x00000000010ef2d9 in locator::token_metadata::first_token (this=0x641000214e98, start=...) at locator/token_metadata.cc:143
#6 0x00000000010e329d in locator::abstract_replication_strategy::get_natural_endpoints (this=0x641000494000, search_token=...)
at locator/abstract_replication_strategy.cc:66
#7 0x0000000001481186 in get_neighbors (hosts=std::vector of length 0, capacity 0, data_centers=std::vector of length 0, capacity 0,
range=<error reading variable: access outside bounds of object referenced via synthetic pointer>, ksname=..., db=...) at repair/repair.cc:196
#8 repair_range<nonwrapping_range<dht::token> > (range=..., ri=...) at repair/repair.cc:781
#9 <lambda(auto:99&)>::<lambda(auto:100&&)>::<lambda(auto:101&)>::<lambda()>::operator() (__closure=0x7efec07f7460) at repair/repair.cc:1005
#10 futurize<future<bool_class<stop_iteration_tag> > >::apply<repair_ranges(repair_info)::<lambda(auto:99&)>::
It is reproduced with
1) while true; do curl -X POST --header "Content-Type: application/json" --header "Accept: application/json" "http://127.0.0.1:10000/storage_service/repair_async/ks3"; done
2) start node 127.0.0.1, stop node 127.0.0.1 in a loop
The problem is, during boot up, the token_metadata is not replicated to all shards until
the node goes into NORMAL status.
To fix, check until node is in NORMAL status before allowing repair.
Fixes#2723
The number of keysapce and column family metrics reported is
proportional to the number of shards times the number of keysapce/column
families.
This can cause a performance issue both on the reporting system and on
the collecting system.
This patch adds a configuration flag (set to false by default) to enable
or disable those metrics.
Fixes#2701
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170821113843.1036-1-amnon@scylladb.com>
"Overly large metadata can hog memory which especially hurts in setups
with bad disk/memory ratio. To ease the pain compress the in-memory
compression-info.
The compression is implemented based on Avi's idea which is to group n
offsets together into segments, where each segment stores a base
absolute offset into the file, the other offsets in the segments being
relative offsets (and thus of reduced size). Also offsets are allocated
only just enough bits to store their maximum value. The offsets are thus
packed in a buffer like so:
arrrarrrarrr...
where n is 4, a is an absolute offset and r are offsets relative to a.
This of course means that stored offsets will not be aligned, not even
on a byte boundary, but the size reduction pretty convincing.
In addition, segments are stored in buckets, where each bucket has its
own base offset. In addition, segments in a buckets are optimized to
address as large of a chunk of the data as possible for a given chunk
size."
Ref #1946.
* 'bdenes/compress-compression-v3' of https://github.com/denesb/scylla:
Add unit test for compress::offsets
Optimise the storage of compression chunk offsets
Add script to precompute segmented compression parameters
To reduce the memory footprint of compression-info, n offsets are
grouped together into segments, where each segment stores a base
absolute offset into the file, the other offsets in the segments being
relative offsets (and thus of reduced size). Also offsets are
allocated only just enough bits to store their maximum value. The
offsets are thus packed in a buffer like so:
arrrarrrarrr...
where n is 4, a is an absolute offset and r are offsets relative to a.
The optimal value of n can be calculated for a given file_size (f) and
chunk_size (c), by finding the minima of the following function:
f(n) = (f/c)/n * (log2(f) + (n - 1)*log2((n-1)*(c + 64)))
This is done in an empirical way, using a script (see below).
Furthermore segments are stored in buckets, where each bucket has its
own base offset. Each bucket therefore can address an equal chunk of the
file and furthermore each segment in a bucket can address an equal
sub-chunk of this area.
The value of a given offset i is thus:
bucket_base_offset_for(i) + segment_base_offset_for(i) + offset(i)
To account for the bucketed storage we calculate a local_f, which is
optimized so that a bucketful of segmented offsets can address the
largest possible chunk of f. As value of this local_f only depends on
the bucket_size (b) and c the value of n can be made independent of f
and therefore only depend on one dynamic value, c. This makes life much
simpler as we don't need to know the size of the file up-front, we can
just append buckets to the storage on demand, while the required storage
is still less than a third [1] of the original storage requirements
(std::deque<uint64>).
The table with the minima(f(n)) for different f and c values is
pre-computed by gen_segmented_compress_params.py and
stored in sstables/segmented_compress_params.hh. This script also
creates a table with the best values of local_f for the given
bucket_size. At runtime we only select the best params based on c.
[1] This was calculated for c=4K and b=4K
Some (most?) users don't read logs or release notes, so they won't notice
that the ByteOrdered and Random partitioners were deprecated in 2.0. Make
them notice by refusing to start with a deprecated partitioner, unless a
switch is explicitly enabled.
Message-Id: <20170820073424.8331-1-avi@scylladb.com>
Limiting summary entry generation to at most one summary entry
per 64k of index data can lead to large index pages, with thousands
of index entries per summary entry. These are slow to parse, and there
is no real gain from the limit, since we already enforce a size
limit on the summary.
Remove the limit and allow summary entry generation based solely on
spanned data size.
Fixes#2711.
Message-Id: <20170819184255.14181-1-avi@scylladb.com>
split_range_to_single_shard() returns a vector of size 4096, with
each element (a partition_range) of size 100. The total of 400k can
cause defragmentation if memory is fragmented.
Fix by using a deque.
Fixes#2707.
Message-Id: <20170819141017.28287-1-avi@scylladb.com>
The script generates sstables/segmented_compress_params.hh which
contains a list with the optimal number of grouped offsets for
different data and chunk sizes as well as a list with the best
nominal data sizes for different chunk sizes, given a bucket size.
Data sizes are in the range of [2**4,2**50] and chunks in the
range of [2**4, 2**30]. Data sizes that are not used with the current
bucket_size are ommited.
See next commit for details of how the calculated values are used.
Large allocations can require cache evictions to be satisfied, and can
therefore induce long latencies. Enable the seastar large allocation
warning so we can hunt them down and fix them.
Message-Id: <20170819135212.25230-1-avi@scylladb.com>
Two reasons for this change:
1) every compaction should be multiplexed to manager which in turn
will make decision when to schedule. improvements on it will
immediately benefit every existing compaction type.
2) active tasks metric will now track ongoing reshard jobs.
Fixes#2671.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170817224334.6402-1-raphaelsc@scylladb.com>
"Now that range queries go through the normal digest path, we rely on
query::result::calculate_counts() to count the amount of partitions
and rows returned.
This series optimizes it, in case it is needed, and also changes the
result message to include the partition and row counts, avoiding the
calculation altogether."
* 'calculate-counts/v3' of github.com:duarten/scylla:
query-result: Send row and partition count over the wire
query::result: Optimize calculate_counts()
These two admin related packages will be packaged under the "app-admin"
category and not the "dev-db" one.
This fixes the detection path of the packages for scylla_setup.
Signed-off-by: Alexys Jacob <ultrabug@gentoo.org>
Message-Id: <20170817094756.21550-1-ultrabug@gentoo.org>
Scylla's configure.py contains stuff we copied from Seastar's
configure.py, but is no longer used. Let's get rid of some of it.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170813150842.12603-1-nyh@scylladb.com>
When user mistakenly forgot to pass parameter for a flag, our scripts misparses
next flag as the parameter.
ex) Correct usage is '--ntp-domain <domain> --setup-nic', but passed
'--ntp-domain --setup-nic'.
Result of that, next flag will ignore by scripts.
To prevent such behavior, reject any parameter that start with '--'.
Fixes#2609
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <20170815114751.6223-1-syuu@scylladb.com>
When compacting a partition for querying we would read an extra row,
to include any tombstones between that one and the previous row.
This is no longer needed since we have a general mechanism to detect
short reads in the storage_proxy.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170811103031.22866-1-duarte@scylladb.com>
incremental_reader_selector assumes the partition_range it receives has a lower
bound, but it was seen in mutation_test that this is not so.
Fix by checking whether the bound exists or not.
Message-Id: <20170815095852.14149-1-avi@scylladb.com>
"Exhausted readers belonging to a combined_mutation_reader can be fast
forwarded, so we have to keep them around. However, if the reader is
not fast forwardable, then we can drop the contained readers and their
buffers."
* 'ff-reader/v2' of github.com:duarten/scylla:
combined_mutation_reader: Drop exhausted readers if not in FF mode
combined_mutation_reader: Remove superfluous mutation_readers list
memtable_snapshot_source: Created readers should be fast forwardable
Exhausted readers can be fast forwarded, so we have to keep them
around. However, if the current reader is not fast forwardable, then
we can drop those readers and their buffers.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Now that range queries go through the normal digest path, we rely on
query::result::calculate_counts() to count the amount of partitions
and rows returned. This patch makes it a bit faster.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
* seastar 7a49ae5...edb73ab (11):
> scripts: perftune.py: change the network module mode auto selection heuristic
> net/tls: explicitly ignore ready future during shutdown
> Use python2 explicitly as an interpreter for Python v2 scripts
> peering_sharded_service: prevent over-run the container
> Add link to documentation to the README.md
> Add guidelines for contributing to Seastar
> sharded: fix move constructor for peering_sharded_service services
> Provide a convenient way to lazy-convert to string the values of pointers
> tutorial: overhaul semaphores section
> simple-stream: Make fragmented::write_substream return simple if possible
> simple-stream: Make simple/fragmented memory output stream top level
quantity prevents index_reader from reading all index entries of a summary
entry that span more than min_index_interval entries. That can happen after
introduction of size-based sampling, and consequently, sstable will not be
able to return a key which logical position in summary entry is beyond
min_index_interval. It's ok to not use quantity because index_reader will
read all indexes until either next summary entry or end of file is reached.
Fixes test_sstable_conforms_to_mutation_source
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170812045821.25269-1-raphaelsc@scylladb.com>
"Fixes #1842."
* 'size_based_sampling_v3' of github.com:raphaelsc/scylla:
tests: test summary entry spanning more keys than min interval
db/config: introduce sstable_summary_ratio option
sstables: introduce size-based sampling for sstable summary
sstables: make components_writer::offset const qualified and uint64_t
sstables: make writer::offset const qualified and uint64_t
"This patch series adds support for the `duration` type in CQL, which
was added to Cassandra in 3.10.
As part of this work, it was necessary also to add support for the
`vint` and `unsigned vint` types to the native protocol implementation,
which are part of v5 of the specification.
To test interactively, it is necessary to use cqlsh distributed with
Cassandra, as the version we distribute does not yet support the
duration type."
* 'jhk/duration_protocol/v5' of https://github.com/hakuch/scylla:
Support `duration` CQL native type
CQL native protocol: Add support for `vint` serialization
duration_test.cc: Add test for printing zero duration
duration.cc: Remove nop `const` qualifier on return type
Change `const` qualifier declaration order for `duration`
duration.cc: Simplify range checking
Rename `duration` to `cql_duration`
Now that we don't go directly to reconciliation for range queries, the
result isn't required to have the row and partition counts calculated
(we no longer transform a reconciled_result to a query::result).
Furthermore, this line was causing a lot of dtests to fail on account
of them not expecting an error line in the logs.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170810225351.12610-1-duarte@scylladb.com>
Currently, a summary entry is added after min_index_interval index
entries were written. Not taking into account size of index entries
becomes a problem with large partitions which may create big index
entries due to promoted indexes. Read performance is affected as a
consequence because index entries spanned by summary are all read
from disk to serve request.
What we wanna do is to also add a summary entry after index reaches
a boundary. To deal with oversampling, we want to write 1 byte to
summary for every 2000 bytes written to data file (this will be
eventually made into an option in the config file).
Both conditions must be met to avoid under or oversampling.
That way, the amount of data needed from index file to satify the
request is drastically reduced.
Fixes#1842.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
`duration` is a new native type that was introduced in Cassandra 3.10 [1].
Support for parsing and the internal representation of the type was added in
8fa47b74e8.
Important note: The version of cqlsh distributed with Scylla does not have
support for durations included (it was added to Cassandra in [2]). To test this
change, you can use cqlsh distributed with Cassandra.
Duration types are useful when working with time-series tables, because they can
be used to manipulate date-time values in relative terms.
Two interesting applications are:
- Aggregation by time intervals [3]:
`SELECT * FROM my_table GROUP BY floor(time, 3h)`
- Querying on changes in date-times:
`SELECT ... WHERE last_heartbeat_time < now() - 3h`
(Note: neither of these is currently supported, though columns with duration
values are.)
Internally, durations are represented as three signed counters: one for months,
for days, and for nanoseconds. Each of these counters is serialized using a
variable-length encoding which is described in version 5 of the CQL native
protocol specification.
The representation of a duration as three counters means that a semantic
ordering on durations doesn't exist: Is `1mo` greater than `1mo1d`? We cannot
know, because some months have more days than others. Durations can only have a
concrete absolute value when they are "attached" to absolute date-time
references. For example, `2015-04-31 at 12:00:00 + 1mo`.
That duration values are not comparable presents some difficulties for the
implementation, because most CQL types are. Like in Cassandra's implementation
[2], I adopted a similar strategy to the way restrictions on the `counter` type
are checked. A type "references" a duration if it is either a duration or it
contains a duration (like a `tuple<..., duration, ...>`, or a UDT with a
duration member).
The following restrictions apply on durations. Note that some of these contexts
are either experimental features (materialized views), or not currently
supported at run-time (though support exists in the parser and code, so it is
prudent to add the restrictions now):
- Durations cannot appear in any part of a primary key, either for tables or
materialized views.
- Durations cannot be directly used as the element type of a `set`, nor can they
be used as the key type of a `map`. Because internal ordering on durations is
based on a byte-level comparison, this property of Cassandra was intended to
help avoid user confusion around ordering of collection elements.
- Secondary indexes on durations are not supported.
- "Slice" relations (<=, <, >=, >) are not supported on durations with `WHERE`
restrictions (like `SELECT ... WHERE span <= 3d`). Multi-column restrictions
only work with clustering columns, which cannot be `duration` due to the
first rule.
- "Slice" relations are not supported on durations with query conditions (like
`UPDATE my_table ... IF span > 5us`).
Backwards incompatibility note:
As described in the documentation [4], duration literals take one of two
forms: either ISO 8601 formats (there are three), or a "standard" format. The ISO
8601 formats start with "P" (like "P5W"). Therefore, identifiers that have this
form are no longer supported.
Fixes#2240.
[1] https://issues.apache.org/jira/browse/CASSANDRA-11873
[2] bfd57d13b7
[3] https://issues.apache.org/jira/browse/CASSANDRA-11871
[4] http://cassandra.apache.org/doc/latest/cql/types.html#working-with-durations
Version 5 of the native protocol for CQL [1] adds the `vint` and `unsigned vint`
types.
An unsigned integer encoded as a `vint` has a variable size based on the
magnitude of the value. The first byte indicates the total number of bytes.
For signed integers, a "zig-zag" encoding scheme ensures that small negative
values are encoded as short-length `vint`s (0 -> 0, -1 -> 1, 1 -> 2, 2 -> 3, -2
-> 4, etc).
[1] https://github.com/apache/cassandra/blob/trunk/doc/native_protocol_v5.spec
"sstables will sometimes have narrow/disjont ranges (e.g. LCS L1+).
This can be exploited when reading from a range of sstables by opening
sstables on-demand thus saving memory, processing and potentially I/O.
To achieve this combined_mutation_reader is refactored such that the
reader selection logic is moved-out into a reader_selector class.
combined_mutation_reader now takes a reader_selector instance in its
constructor and asks it for new readers for the current ring position
on every call to operator()().
At the moment two specializations of reader_selector are provided:
* list_reader_selector which implements the current logic, that is using
a provided mutation_reader list, and
* incremental_reader_selector which implements the on-demand opening
logic discussed above.
Fixes#1935"
* 'bdenes/optimize_combined_reader-v6' of https://github.com/denesb/scylla:
Add combined_mutation_reader_test unit test
Remove range_sstable_reader
Add incremental_reader_selector
Add reader_selector to combined_mutation_reader
sstable_set::incremental_selector: select() now returns a selection
incremental_reader_selector is a specialization of reader_selector for
the case when sstables have narrow and/or disjoint token ranges. To
exploit this it creates new readers on-demand when their sstable's
token range intersects with the current ring position.
combined_mutation_reader now accepts as a constructor argument a
reader_selector instance whoose task is to create new readers on
each call to operator()() if needed and possible.
This way it is possible to control how readers are created through
different specializations of reader_selector.
The previous logic is refactored into list_reader_selector which
is using a pre-provided mutation_reader list and forwards all of them to
combined_mutation_reader at once.
We are moving to aptly to release .deb packages, that requires debian repository
structure changes.
After the change, we will share 'pool' directory between distributions.
However, our .deb package name on specific release is exactly same between
distributions, so we have file name confliction.
To avoid the problem, we need to append distribution name on package version.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1502312935-22348-1-git-send-email-syuu@scylladb.com>
A seletion contains - in addition to the list of sstables - a next_token
which is a hint as to what is the next best token to call select() with.
This should be the smallest token such that at the next call to
select() the least number of new sstables will be returned, without
skipping any.
"With this series, all the following cluster operations:
- bootstrap
- rebuild
- decommission
- removenode
will use the same code to do the streaming.
The range_streamer is now extended to support both fetch from and push
to peer node. Another big change is now the range_streamer will stream
less ranges at a time, so less data, per stream_plan and range_streamer
will remember which ranges are failed to stream and can retry later.
The retry policy is very simple at the moment it retries at most 5 times
and sleep 1 minutes, 1.5^2 minutes, 1.5^3 minutes ....
Later, we can introduce api for user to decide when to stop retrying and
the retry interval.
The benefits:
- All the cluster operation shares the same code to stream
- We can know the operation progress, e.g., we can know total number of
ranges need to be streamed and number of ranges finished in
bootstrap, decommission and etc.
- All the cluster operation can survive peer node down during the
operation which usually takes long time to complete, e.g., when adding
a new node, currently if any of the existing node which streams data to
the new node had issue sending data to the new node, the whole bootstrap
process will fail. After this patch, we can fix the problematic node
and restart it, the joining node will retry streaming from the node
again.
- We can fail streaming early and timeout early and retry less because
all the operations use stream can survive failure of a single
stream_plan. It is not that important for now to have to make a single
stream_plan successful. Note, another user of streaming, repair, is now
using small stream_plan as well and can rerun the repair for the
failed ranges too.
This is one step closer to supporting the resumable add/remove node
opeartions."
* tag 'asias/use_range_streamer_everywhere_v4' of github.com:cloudius-systems/seastar-dev:
storage_service: Use the new range_streamer interface for removenode
storage_service: Use the new range_streamer interface for decommission
storage_service: Use the new range_streamer interface for rebuild
storage_service: Use the new range_streamer interface for bootstrap
dht: Extend range_streamer interface
We experienced 'Constructing RAID volume...' takes too much time on some AMIs,
this is because setup script stuck at 'yum -y install mdadm xfsprogs'.
We don't have to install these packages on AMI startup time, we should
preinstall them on AMI creating time.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1502192796-21040-1-git-send-email-syuu@scylladb.com>
* Configuration for cluster_name is commented-out in config file.
* Default value set to empty string and if not rewritten by user then
warning is printed and value is reset to "ScyllaDB Cluster".
Fixes#2648.
Message-Id: <20170808113322.9313-1-daniel@scylladb.com>
ScyllaDB loves python & python loves ScyllaDB.
It would benefit the project to start enforcing some code guidelines
and basic QA with a linter along a PEP8 respect thanks to flake8.
This patch adds a tox config to at least start with an assessment
of the work to be done on all .py files in the code base.
To reduce its noise, tests on long lines (> 80char) are ignored
for now.
Signed-off-by: Ultrabug <ultrabug@gentoo.org>
Message-Id: <20170726134242.8927-1-ultrabug@gentoo.org>
index's file output stream uses write behind but it's not closed
when sstable write fails and that may lead to crash.
It happened before for data file (which is obviously easier to
reproduce for it) and was fixed by 0977f4fdf8.
Fixes#2673.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170807171146.10243-1-raphaelsc@scylladb.com>
After this patch and the following patches to use the new
range_streamder interface, all the following cluster operations:
- bootstrap
- rebuild
- decommission
- removenode
will use the same code to do the streaming.
The range_streamer is now extended to support both fetch from and push
to peer node. Another big change is now the range_streamer will stream
less ranges at a time, so less data, per stream_plan and range_streamer
will remember which ranges are failed to stream and can retry later.
The retry policy is very simple at the moment it retries at most 5 times
and sleep 1 minutes, 1.5^2 minutes, 1.5^3 minutes ....
Later, we can introduce api for user to decide when to stop retrying and
the retry interval.
The benefits:
- All the cluster operation shares the same code to stream
- We can know the operation progress, e.g., we can know total number of
ranges need to be streamed and number of ranges finished in
bootstrap, decommission and etc.
- All the cluster operation can survive peer node down during the
operation which usually takes long time to complete, e.g., when adding
a new node, currently if any of the existing node which streams data to
the new node had issue sending data to the new node, the whole bootstrap
process will fail. After this patch, we can fix the problematic node
and restart it, the joining node will retry streaming from the node
again.
- We can fail streaming early and timeout early and retry less because
all the operations use stream can survive failure of a single
stream_plan. It is not that important for now to have to make a single
stream_plan successful. Note, another user of streaming, repair, is now
using small stream_plan as well and can rerun the repair for the
failed ranges too.
This is one step closer to supporting the resumable add/remove node
opeartions.
* seastar f14d2a3...7a49ae5 (8):
> sharded: improve support for cooperating sharded<> services
> sharded: support for peer services
> semaphore: add a version of with_semaphore that takes a duration timeout
> scripts: perftune.py: fix the CPU mask generation for more than 64 CPUs
> Revert "future-utils: make when_all() (vector variant) exception safe"
> Revert "future-utils: fix gross compilation errors in when_all()"
> future-utils: fix gross compilation errors in when_all()
> future-utils: make when_all() (vector variant) exception safe
Includes change to batchlog_manager constructor to adapt it to
seastar::sharded::start() change.
This reverts commit 98757069a5. We have the
failure detector which will detect an unresponsive node and fail the RPC.
Adding a timeout can just introduce false positives.
In commit f38e4ff3f, we have separated streaming reads from normal reads
for the purpose of determining the maximum number of reads going on.
However, we'll now be totally unaware of how many reads will be
happening on behalf of streaming and that can be important information
when debugging issues.
This patch adds this metric so we don't fly blind.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1501909973-32519-1-git-send-email-glauber@scylladb.com>
We have problem to run fstrim with nomerges=2, so we need to change
the parameter to 1 during fstrim execution.
To do this, this fix changes follow things:
- revert dropping scylla_fstrim on Ubuntu 16.04/CentOS
- disable distribution provided fstrim script
- enable scylla_fstrim on all distributions
- introduce --set-nomerges on scylla-blocktune
- scylla_fstrim call scylla-blocktune by following order:
- 'scylla-blocktune --set-nomerges 1'
- 'fstrim' for each devices
- 'scylla-blocktune --set-nomerges 2'
Fixes#2649
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1501531393-21109-1-git-send-email-syuu@scylladb.com>
Streaming reads and normal reads share a semaphore, so if a bunch of
streaming reads use all available slots, no normal reads can proceed.
Fix by assigning streaming reads their own semaphore; they will compete
with normal reads once issued, and the I/O scheduler will determine the
winner.
Fixes#2663.
Message-Id: <20170802153107.939-1-avi@scylladb.com>
If we fail a streaming read due queue overload, we will fail the entire repair.
Remove the limit for streaming, and trust the caller (repair) to have bounded
concurrency.
Fixes#2659.
Message-Id: <20170802143448.28311-1-avi@scylladb.com>
"Currently scanning reads go to reconciliation stage directly which
requires asking for mutation data from all peers. This series makes
it to try matching digests first like a single partition read."
Fixes#2666.
* 'gleb/digest_scan' of github.com:cloudius-systems/seastar-dev:
storage_proxy: make range_slice_read_executor go through digest matching state
storage_proxy: add capability to read data/digest for non singular ranges
storage_proxy: remove redundant parameter from never_speculating_read_executor constructor
Currently scanning reads go to reconciliation stage directly which
requires asking for mutation data from all peers. This patch makes
it to try matching digests first like a single partition read.
The change requires internode protocol changes since currently it is not
possible to ask for multi partition data/digest over RPC. It means that
the capability has to be guarded by new gossip feature flag which the
patch also adds.
Using --hostname to give the container a meaningful name is a good
practice, and make the monitoring dashboard easier to understand
Signed-off-by: Tzach Livyatan <tzach@scylladb.com>
Message-Id: <20170803081027.6675-1-tzach@scylladb.com>
never_speculating_read_executor always waits for all targets so
block_for parameter is always equal to targets.size(). No need to
to pass it explicitly.
"This series adds an option to use paging in internal query and use that for the
get compaction history function.
Internal paging will be done explicitly, to use paging, you first create a
state object (that contains the query as well) and use that state to get the
first page, the result will contain both the query result and a new state that
can be used to get the next page.
Fixes#2366"
* 'amnon/paged_compaction_history_v5' of github.com:cloudius-systems/seastar-dev:
system_keyspace: Use paging for get compaction history
Add paging for internal queries
query_options: Allows creating query_options from query_options
"This series tries to fix possible repair stuck."
Fixes#2660, #2661, #2662.
* tag 'asias/repair_stuck_v2.1' of github.com:cloudius-systems/seastar-dev:
repair: Make send_repair_checksum_range timeout
repair: Singal parallelism_semaphore in case of error
repair: Fix repair_tracker done
It specifies the maximum gossip shadow round time. It can be used to
reduce the gossip feature check time during node boot up.
For instance, when the first node in the cluster, which listed both
itself and other node as seed in the yaml config, boots up, it will try
to talk to other seed nodes which are not started yet. The gossip shadow
round will be used to fetch the feature info of the cluster. Since there
is no other seed node in the cluster, the shadow round will fail. User
can reduce the default shadow_round_ms option to reduce the boot time.
Fixes#2615
Message-Id: <10916ce9059f3c7f1a1fb465919ae57de3b67d59.1500540297.git.asias@scylladb.com>
The timer is armed inside the section guarded by the _timer_reads_gate
therefore it has to be canceled after the gate is closed.
Otherwise we may end up with the armed timer after stop() method has
returned a ready future.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1501603059-32515-1-git-send-email-vladz@scylladb.com>
"This series ensures the always write correct cell names to promoted
index cell blocks, taking into account the eoc of range tombstones.
Fixes#2333"
* 'pi-cell-name/v1' of github.com:duarten/scylla:
tests/sstable_mutation_test: Test promoted index blocks are monotonic
sstables: Consider eoc when flushing pi block
sstables: Extract out converting bound_kind to eoc
It was meant to be run in the foreground since it is waited upon during
stop(), but as it is now from the stop() perspective it is completed
after first connection is accepted.
Fixes#2652
Message-Id: <20170801125558.GS20001@scylladb.com>
Queries with query page size equal or smaller than
zero are unpaged queries.
Count these kind of queries and make them a metrics
since they can ruin the performance of the system.
Message-Id: <20170731130004.25807-2-benoit@scylladb.com>
"This series reduce that effect in two ways:
1. Remove the latency counters from the system keyspaces
2. Reduce the histogram size by limiting the maximum number of buckets and
stop the last bucket."
Fixes#2650.
* 'amnon/remove_cf_latency_v2' of github.com:cloudius-systems/seastar-dev:
database: remove latency from the system table
estimated histogram: return a smaller histogram
Since the discovery of std::exchange(x, {}) move_and_clear has become
obsolete. Beside, the name was wrong, it did not clear the vector but
recreated it meaning that any allocated memory wasn't reused (not that
it mattered in the existing usages).
Message-Id: <20170731123549.10887-1-pdziepak@scylladb.com>
"These patches optimise combined_mutation_reader for cases where the majority
of mutation_readers is disjoint.
perf_fast_forward:
Results are medians of 3 of fragments/s as reported by perf_fast_forward.
Command:
perf_fast_forward -c1 --enable-cache
small: small-partition-skips (read=1, skip=0)
large: large-partition-skips (read=1, skip=0)
before after diff
small 195753 238196 +22%
large 1244325 1359096 +9%
perf_simple_query:
Results are medians of 10 of reads/s as reported by perf_simple_query.
Command:
perf_simple_query -c1
before 98651.40
after 104554.85
diff +6%"
* tag 'avoid-merge_mutations/v1' of https://github.com/pdziepak/scylla:
combined_mutation_reader: avoid unnecessary merge_mutations()
combined_mutation_reader: do not pop mutation with different key
"This series ensure that when we retry a memtable flush, we re-acquire the
flush permit that was previously released. It also ensures we don't hold
the sstable read lock for the duration of the sleep leading to the retry.
To achieve that cleanly we refactor the way the permit lifecycle is managed
by employing a RAII-based approach.
We also improve the latency of writes blocked on virtual dirty by releasing
the flush permit before fsyncing the sstables. There are additional avenues
for performance improvements on top of this one."
* 'memtable-flush-additional-fixes/v4' of github.com:duarten/scylla:
column_family: Re-acquire flush permit in case of error
column_family: Don't hold sstable read lock when retrying flush
sstables: Release the flush permit before fsyncing
sstables: Introduce write_monitor
database: Extract out dirty_memory_manager
dirty_memory_manager: Refactor flush permit lifetime management
dirty_memory_manager: Invert permit acquisition order
memtable_list: Register different seal functions for each behaviour
Merging mutations is quite an expensive operation. The creation of
streamed mutation merger involves several allocations (mostly coming
from various std::vector) and then all mutation_fragments need to go
through a heap.
All this is completely unnecessary if there is only one mutation, so
let's skip a call to merge_mutations() in such cases. This also means
that we can reuse memory allocated by _current vector if merge is not
required.
Originally, the loop insidecombined_mutation_reader::next() so that it
was popping mutation from the heap and when it encountered one with a
different decorated key it was pushed back and the ones accumulated so
far merged and emitted. In other words, every time the reader progressed
to the next mutation it did needless pop and push operations on the
heap.
This patch rearranges the code so that the key of the next mutation is
compared before it is popped from the heap.
If we fail to flush an sstable, after creating the flush_reader, then
we will have released the flush permit when we retry the flush. Ensure
that when retrying, we re-acquire the flush permit.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This allows a queued flush to start while we fsync the current
sstable, which helps reduce the overall time new writes are blocked on
dirty memory.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The write_monitor provides callbacks to inform an observer of the
state of the ongoing sstable write.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch refactors how the flush permit lifetime is managed,
dropping the current hash table in favour of a RAII approach.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
For an upcoming fix it is required to invert the permit acquisition
order: first we acquire the background work permit and then the single
flush permit.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Instead of passing a flush_behaviour to the seal function, use two
different functions for each of the behaviours.
This will be important in the forthcoming patches, which will require
the signatures of those functions to differ.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
serialized_action_tests depends on the fact that first part of the
serialized_action is executed at cetrtain points (in which it reads a
global variable that is later updated by the main thread).
This worked well in the release mode before ready continuations were
inlined and run immediately, but not in the debug mode since inlining
was not happening and the main seastar::thread was missing some yield
points.
Message-Id: <20170731103013.26542-1-pdziepak@scylladb.com>
1. assert() is not constexpr.
2. can't use static_assert(), because the contructor may be called in a non-constexpr
environment; moved to log_histogram
3. pow2_rank() uses count_leading_zeros() which is not constexpr; split
into constexpr and non-constexpr versions
4. duplicated number_of_buckets() because bucket_of() can't be constexpr due to pow2_rank
Message-Id: <20170726105444.32698-1-avi@scylladb.com>
"This series ensure that when we retry a memtable flush, we re-acquire the
flush permit that was previously released. It also ensures we don't hold
the sstable read lock for the duration of the sleep leading to the retry.
To achieve that cleanly we refactor the way the permit lifecycle is managed
by employing a RAII-based approach.
We also improve the latency of writes blocked on virtual dirty by releasing
the flush permit before fsyncing the sstables. There are additional avenues
for performance improvements on top of this one."
* 'memtable-flush-additional-fixes/v3' of github.com:duarten/scylla:
column_family: Re-acquire flush permit in case of error
column_family: Don't hold sstable read lock when retrying flush
sstables: Release the flush permit before fsyncing
sstables: Introduce write_monitor
database: Extract out dirty_memory_manager
dirty_memory_manager: Refactor flush permit lifetime management
dirty_memory_manager: Invert permit acquisition order
memtable_list: Register different seal functions for each behaviour
main: Don't catch polymorphic exceptions by value
* seastar a14d667...54e940f (8):
> Merge "Prometheus to use output stream" from Amnon
> http_test: Fix an http output stream test
> build: harden try_compile_and_link output temporary file
> configure: disable exception scalability hack on debug build
> build: don't perform test compiles to /dev/null
> Provide workaround for non scaleable c++ exception runtime
> Merge "Add output stream to http message reply" from Amnon
> configure.py: use user provided compiler flags when checking for features
Commitlog replay fails when upgrade from <1.7.3 to 2.0, we need to refuse
updating package if current scylla < 1.7.3 && commitlog remains.
Note: We have the problem on scylla-server package, but to prevent
scylla-conf package upgrade, %pretrans should be define on scylla-conf.
Fixes#2551
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1501187555-4629-1-git-send-email-syuu@scylladb.com>
"This patch series backports scalability fix for _Unwind_Find_FDE and modifies
out CentOS package to use our libgcc and libstdc++ which are needed to make
use of the fix instead of locally installed ones."
Ref #2646 (fixes on RHEL 7 and related only)
* 'gleb/exception-gcc-fix-v2' of github.com:cloudius-systems/seastar-dev:
dist/redhat: Make scylla rpm depend on scylla-libgcc and scylla-libstdc++ and use it instead of locally installed one
dist/redhat: Backport scalability fix of _Unwind_Find_FDE to out gcc
If we fail to flush an sstable, after creating the flush_reader, then
we will have released the flush permit when we retry the flush. Ensure
that when retrying, we re-acquire the flush permit.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This allows a queued flush to start while we fsync the current
sstable, which helps reduce the overall time new writes are blocked on
dirty memory.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The write_monitor provides callbacks to inform an observer of the
state of the ongoing sstable write.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch refactors how the flush permit lifetime is managed,
dropping the current hash table in favour of a RAII approach.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
For an upcoming fix it is required to invert the permit acquisition
order: first we acquire the background work permit and then the single
flush permit.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Instead of passing a flush_behaviour to the seal function, use two
different functions for each of the behaviours.
This will be important in the forthcoming patches, which will require
the signatures of those functions to differ.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
If schema merging completes at lower rate than incoming pull requests,
then merge processes will accumulate and needlessly request and hold schema mutations.
In rare cases, when there are constant schema changes, they may even
overflow memory. This was seen in dtest:
concurrent_schema_changes_test.py:TestConcurrentSchemaChanges.create_lots_of_schema_churn_test
Allowing only one active and one queued pull request per remote
endpoint is enough.
* tag 'tgrabiec/dont-accumulate-schema-pulls-v2' of github.com:scylladb/seastar-dev:
migration_manager: Log schema pulls
migration_manager: Prevent pull requests from accumulating
utils: Introduce serialized_action
If schema merging completes at lower rate than incoming pull requests,
then merge processes will accumulate and needlessly request and hold schema mutations.
In rare cases, when there are constant schema changes, they may even
overflow memory. This was seen in dtest:
concurrent_schema_changes_test.py:TestConcurrentSchemaChanges.create_lots_of_schema_churn_test
Allowing only one active and one queued pull request per remote
endpoint is enough.
When flushing a promoted index block using a range tombstone cell name
as a bound, use the right eoc value instead of always writing
composite::eoc::none.
Fixes#2333
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
consume_mutation_fragments_until() allows consuming mutation fragments
until a specified condition happens. This patch reorganises its
implementation so that we avoid situations when fill_buffer() is called
with stop condition being true.
Message-Id: <20170727122218.7703-1-pdziepak@scylladb.com>
When we install scylla metapackage with version (ex: scylla-1.7.1),
it just always install newest scylla-server/-jmx/-tools on the repo,
instead of installing specified version of packages.
To install same version packages with the metapackage, limited dependencies to
current package version.
Fixes#2642
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <20170726193321.7399-1-syuu@scylladb.com>
Commitlog replay fails when upgrade from <1.7.3 to 2.0, we need to refuse
updating package if current scylla < 1.7.3 && commitlog remains.
Note: We have the problem on scylla-server package, but to prevent
scylla-conf package upgrade, %pretrans should be define on scylla-conf.
Fixes#2551
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <20170727110730.613-1-syuu@scylladb.com>
database::make_sstable_reader() creates a reader which will need to
obtain a semaphore permit when invoked. Therefore, each read may
create at most one such reader in order to be guaranteed to make
progress. If the reader tries to create another reader, that may
deadlock (or for non-system tables, timeout), if enough number of such
readers tries to do the same thing at the same time.
Avoid the problem by dropping previous reader before creating a new
one.
Refs #2644.
Message-Id: <1501152454-4866-1-git-send-email-tgrabiec@scylladb.com>
Initialize the system_auth and system_traces keyspaces and their tables after
the Node joins the token ring because as a part of system_auth initialization
there are going to be issues SELECT and possible INSERT CQL statements.
This patch effectively reverts the d3b8b67 patch and brings the initialization order
to how it was before that patch.
Fixes#2273
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1500417217-16677-1-git-send-email-vladz@scylladb.com>
This patch remove the latency histograms from the system table, it also
extend the already existing exclusion to all system keyspaces.
It also uses the new get_histogram API to set a minimal bucket size to
100 microseconds.
The current histogram contains 91 buckets, this is a very high
resolution with a high upper limit.
To reduce traffic passed, between scylla and the prometheus, this patch
generate a smaller histogram.
It limit the number of buckets (16 by default), set a lower limit to the
lowest bucket, and uses 2 as the bucket coeficient.
Highest empty buckets will not be reported.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
estimated histogram
These patches contain some minor fixes for performance regression reported
by perf_fast_forward after partial cache was merged. The solution is still
far from perfect, there is one case that still has 30% degradation, but
there is some improvement so there is no reason to hold these changes back.
Refs #2582.
Some numbers:
before - before cache changes were merged
(555621b537)
cache - at the commit that introduced the partial cache
(9b21a9bfb6)
after - recent master + this series
(based on e988121dbb)
Differences are shown relative to "before".
Testing effectiveness of caching of large partition, single-key slicing reads:
Large partitions, range [0, 500000], populating cache
before cache after
1636840 1013688 1234606
-38% -25%
Large partitions, range [0, 500000], reading from cache
before cache after
2012615 3076812 3035423
+53% +51%
Testing scanning small partitions with skips.
reading small partitions (skip 0)
before cache after
227060 165261 200639
-27% -11%
skipping small partitions (skip 1)
before cache after
29813 27312 38210
-8% +28%
Testing slicing small partitions:
slicing small partitions (offset 0, read 4096)
before cache after
195282 149695 180497
-23% -8%
* https://github.com/pdziepak/scylla.git perf_fast_forward-regression/v3:
sstables: make sure that fill_buffer() actually fills buffer
mutation_merger: improve handling of non-deferring fill_buffer()s
partition_snapshot_row_cursor: avoid apply() in single-version cases
sstables: introduce decorated_key_view
ring_position_comparator: accept sstables::decorated_key_view
sstable: keep a pre-computed token in summary_entry
sstables: cache token in index entries
index_reader: advance_and_check_if_present() use index_comparator
ring_position_comparator: drop unused overloads
cache_streamed_mutation: avoid moving clustering_row
streamed_mutation: introduce consume_mutation_fragments_until()
cache_streamed_mutation: use consumer based read_context reader
rows_entry: make position() inlineable
mutation_fragment: make destructor always_inline
keys: introduce compound_wrapper::from_exploded_view()
sstables: avoid copying key components
compound_compat: explode: reserve some elements in a vector
cache: short-circut static row logic if there are no static columns
cache: use equality comparators instead of tri_compare
sstables: avoid indirect calls to abstract_type::is_multi_cell()
loading_cache invokes a timer that may issue asynchronous operations
(queries) that would end with writing into the internal fields.
We have to ensure that these operations are over before we can destroy
the loading_cache object.
Fixes#2624
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1501096256-10949-1-git-send-email-vladz@scylladb.com>
We need to consider the _live_endpoints size. The nr_live_nodes should
not be larger than _live_endpoints size, otherwise the loop to collect
the live node can run forever.
It is a regression introduced in commit 437899909d
(gossip: Talk to more live nodes in each gossip round).
Fixes#2637
Message-Id: <863ec3890647038ae1dfcffc73dde0163e29db20.1501026478.git.asias@scylladb.com>
Equality comparator may be much cheaper than the fully fledged
trichotomic comparator, especially if the component types are byte order
equal but not byte order comparable.
When we are exploding a compound key we know already that there is more
than one component, but we have no easy way of determining how many of
them are going to be there. Let's reserve space for a few elements so
that we avoid an excessive number of reallocations in case of
medium-sized keys.
mutation_fragment destructor was already made inline-friendly by moving
most of the logic to a separate function. However, the compiler still is
quite reluctant to inline it in certain cases, so let's give it a
stronger hint.
consume_mutation_fragments_until() is a consumer based interface that
avoids indirect calls and continuation overhead present in the naive
streamed_mutation::operator() approach.
clustering_row can stores quite a lot of data internally which makes its
move constructor not exactly cheap.
If possible it is better to move mutation_fragment around as it keeps
everything externally. This also avoids some cases when clustering row
would be extracted from mutation_fragment only to be made to create
another mutation_fragment later.
When a sstable reader is fast forwarded some index entries may be read
(and compared) multiple times. This patch makes sure that once a token
is computed we keep it around and reuse if the entry is accessed again.
Each sstable index lookup involves a binary search in the summary and
each time a partition key of summary entry is compared with anything its
token needs to be calculated.
Since we keep summary in the memory all the time it is better to also
keep the tokens around.
ring_position_comparator has overloads for comparing ring_positions as
well as sstables::key_view. In the case of the latter it needs to
compute the token of the key. However, the sstable layer could cache
some tokens so let's allow the comparator callers to provide it
directly.
It is possible that a call to fill_buffer() will return an immediately
ready future. This patch avoids uncontrolled recursion in case when all
merged streamed mutation do not defer ini fill_buffer() and also
optimises for non-deferring case by avoiding some of the logic.
streamed_mutation::impl::fill_buffer() is supposed to either push
mutation fragments to the buffer or set EOS flag. However, it was
possible that mp_row_consumer would return proceed::no if a skip was
needed without satisfying any of these conditions.
"This patch series addresses some feedback from the preliminary
HACKING.md, adds some new content, and updates the README file with
some quick-start information."
* 'jhk/better_hacking/v3' of github.com:hakuch/scylla:
README.md: Add quick-start section and defer to `HACKING.md`
HACKING.md: `CMakeLists.txt` for analysis works for other IDEs too
HACKING.md: Add details and examples for unit tests
HACKING.md: Add section for project dependencies
HACKING.md: Describe releases and tags
HACKING.md: Re-work "building" section, including memory needs
HACKING.md: Update ccache recommendations
HACKING.md: Update "Contributing" URL
If a node is notified of a schema change where the schema's dropped
columns have changes, that node will miss the changes to the dropped
columns. A scenario where this can happen is where a column c is
dropped, then added as a different typed, and then dropped again, with
a node n having seen the first drop and being notified of the
subsequent add and drop.
Fixes#2616
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170725170622.4380-1-duarte@scylladb.com>
Counters write path on leader is completely different than on any other
replica (non-leaders share write path between counters and regular
columns). This patch makes sure that counter writes performed on leader
are added to appropriate metrics.
Message-Id: <20170725153346.31238-1-pdziepak@scylladb.com>
* git@github.com:raphaelsc/scylla.git row_cache_fixes:
db: atomically synchronize cache with changes to the snapshot
db: refresh row cache's underlying data source after compaction
Empty clustering key range is perfectly valid and signifies that the
reader is not interested in anything but the static row. Let's not
make it mean anything else.
Message-Id: <20170725131220.17467-2-pdziepak@scylladb.com>
cache_streamed_mutation assumed that at least one clustering range was
specified. That was wrong since the readers are allowed to query just
for a static row (e.g. counter update that modifies only static
columns).
Fixes#2604.
Message-Id: <20170725131220.17467-1-pdziepak@scylladb.com>
To support both version and patch release, the version server now returns
a patchversion parameter that include the latest minor version's patch
release.
The housekeeping should return a separate message if the current
minor version is not with the latest patch release, and a message if the version was
changed.
For example, if a user is using version 1.6.1 it should get a warning
that he need to update if 1.6.2 is available and in addition a warning it
should upgrade if version 1.7 is out.
Examples:
$ scylla-housekeeping version --version 1.6.2
Your current Scylla release is 1.6.2, while the latest patch release is 1.6.4, and the latest minor release is 1.7.2 (recommended)
$ scylla-housekeeping version --version 1.7.1
You current Scylla release is 1.7.1 while the latest patch release is 1.7.2 is available, update for the latest bug fixes
$ scylla-housekeeping version --version 1.7.1
You current Scylla release is 1.7.1 while the latest patch release is 1.7.2, update for the latest bug fixes and improvements
Fixes#1972
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Acked-by: Tzach Livyatan <tzach@scylladb.com>
Message-Id: <20170725095455.6450-1-amnon@scylladb.com>
Underlying data source in row cache holds a reference to sstable set
prior to compaction which isn't released until a memtable flush, which
means file descriptors of deleted sstables remains opened, wasting
disk space.
The fix is to refresh underlying data source in row cache.
Fixes#2570.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
updates to cache and snapshot (i.e. sstable set) aren't synchronized, so
it may happen that cache update for memtable flush will use wrong snapshot
version, and that violates cache invariant of each partition entry only
reflecting one snapshot.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Boost 1.55 accidentally removed support for "range for" on
recursive_directory_iterator (previous and latter versions do
support it). Use old-style iteration instead.
Message-Id: <20170724080128.8824-1-avi@scylladb.com>
The 'free' attribute is not updated for all pages belonging to a large
object, so we can't use it to determine if the page is allocated or
not. More reliable way is to check if it belongs to any free span.
Message-Id: <1500648094-20039-1-git-send-email-tgrabiec@scylladb.com>
This is in order to avoid frequent misses which have a relatively high
cost. A miss means we need to fetch schema definition from another
node and in case of writes do a schema merge.
If the schema is kept alive only by the incoming request, then it
will be forgotten immediately when the request is done, and the next
request using the same schema version will miss again.
Refs #2608.
Message-Id: <1500632447-10104-1-git-send-email-tgrabiec@scylladb.com>
"Fixes issues uncovered in longevity test (#2608).
Main problem is that due to time drift scylla_tables.version column
may not get deleted on all nodes doing the schema merge, which will
make some nodes come up with different table schema version than others.
The inconsistency will not heal because scylla_tables doesn't
take part in the schema sync. This is fixed by the last patch.
This will cause nodes to constantly try to sync the schema, which under
some conditions triggers #2617."
* tag 'tgrabiec/fix-table-schema-version-inconsistency-v1' of github.com:scylladb/seastar-dev:
schema_tables: Add scylla_tables to ALL
schema: Make schema_mutations equality consistent with digest
schema_tables: Extract compact_for_schema_digest()
schema_tables: Always drop scylla_tables::version
global_schema_ptr ensures that schema object is replicated to other
cores on access. It was replicating the "synced" state as well, but
only when the shard didn't know about the schema. It could happen that
the other shard has the entry, but it's not yet synced, in which case
we would fail to replicate the "synced" state. This will result in
exception from mutate(), which rejects attempts to mutate using an
unsynced schema.
The fix is to always replicate the "synced" state. If the entry is
syncing, we will preemptively mark it as synced earlier. The syncing
code is already prepared for this.
Refs #2617.
Message-Id: <1500555224-15825-1-git-send-email-tgrabiec@scylladb.com>
there could be a lot of compactions when querying for compaction
history.
This patch changes the query to use paging. It would collect all results
when returning to the caller.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Usually, internal queries are used for short queries. Sometimes though,
like in the case of get compaction history, there could be a large
amount of results. Without paging it will overload the system.
This patch adds the ability to use paging internally.
Using paging will be done explicitely, all the relevant information
would be store in an internal_query_state, that would hold both the
paging state but also the query so consecutive calls can be made.
To use paging use the query method with a function.
The function gets beside a statement and its parameters a function that
will be used for each of the returned rows.
For example if qp is a query_processor:
qp.query("SELECT * from system.compaction_history", [] (const cql3::untyped_result_set::row& row) {
....
// do something with row
...
return stop_iteration::no; // keep on reading
});
Will run the function on each of the compaction history table rows.
To stop the iteration, the function can return stop_iteration::yes.
* https://github.com/pdziepak/scylla.git perf_fast_forward_improvements/v1:
perf_fast_forward: move global state to global scope
perf_fast_forward: move tests groups to separate functions
perf_fast_forward: allow running only selected test groups
perf_fast_forward: use consumer interface for reading
streamed_mutation
So that scylla_tables takes part in the digest and in mutations sent
as part of schema sync. Otherwise inconsistencies in scylla_tables
will not heal.
Refs #2608.
Digest only looks like live values, ignoring deletion
information. Equality should be consistent with that, so that schemas
considered equal do not trigger the alter path unnecessarily.
It can happen that due to time drift between nodes, the incoming
"version" cell will have higher timestamp than api::new_timestamp().
In such case the column would not be dropped and would cause version
mismatch between nodes.
Ensure it's always covered by using max of current time and cell's
timestamp.
Refs #2608.
By default build_deb.sh destroys all previous build image to make sure we don't
have environment dependent issue, but it's takes time to build distribution root
image (.tgz in pbuilder) from scratch.
--no-clean option is for skipping create .tgz stage, use previously built image,
to make build time shorter.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1500542094-12946-1-git-send-email-syuu@scylladb.com>
Using streamed_mutation::operator() is undesirable as it introduces an
indirect call and a continuation overhead for each emitted mutation
fragment. Consumer interface is the preferred method of reading streamed
mutations.
All test perf_fast_forward test cases currently live in the main
function. This patch moves the state they rely on to a global scope
so that it will be easier to extract these tests to individual
functions.
"This patchset restricts background writers - such as compactions,
streaming flushes and memtable flushes to a maximum amount of CPU usage
through a seastar::thread_scheduling_group.
The said maximum is recommended to be set 50 % - it is default
disabled, but can be adjusted through a configuration option until we
are able to auto-tune this.
The second patch in this series provides a preview on how such auto-tune
would look like. By implementing a simple controller we automatically
adjust the quota for the memtable writer processes, so that the rate at
which bytes come in is equal to the rates at which bytes are flushed.
Tail latencies are greatly reduced by this series, and heavy spikes that
previously appeared on CPU-bound workloads are no more."
* 'memtable-controller-v5' of https://github.com/glommer/scylla:
simple controller for memtable/streaming writer shares.
restrict background writers to 50 % of CPU.
Some places remained where code looked directly at
system_keyspace::NAME to determine iff a ks is
considered special/system/protected. Including
schema digest calculation.
Export "is_system_keyspace" and use accordingly.
Message-Id: <1500469809-23546-1-git-send-email-calle@scylladb.com>
"Fixes schema layout incompatibility in a mixed 1.7 and 2.0 cluster (#2555)
by reverting back to using the old layout in memory and thus also
in across-node requests. We still use the new v3 layout in schema
tables (needed by drivers and external tools). Translations happen
when converting to/from schema mutations."
* tag 'tgrabiec/use-v2-schema-layout-in-memory-v2' of github.com:scylladb/seastar-dev:
schema: Revert back to the 1.7 layout of static compact tables in memory
schema: Use v3 column layout when converting to/from schema mutations
schema: Encapsulate column layout translations in the v3_columns class
"This series improves the streaming error handling so that when one side of the
streaming failed, it will propagate the error to the other side and the peer
will close the failed session accordingly. This removes the unnecessary wait and
timeout time for the peer to discover the failed session and fail eventually.
Fix it by:
- Use the complete message to notify peer node local session is failed
- Listen on shutdown gossip callback so that we can detect the peer is shutdown
can close the session with the peer
Fixes#1743"
* tag 'asias/streaming/error_handling_v2' of github.com:cloudius-systems/seastar-dev:
streaming: Listen on shutdown gossip callback
gms: Add is_shutdown helper for endpoint_state class
streaming: Send complete message with failed flag when session is failed
streaming: Handle failed flag in complete message
streaming: Do not fail the session when failed to send complete message
streaming: Introduce send_failed_complete_message
streaming: Do not send complete message when session is successful
streaming: Introduce the failed parameter for complete message
streaming: Remove unused session_failed function
streaming: Less verbose in logging
streaming: Better stats
Due to the asynchronous nature of view update propagation, results
might still be absent from views when we query them. To be able to
deterministically assert on view rows, this patch retries a query a
bounded number of times until it succeeds.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170718212646.2958-1-duarte@scylladb.com>
We are using C* 3.x compatible layout in schema tables but want to
keep using the 1.7 layout in memory for compatibility during rolling
upgrade. This patch switches the schema and schema_builder classes
back to the old layout. Translation of layout happens when converting
to/from schema mutations.
Notable changes:
1) Includes a revert of commit 6260f31e08
"thrift: Update CQL mapping of static CFs".
2) Brings back the "default_validation_class" schema attribute. In v3
it can be dervied from column definitions, but in v2 it can't, so
we have to store it.
3) legacy_schema_migrator and schema_builder don't have to do
conversions to v3, this is now handled by the v3_columns
class. schema_builder works with the same layout as schema, that
is v2.
4) Includes a revert of commit 66991a7ccb
"v3 schema test fixes"
Fixes#2555.
"Time window strategy was introduced to address several limitations of
date tiered strategy. In addition, its options are much easier to reason
about, basically just window size and window unit.
TWCS will work to keep only one sstable in each window. So the only real
optimization needed is to align partition key to the window.
Size tiered strategy is used to reduce write amplification when compacting
the incoming window.
For more details: https://issues.apache.org/jira/browse/CASSANDRA-9666
Fixes #1432."
* 'twcs_v2' of github.com:raphaelsc/scylla:
tests: add tests for time window compaction strategy
compaction: wire up time window compaction strategy
compaction/twcs: override default values with options in schema
sstables: implement time window compaction strategy
sstables: import TimeWindowCompactionStrategy.java
This patch provides a rather trivial implementation of hash() for
collection types.
It is needed for view building, where we hold mutations in a map
indexed by partition keys (and frozen collection types can be part of
the key).
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170718192107.13746-1-duarte@scylladb.com>
This patch introduces a simple controller that will adjust memtables CPU
shares, trying to keep it around the soft limit: if we start going below
it means we're too fast (unless we are idle) and shares are adjusted
downwards. If we start going above it means we're too fast and shares
are adjusted upwards.
I have tested this extensively in a single-CPU setup with various
CPU-bound workloads while tracking virtual dirty and the results are
good, with virtual dirty fluctuating only slightly, somewhere within the
desired range.
Exceptions to this are:
1) when the load is very light - the idle system goes faster, and that's
ok
2) when the load is very high - as foreground requests dominate we can't
flush fast enough and hit the hard limit. However, in such scenarios
the memtable shares do hit its maximum, and the results are no worse
than they are right now and this will only be fixed by CPU-limiting the
actual requests.
This feature can be disabled with a config option - that is scheduled to
go away as we acquire more confidence in this. When the feature is
disabled, all background writers (streaming, compaction, memtables) will
share the same scheduling group, with static quotas.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
In scylla, we have foreground processes, which are latency sensitive and
need to be responded to as fast as possible in order to maintain good
latency profiles, and background process, which are less so.
The most important background processes we have during normal write
workload operations are memtable writes and sstable compactions. Those
processes are quite CPU-intensive, and left unchecked will easily
dominate the CPU. Lower values of task-quota usually help, as it will
force those processes to preempt more, but aren't enough to guarantee
good isolation. We have seen boxes with good NVMe storage having their
throughput reduced to less than half of the original baseline in a short
dive down for the duration of a compaction.
In the long run, our goal is to leverage the CPU scheduler to make sure
that those processes are balanced with respect to all the others.
However, the current state of affairs is causing grievances as this very
moment. Thankfully, those processes live in a seastar::thread, that
ships with its own rudimentary bandwidth control mechanism: the
scheduling group.
The goal of this patch is to wrap background processes together in a
scheduling group, and assign to such group 50 % of our CPU power; the
remainder being left to foreground processes.
While we pride ourselves in dynamically adjusting things to the
workload, we won't be able to do this properly before the CPU scheduler
lands - and let's face it, leaving background processes run wild is not
adaptative either. Every workload would benefit most from a different
value for such shares, but 50 % is as fair as it gets if we really need
static partitining in the mean time.
As a defense against unforeseen consequences, we'll leave the actual
value as an option, but will do our best to hide it - as this is not a
tunable that we want to be part of a normal Scylla setup. The most
convenient place for this tunable is still db::config, so we can easily
pass it down to the database layer - but we will not document it in the
yaml, and will clearly note in the help string that it is not supposed
to be tuned.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
When a node shutdown itself, it will send a shutdown status to peer
nodes. When peer nodes receives the shtudown status update, they are
supposed to close all the sessions with that node becasue the node is
shutdown, no need to wait and timeout, then fail the session.
This change can speed up the closing of sessions.
Currently, send_complete_message is not used. We will use it shortly in
case the local session is failed. Send a complete message with failed
flag to notify peer node that the session is failed so that peer can
close the session. This can speed up the closing of failed session.
Also rename it to send_failed_complete_message.
it will be later converted to C++. Imported from latest scylla-
tools-java repository. Checked that it doesn't lack anything.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
A user reported scylla-server.service does not able to run on their cloud instance, because of hugeadm.
(hugeadm says the kernel does not support huge pages.)
We don't need it for posix mode, so move it in dpdk mode.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1500367219-8728-1-git-send-email-syuu@scylladb.com>
Instead of retrying, just drop mutations that raced with a truncate.
* git@github.com:duarten/scylla.git truncate-reorder/v1:
database: Rename replay_position_reordered_exception
database: Drop mutations that raced with truncate
The complete_message is not needed and the handler of this rpc message
does nothing but returns a ready future. The patch to remove it did not
make into the Scylla 1.0 release so it was left there.
Use this flag to notify the peer that the session is failed so that the
peer can close the failed session more quickly.
The flag is used as a rpc::optional so it is compatible use old
version of the verb.
It is useful for larger cluster with larger gossip message latency. By
default the fd_max_interval_ms is 2 seconds which means the
failure_detector will ignore any gossip message update interval larger
than 2 seconds. However, in larger cluster, the gossip message udpate
interval can be larger than 2 seconds.
Fixes#2603.
Message-Id: <49b387955fbf439e49f22e109723d3a19d11a1b9.1500278434.git.asias@scylladb.com>
branch 'tgrabiec/schema-migration-fixes' of github.com:scylladb/seastar-dev:
schema: Use proper name comparator
legacy_schema_migrator: Properly migrate non-UTF8 named columns
schema_tables: Store column_name in text form
legacy_schema_migrator: Migrate columns like Cassandra
schema_builder: Add factory method for default_names
legacy_schema_migrator: Simplify logic
thrift: Don't set regular_column_name_type
schema: Use proper column name type for static columns
schema: Fix column_name_type() for static compact tables
schema: Introduce clustering_column_at()
thrift: Reuse cell_comparator::to_sstring() for obtaining comparator type
partition_slice_builder: Use proper column's type instead of regular_column_name_type()
This replaces column_definition::name_comparator, which incorrectly
assumes that names are always utf8, with name_compare moved from
schema::rebuild() and unifies usages.
Currently migrator assumed all columns are utf8-named, which
doesn't have to be the case for static compact tables.
Refs #2597.
Due to #2573, we can assume that Scylla wasn't used with non-utf8
column names, and that old names are always in textual form.
This fixes generation of synthetic columns for static compact tables.
Current code always generates synthetic clustering column with utf8
type and synthetic regular column with bytes type (in schema_builder).
That's fine when creating a new CQL table, but not when migrating
existing tables created via thrift API.
Fixes#2584.
This also migrates empty compact value columns like Cassandra
does. Such columns are present in compact tables without regular
columns, e.g.:
create table test (k int, ck int, primary key (k, ck)) with compact storage;
They should be migrated to a synthetic regular column with
empty_type type and a non-empty name.
The expression "is_dense.value_or(true)" is always true inside the if,
so drop it.
This allows us to drop temporary calulated_is_dense.
We can also get rid of one of the if branches by extracting
builder.set_is_dense() outside.
After f5dae826ce, static columns not
always have utf8 column names. For static compact tables it's
determined by the cell name comparator type, which is equal to the
type of the synthetic clustering column.
Caused various errors with static thrift tables with non-utf8
comparator.
Reduces view_schema_test runtime to 5 seconds, from 53 seconds on an NVMe disk
with write-back cache, and forever on a spinning disk.
Message-Id: <20170716081653.10018-1-avi@scylladb.com>
We will be creating links to those sstable's files, and those don't work
if the data directory and the test sstable are on different devices.
Copying the files to the same directory fixes the problem.
Message-Id: <20170716090405.14307-1-avi@scylladb.com>
Rename replay_position_reordered_exception to
mutation_reordered_with_truncate_exception for more precision, since
this is the only situation where this exception can be thrown.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We don't ensure mutations are applied in memory following the order of their
replay positions. A memtable can thus be flushed with replay position rp,
with the new one being at replay position rp', where rp' < rp. This breaks
an intrinsic assumption in the code, which this series addresses.
Fixes#2074
branch memtable-flush/v3 of git@github.com:duarten/scylla.git:
commitlog: Always flush latest memtable
column_family: More precise count of switched memtables
column_family: Fix typo in pending_tasks metric name
column_family: More precise count of pending flushes
dirty_memory_manager: Remove unnecessary check from flush_one()
column_family: Don't rely on flush_queue to guarantee flushes finished
column_family: Don't bother closing the flush_queue on stop()
column_family: Stop using flush_queue
column_family: Remove outdated comment about the flush_queue
memtable: Stop tracking the highest flushed rp
Clang is happy to create a vector<data_value> from a {}, a {1, 2}, but not a {1}.
No doubt it is correct, but sheesh.
Make the data_value explicit to humor it.
Message-Id: <20170713074315.9857-1-avi@scylladb.com>
In storage_proxy we arrange the mutations sent by the replicas in a
vector of vectors, such that each row corresponds to a partition key
and each column contains the mutation, possibly empty, as sent by a
particular replica.
There is reconciliation-related code that assumes that all the
mutations sent by a particular replica can be found in a single
column, but that isn't guaranteed by the way we initially arrange the
mutations.
This patch fixes this and enforces the expected order.
Fixes#2531Fixes#2593
Signed-off-by: Gleb Natapov <gleb@scylladb.com>
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170713162014.15343-1-duarte@scylladb.com>
Since we no longer enforce that mutations are applied in memory
ordered by their replay_positions, the way the highest_flush_rp is
being tracked is no longer correct.
The invariant it was used to maintain no longer exists, so we can get
rid of it together with the assertion on the highest_flush_rp on
flush().
Fixes#2074
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Since commitlog ordering requirements have been relaxed, we now keep
the set of replay_positions seen by a memtable in a set, which we then
use to clean up relevant segments in the commitlog. This means that
the guarantees provided by the flush_queue are no longer necessary.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
When stopping a column family we issue a flush(), for which we wait.
Since writes are supposed to have stopped coming in, and also new
flush requests, there's no need to call and wait for the flush_queue
to be closed.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We now don't ensure mutations are applied in memory following the
order of their replay positions, so we can't rely on the replay
position to order memtable flushes. So, use a phased_barrier() to
ensure that calling flush() returns a future that completes when all
flushes up to that point have finished.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We don't need to check whether a memtable is empty in flush_one(), as
that must be checked later, during the actual sealing.
The condition itself is rare and is checked already after the potentially
contented semaphore has been acquired.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures we update the count of pending flushes in the same
place as we update the stats across column families, which is more
correct since it only accounts for actual flushes and not those of
empty memtables or that have been coalesced together.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The memtable_switch_count metric is supposed to count the number of
times a flush has resulted in the memtable being switched out, but we
were incrementing the count regardless of whether we tried to flush an
empty memtable or two or more flushes were coalesced into one. This
patch fixes this by moving the metric to where the memtable is
actually switched.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We now don't ensure mutations are applied in memory following the
order of their replay positions, so we can't rely on the replay
position to order memtable flushes. When flushing commit log segments,
ensure we flush the latest memtable.
Refs #2074
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"This series aims to fix the "serving invalid (old) values" issue in the
loading_cache (issue #2590) by arming the timer with a period that equals
min(expire, refresh).
We are still trying to optimize the main case where 'expire' is
significantly longer than 'refresh' period.
We don't want to add any additional logic in the fast path and this
series gives the immediate solution for the issue above while not adding
any additional CPU cycle to the fast path."
* 'loading_cache_short_expired-v2' of https://github.com/vladzcloudius/scylla:
utils::loading_cache: arm the timer with a period equal to min(_expire, _update)
utils::loading_cache: make a timer use a loading_cache_clock_type clock as a source
Make the descriptions of permissions_validity_in_ms, permissions_update_interval_in_ms
and permissions_cache_max_entries more readable and more related to what they really
do.
Mention the none-zero value requirement for the permissions_update_interval_in_ms and
the permissions_cache_max_entries when the permissions cache is enabled.
Adjust the parameters description in the scylla.yaml too.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1499957053-31792-1-git-send-email-vladz@scylladb.com>
Arm the timer with a period that is not greater than either the permissions_validity_in_ms
or the permissions_update_interval_in_ms in order to ensure that we are not stuck with
the values older than permissions_validity_in_ms.
Fixes#2590
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Current algorithm was marking tables with regular columns not named
"value" as not dense, which doesn't have to be the case. It can be
either way.
It should be enough to look at clustering components. If there is a
clustering key, then table is dense if and only if all comparator
components belong to the clustering key.
If there is no clustering key, then if there are any regular columns
we're sure it's not dense.
Fixes#2587.
Message-Id: <1499877777-7083-1-git-send-email-tgrabiec@scylladb.com>
Cassandra 3.10 added the `duration` type [1], intended to manipulate date-time
values with offsets (for example, `now() - 2y3h`).
The full implementation of the `duration` type in Scylla requires support
for version 5 of the binary protocol, which is not yet available.
In the meantime, this patch patch adds the implementation of the underlying type
for the eventual `duration` type. Included is also the ported test suite from
the reference implementation and additional tests.
Related to #2240.
[1] https://issues.apache.org/jira/browse/CASSANDRA-11873
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <b1e481da103efee82106bf31f261c5a1f4f8d9ca.1499885803.git.jhaberku@scylladb.com>
query_options object cannot be changed after it was created. For
internal uses, like internal query paging, it is needed to create a new
object based on some of the data from an existing one with a new paging
state.
This patch adds a constructor from a unique_ptr and paging state.
using unique_ptr behave similar to move modify constructor.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Since base tables no longer look for their views, we need to parse
base tables first so that when we add a view we can fetch and connect
it to its base table.
When announcing view table mutations to other nodes we always include
the base table mutations, so there's no need to expect a view being
added before its base table.
Found out while testing view building.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170712172115.2960-1-duarte@scylladb.com>
"Currently new nodes calculate digests based on v3 schema mutations,
which are very different from v2 mutations. As a result they will
use schemas with different table_schema_version that the old nodes.
The old nodes will not recognize the version and will try to request
its definition. That will fail, because old nodes don't understand
v3 schema mutations.
To fix this problem, let's preserve the digests during migration,
so that they're the same on new and old nodes. This will allow
requests to proceed as usual.
This does not solve the problem of schema being changed during
the rolling upgrade. This is not allowed, as it would bring the
same problem back.
Fixes #2549."
* tag 'tgrabiec/use-consistent-schema-table-digests-v2' of github.com:cloudius-systems/seastar-dev:
tests: Add test for concurrent column addition
legacy_schema_migrator: Set digest to one compatible with the old nodes
schema_tables: Persist table_schema_version
schema_tables: Introduce system_schema.scylla_tables
schema_tables: Simplify read_table_mutations()
schema_tables: Resurrect v2 read_table_mutations()
system_keyspace: Forward-declare legacy schemas
legacy_schema_migrator: Take storage_proxy as dependency
Use name of the existing preceeding column with restriction
(last_column) instead of assuming that the column right after the
current column already has restrictions.
This will yield an error message that is different from that of
Cassandra, albeit still a correct one.
Fixes#2421
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <40335768a2c8bd6c911b881c27e9ea55745c442e.1499781685.git.bdenes@scylladb.com>
DowngradingConsistencyRetryPolicy uses live replicas count from
Unavailable exception to adjust CL for retry, but when there are pending
nodes CL is increased internally by a coordinator and that may prevent
retried query from succeeding. Adjust live replica count in case of
pending node presence so that retried query will be able to proceed.
Fixes#2535
Message-Id: <20170710085238.GY2324@scylladb.com>
"most of changes are to improve maintainability of the strategy but
the ones that are introduced by the following patches:
lcs: do not check if level 0 can be promoted twice
lcs: remove quadratic behavior from L0 compaction
lcs: partially sort candidates that will be trimmed
lcs: only demote sstable from level higher than target one"
* 'lcs_improvements_2' of github.com:raphaelsc/scylla:
lcs: only demote sstable from level higher than target one
lcs: improve indentation for get_overlapping_starved_sstables
lcs: improve indentation for get_compaction_candidates
lcs: partially sort candidates that will be trimmed
lcs: remove quadratic behavior from L0 compaction
lcs: introduce private interface
lcs: make some member functions static
lcs: make some functions const qualified
lcs: remove add method
lcs: extract code for higher levels compaction from get_candidates_for
lcs: simplify code to get candidates for higher levels
lcs: extract round-robin heuristic for even distribution of keys into function
lcs: update outdated comments for level 0 compaction
lcs: improve worth_promoting_L0_candidates interface
lcs: do not check if level 0 can be promoted twice
lcs: extract code for level 0 compaction from get_candidates_for
CQL reply may contain metadata that describes columns present in the
response including the information about their type.
However, Scylla incorrectly reports counter types as bigint. The
serialised format of counters and bigint is exactly the same, which
could explain why the problem hasn't been noticed earlier but it is a
bug nevertheless.
Fixes#2569.
Message-Id: <20170711130520.27603-1-pdziepak@scylladb.com>
Calculate and set digest using v2 mutations so that digests are the
same before and after migration. This is neeed so that no schema
definition exchange is required during rolling upgrade.
Fixes#2549.
When migrating schema tables from v2 to v3, mutations underlying
table schema will change, and so will their digest. However, we want
the digest to be the same on new nodes as on the old nodes, because
schema exchange is not possible between the two nodes, so they
must to request schema definitions from each other.
The solution is to make the digest persistable, so that it sticks to
given table schema, surviving both migration and node restarts. On
migration from v2, the digest will be calculated from v2 mutations, so
it will be the same on new and old nodes.
It will be used to store Scylla spcific table metadata. We cannot
store it in the standard "tables" table for compatibility reasons -
Cassandra will fail to read schema if it encounteres columns it is not
expecting.
if we are compacting level 1 into level 2, we only want to demote
a sstable from level 3 or higher.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
L0 compaction triggers quadratic behavior when many newly created
sstables are needed for promotion due to their size being relatively
low to max sstable size parameter. So until L0 is worth promoting,
the strategy will compact every new sstable with all the existing
ones in L0. To fix it, let's do STCS on level 0 until it becomes
worth promoting.
Fixes#2432.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
get rid of unneeded loop for dealing with suspect sstables and
std::advance because vector allows random access.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
some comments are no longer relevant, especially the ones that
talk about dealing with busy sstables due to parallel compaction,
which isn't done by us for lcs.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
can_promote flag will be used to carry info about whether or not
level 0 can promoted. That will avoid a single iteration for higher
levels too which can contain tens of thousands of sstables.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
I will split code for higher levels compaction into functions first
before putting it into its own function too.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
The default of 2ms is somewhat arbitrary. Now that we have a lot more
mileage deploying Scylla applications in production it does sound not
only arbitrary, but high.
In particular, it is really hard to achieve 1ms latencies in the face of
CPU-heavy workloads with it.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1499354495-27173-1-git-send-email-glauber@scylladb.com>
Otherwise we may deadlock, as explained in commit 5e8f0efc8:
Table drop starts with creating a snapshot on all shards. All shards
must use the same snapshot timestamp which, among other things, is
part of the snapshot name. The timestamp is generated using supplied
timestamp generating function (joinpoint object). The joinpoint object
will wait for all shards to arrive and then generate and return the
timestamp.
However, we drop tables in parallel, using the same joinpoint
instance. So joinpoint may be contacted by snapshotting shards of
tables A and B concurrently, generating timestamp t1 for some shards
of table A and some shards of table B. Later the remaining shards of
table A will get a different timestamp. As a result, different shards
may use different snapshot names for the same table. The snapshot
creation will never complete because the sealing fiber waits for all
shards to signal it, on the same name.
Message-Id: <1499762663-21967-1-git-send-email-tgrabiec@scylladb.com>
"most of changes are to improve maintainability of the strategy but
the ones that are introduced by the following patches:
lcs: do not check if level 0 can be promoted twice
lcs: remove quadratic behavior from L0 compaction
lcs: partially sort candidates that will be trimmed
lcs: only demote sstable from level higher than target one"
* 'lcs_improvements' of github.com:raphaelsc/scylla: (21 commits)
lcs: only demote sstable from level higher than target one
lcs: improve indentation for get_overlapping_starved_sstables
lcs: improve indentation for get_compaction_candidates
lcs: partially sort candidates that will be trimmed
lcs: remove quadratic behavior from L0 compaction
lcs: introduce private interface
lcs: make some member functions static
lcs: make some functions const qualified
lcs: remove add method
lcs: extract code for higher levels compaction from get_candidates_for
lcs: simplify code to get candidates for higher levels
lcs: extract round-robin heuristic for even distribution of keys into function
lcs: update outdated comments for level 0 compaction
lcs: improve worth_promoting_L0_candidates interface
lcs: do not check if level 0 can be promoted twice
lcs: extract code for level 0 compaction from get_candidates_for
dist/offline_installer: add --skip-setup option to offline installer
dist/offline_installer/debian: install python-minimal package before installing scylla deps
migration_manager: Give empty response to schema pulls from incompatible nodes
migration_manager: Don't pull schema from incompatible nodes
...
if we are compacting level 1 into level 2, we only want to demote
a sstable from level 3 or higher.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
L0 compaction triggers quadratic behavior when many newly created
sstables are needed for promotion due to their size being relatively
low to max sstable size parameter. So until L0 is worth promoting,
the strategy will compact every new sstable with all the existing
ones in L0. To fix it, let's do STCS on level 0 until it becomes
worth promoting.
Fixes#2432.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
get rid of unneeded loop for dealing with suspect sstables and
std::advance because vector allows random access.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
some comments are no longer relevant, especially the ones that
talk about dealing with busy sstables due to parallel compaction,
which isn't done by us for lcs.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
can_promote flag will be used to carry info about whether or not
level 0 can promoted. That will avoid a single iteration for higher
levels too which can contain tens of thousands of sstables.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
I will split code for higher levels compaction into functions first
before putting it into its own function too.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
The old nodes which are still using v2 schema tables will fail to
apply our response, with error messages complaining about not being
able to locate schema of certain versions (new schema tables). This
change inhibits such errors by responding with an empty mutation list.
Currently it results in scary error messages in logs about not being
able to find schema of given version. It's benign, but may scare
users. It the future incompatibilities could result in more subtle
errors. Better to inhibit it completely.
"Old and new nodes will advertise different schema version because
of different format of schema tables. This will result in attempts
to sync the schema by each of the node.
Currently this will result in scary error messages in logs about
sync failing due to not being able to find schema of given version.
It's benign, but may scare users. It the future incompatibilities
could result in more subtle errors. Better to inhibit it completely."
* 'tgrabiec/fix-schema-pull-errors-during-upgrade' of github.com:cloudius-systems/seastar-dev:
migration_manager: Give empty response to schema pulls from incompatible nodes
migration_manager: Don't pull schema from incompatible nodes
service: Advertise schema tables format version through gossip
So that they are not left on disk even though we did a clean shutdown.
First part of the fix is to ensure that closed segments are recognized
as not allocating (_closed flag). Not doing this prevents them from
being collected by discard_unused_segments(). Second part is to
actually call discard_unused_segments() on shutdown after all segments
were shut down, so that those whose position are cleared can be
removed.
Fixes#2550.
Message-Id: <1499358825-17855-1-git-send-email-tgrabiec@scylladb.com>
The old nodes which are still using v2 schema tables will fail to
apply our response, with error messages complaining about not being
able to locate schema of certain versions (new schema tables). This
change inhibits such errors by responding with an empty mutation list.
Currently it results in scary error messages in logs about not being
able to find schema of given version. It's benign, but may scare
users. It the future incompatibilities could result in more subtle
errors. Better to inhibit it completely.
This change adds the start of what will hopefully be a continually evolving and
improving document for helping developers and contributors to get started with
Scylla development.
The first part of the document is general advice and information that is broadly
applicable.
The second part is an opinionated example of a particular work-flow and set of
tools. This is intended to serve as a starting point and inspire contributors to
develop their own work-flow.
The section on branching is marked "TODO" for now, and will be addressed by a
subsequent change.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <470a542a92aff20d6205fb94b3fb26168735ae6f.1499319310.git.jhaberku@scylladb.com>
Instead of copying and moving the bound, pass it by reference so the
transformer can decide whether it wants to copy or not. The only
caller so far doesn't want a copy and takes the value by reference,
which would be capturing a temporary value. Caught by the
view_schema_test with gcc7.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170705210255.29669-1-duarte@scylladb.com>
The state machines generated by antlr allocate many local variables per function.
In release mode, the stack space occupied by the variables is reused, but in debug
build, it is not, due to Address Sanitizer setting -fstack-reuse=none. This causes
a single function to take above 100k of stack space.
Fix by hacking the generated code to use just one variable.
Fixes#2546
Message-Id: <20170704135824.13225-1-avi@scylladb.com>
* tag 'tgrabiec/row-cache-metrics-v2' of github.com:cloudius-systems/seastar-dev:
row_cache: Switch _stats.hits/misses to row granularity
row_cache: Rename num_entries() to partitions() for clarity
row_cache: Track mispopulations also at row level
row_cache: Track row insertions
row_cache: Track row hits and misses
row_cache: Make mispopulation counter also apply for continuity information
row_cache: Add partition_ prefix to current counters
misc_services: Switch to using reads_with[_no]_misses counters
row_cache: Add metrics for operations on underlying reader
row_cache: Add reader-related metrics
row_cache: Remove dead code
"This series introduces selective_token_range_sharder and uses it in repair to
generate dht::token_range belongs to a specific shard."
* tag 'asias/repair-selective_token_range_sharder-v3' of github.com:cloudius-systems/seastar-dev:
repair: Use selective_token_range_sharder
tests: Add test_selective_token_range_sharder
dht: Add selective_token_range_sharder
With this change, we ask all the shard to handle the ranges provided by
user and we use selective_token_range_sharder to split the ranges and
ignore the ranges do not belong to the current shard.
The test put a wrapping range into a non-wrapping range variable.
This was harmless at the time this test was written, but newer code
may not be as forgiving so better use a non-wrapping range as intended.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170704103128.29689-1-nyh@scylladb.com>
`r` is moved-from, and later captured in a different lambda. The compiler may
choose to move and perform the other capture later, resulting in a use-after-free.
Fix by copying `r` instead of moving it.
Discovered by sstable_test in debug mode.
Message-Id: <20170702082546.20570-1-avi@scylladb.com>
Currently, lcs will choose, for tombstone compaction, sstable with
the lowest ratio from the ones which ratio is at least above threshold
(0.2 by default).
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170703185633.6644-1-raphaelsc@scylladb.com>
* 'lcs_improvements_part_2' of github.com:raphaelsc/scylla:
lcs: Match estimated tasks arithmetic to score in LCS
lcs: prevent leveled_compaction_strategy.hh from being included more than once
lcs: use vector instead for storing a level of sstables
compaction: keep only one variant of size_tiered_most_interesting_bucket
lcs: get rid of unused code in leveled_manifest
Contains fix for CASSANDRA-8904.
Added TARGET_SCORE to get rid of magic number for target score which
is now used more than once.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
list is no longer needed because lcs no longer moves a sstable breaking
invariant at its level to level 0. Now lcs incrementally restores invariant
by compacting together first set of overlapping tables.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
two variants of size_tiered_most_interesting_bucket existed to avoid copy,
but subsequent work will make lcs use vector for each level of sstables,
so let's only keep one variant.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Repair today has a semaphore limiting the number of ongoing checksum
comparisons running in parallel (on one shard) to 100. We needed this
number to be fairly high, because a "checksum comparison" can involve
high latency operations - namely, sending an RPC request to another node
in a remote DC and waiting for it to calculate a checksum there, and while
waiting for a response we need to proceed calculating checksums in parallel.
But as a consequence, in the current code, we can end up with as many as
100 fibers all at the same stage of reading partitions to checksum from
sstables. This requires tons of memory, to hold at least 128K of buffer
(even more with read-ahead) for each of these fibers, plus partition data
for each. But doing 100 reads in parallel is pointless - one (or very few)
should be enough.
So this patch adds another semaphore to limit the number of checksum
*calculations* (including the read and checksum calculation) on each shard
to just 2. There may still be 100 ongoing checksum *comparisons*, in
other stages of the comparisons (sending the checksum requests to other
and waiting for them to return), but only 2 will ever be in the stage of
reading from disk and checksumming them.
The limit of 2 checksum calculations (per shard) applies on the repair
slave, not just to the master: The slave may receive many checksum
requests in parallel, but will only actually work on 2 at a time.
Because the parallelism=100 now rate-limits operations which use very little
memory, in the future we can safely increase it even more, to support
situations where the disk is very fast but the link between nodes has
very high latency.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170703151329.25716-1-nyh@scylladb.com>
when do_for_each is in its last iteration and with_semaphore defers
because there's an ongoing cleanup, sstable object will be used after
freed because it was taken by ref and the container it lives in was
destroyed prematurely.
Let's fix it with a do_with, also making code nicer.
Fixes#2537.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170630035324.19881-1-raphaelsc@scylladb.com>
"compaction_strategy.cc keeps the full implementation of size tiered,
major, and null strategies, and partial implementation of leveled
and date tiered strategies. It's a mess. In the future, we will also
need space for time window strategy. The file is hard to read and
maintain.
My goal here is to improve maintainability of the strategies by
putting each of them into its own header.
NOTE: No semantic change is introduced here."
* 'improve_compaction_strategy_maintainability' of github.com:raphaelsc/scylla:
compaction_strategy: move dtcs to its existing header
compaction_strategy: move lcs implementation to its own header
compaction_strategy: move stcs implementation to its own header
compaction_strategy: move compaction_strategy_impl to its own header
Configuring cpufreq service on VMs/IaaS causes an error because it doesn't supported cpufreq.
To prevent causing error, skip whole configuration when the driver not loaded.
Fixes#2051
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1498809504-27029-1-git-send-email-syuu@scylladb.com>
A comment states that we want the file to be old enough, but sets
a timestamp of max(), which is in the future. This may have passed
because the conversion from numeric_limits<time_t>::max() to
db_clock::time_point is not well defined (their dynamic range is
different), so truncation may have converted the large number to a
low one.
Message-Id: <20170702082903.20879-1-avi@scylladb.com>
Error messages incorrectly used the debug representation of the receiver,
rather than the text representation of the operation itself.
Fixes#113.
Message-Id: <20170701101325.3163-1-avi@scylladb.com>
Boot should not continue until a future returned by
wait_for_gossip_to_settle() is resolved. Commit 991ec4a16 mistakenly
broke that, so restore it back. Also fix calls for supervisor::notify()
to be in the right places.
Message-Id: <20170702082355.GQ14563@scylladb.com>
This reverts commit db5bf363d0. Causes
errors of the sort
Exiting on unhandled exception: exceptions::invalid_request_exception
(Keyspace 'system_traces' does not exist)
Previously, lexing and parsing errors were aggregated while CQL queries were
evaluated. Afterwards, the first collected error (if present) would be thrown as
an exception.
The problem was that when parsing and lexing errors were aggregated this way,
the parser would continue even in spite of errors like "no viable alternative".
Semantic actions attached to grammar rules would still execute, though with
variables that had not yet been initialized. This would crash Scylla.
This change modifies the error-handling strategy of CQL parsing. Rather than
aggregate errors, we throw an exception on the first error we encounter. This
ensures that grammar actions never execute unless there is a precise match.
One possible issue with this approach is that the generated C++ code from the
ANTLR grammar may not be exception-safe. I compiled Scylla in debug-mode with
ASan support and executed several erroneous CQL queries with `cqlsh`. No memory
leaks were reported.
Fixes#2466.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <db1f650a2bbb615b506d9015486eece45375a440.1498836703.git.jhaberku@scylladb.com>
compaction_strategy.cc keeps the full implementation of size tiered,
major, and null strategies, and partial implementation of leveled
and date tiered strategies. It's a mess. In the future, we will also
need space for time window strategy. The file is hard to read and
maintain.
My goal here is to eventually improve maintainability of the
strategies by putting each of them into its own header.
This is the first step towards that goal.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
"This feature is intended to make compaction more efficient at getting rid of
droppable tombstone and expired data wasting disk space. So far, people have
been dealing with it manually through major compaction.
With strategies other than date tiered, large sstables will be left untouched
for a long time even though it's all expired. Date tiered suffers from it when
mixing data with different TTL because it only includes for compaction sstable
that is fully expired.
sstables keeps as metadata a histogram which allows us to easily estimate
droppable data ratio from gc_before. sstables which droppable data ratio is
above 20% (default value for tombstone_threshold option) will be considered
candidates for the operation.
Like in C*, we will only do tombstone removal compaction when there's nothing
to compact in standard way. It would be interesting to trigger it too when
disk usage is above a given threshold, but I decided to leave this for later.
Fixes #2306."
* 'tombstone_removal_compaction_v4' of github.com:raphaelsc/scylla:
tests: more testing for tombstone compaction options
tests: basic tombstone compaction test for date tiered
compaction/dtcs: add support for tombstone compaction
tests: basic test of tombstone compaction with lcs
compaction/lcs: add support for tombstone compaction
tests: basic tombstone compaction test for size tiered
compaction/stcs: add support for tombstone compaction
tests: add test for estimation of droppable tombstone ratio
sstables: introduce function to estimate droppable tombstone ratio
compaction_manager: periodically submit cfs for compaction
streaming_histogram: fix coding style
tests: add streaming_histogram_test
streaming_histogram: implement sum
tests: add test for sstable with bad tombstone histogram
sstables: discard bad streaming histogram for future use
tests: add sstable tombstone histogram test
streaming_histogram: fix update
streaming_histogram: move it to utils
streaming_histogram: do not limit it to be used by sstables
sstables: update tombstone_histogram for cells with expiration time
Unlike other strategies, dtcs has tombstone compaction disabled by
default due to:
- deletion shouldn't be used with DTCS; rather data is deleted through TTL.
- with time series workloads, it's usually better to wait for whole sstable
to be expired rather than compacting a single sstable when it's more than
20% (default value) expired.
See CASSANDRA-9234 for more details.
For tombstone compaction, unworthy sstables are filtered out and the oldest
one is chosen because it's the one less likely to shadow data and it's also
relatively big.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
LCS will choose its candidate by starting from highest level and
getting sstable which has highest droppable tombstone ratio.
Unlike STCS which needs to choose oldest sstable from biggest tier,
LCS can choose the one with highest d__t__r because sstables in
a given level don't overlap.
Sstable picked up for tombstone removal compaction won't be demoted
or promoted.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Larger sstables are hard to find sstable peers and therefore are
left uncompacted for a long time. Expired data and tombstones which
can be purged will waste disk space meanwhile.
sstable tracks droppable tombstone from which ratio can be calculated.
If ratio is greater than threshold (0.2 by default), sstable will
be eligible for compaction. Oldest sstables from biggest tiers are
preferrable because droppable data in them are more likely to satisfy
the conditions for purge, like not shadowing data in another sstable.
Subsequent patches will add support in leveled and date tiered
strategies.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Function used to estimate ratio of droppable tombstone.
A tombstone is considered droppable for cells expired before
gc_before and regular tombstones older than gc_before.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
This is useful for a column family which isn't generating new content
and will have lots of expired data later on that can be purged.
Compaction submission is NO-OP if there's nothing to do, so I think
it's reasonable to do it at an interval of 1 hour.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
This function is used to estimate number of points in interval
[-inf,b]. It will be useful for estimating droppable tombstone
ratio in a given sstable.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Find bad histogram which had incorrect elements merged due to use of
unordered map. The keys will be unordered. Histogram which size is
less than max allowed will be correct because no entries needed to be
merged, so we can avoid discarding those.
This is important because histogram for tombstone will be used to
estimate droppable tombstone ratio. If it's incorrectly high for many
of existing sstables, we will needlessly compact lots of them.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
This bug was introduced when converting java code. Return value
of map::erase() was used as if it were the value of the removed
entry, but it's actually the number of removed entries.
update() also relies on ordered keys, so map is used instead
by histogram.
In addition, histograms will be written in sorted order (like C*
does) such that we can detect bad histograms, using disk_array.
disk_array is also used from now on to read histograms.
The conversion from array to map is fine because histograms for
sstables are limited to 100 elements.
Coming patch will detect bad histograms (generated only by us)
and discard them, because we can't rely on their information.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
As a preperation for the http stream support, creation of empty reply
should be avoided.
This patch removes a line that cannot be reached but causes the compiler
to complain.
It has no effect aside of removing the reply creation.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170628130202.8132-1-amnon@scylladb.com>
If cache is missing given key, but the range is marked as continuous,
it means sstables don't have that entry and we can insert it without
asking the presence checker (bloom filter based). The latter is more
expensive and gives false positives. So this improves update
performance and hit ratio.
Another positive effect is that we don't have to clear continuity now.
Fixes#1999.
Message-Id: <1498643043-21117-1-git-send-email-tgrabiec@scylladb.com>
Region comparator, used by the two, calls region_impl::min_occupancy(),
which calls log_histogram::largest(). The latter is O(N) in terms of
the number of segments, and is supposed to be used only in tests.
We should call one_of_largest() instead, which is O(1).
This caused compact_on_idle() to take more CPU as the number of
segments grew (even when there was nothing to compact). Eviction
would see the same kind of slow down as well.
Introduced in 11b5076b3c.
Message-Id: <1498641973-20054-1-git-send-email-tgrabiec@scylladb.com>
It's been linked with various performance issues, either by causing
them or making them worse. One example is #1634, and also recently
I have investigated continuous performance degradation that was also
linked to defrag on idle activity.
Until we can figure out how to reduce its impact, we should disable it.
Signed-off-by: Glauber Costa <glauber@glauber.scylladb>
Message-Id: <20170627201109.10775-1-glauber@scylladb.com>
streaming histogram will later be placed in /utils, so we want
it to use std::unordered_map<> instead of disk_hash<>.
That also requires implementing serialization/deserialization
functions for it.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
That tombstone_histogram is used to determine droppable data ratio
for a sstable, and unlike C*, we were only updating it for
tombstones. We need to update it with expiration time of cells too,
if any. Creation time (expiration - ttl) cannot be used because if
ttl > gc_grace_seconds, the resulting sstable could be considered
worth dropping by tomstone compaction before any data is actually
expired.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Currently our build script only supports Ubuntu 14.04/16.04 and Debian 8, this
change extends support to Ubuntu non-LTS versions / Debian non-stable versions.
Note that this is unofficial support, users should build the package for these
distributions theirselves.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1498491473-28691-1-git-send-email-syuu@scylladb.com>
When peer nodes have the same partition data, i.e., with the same
checksum, we currently choose to stream from any of them randomly.
To improve streaming performance, select the peer within the same DC.
This patch is supposed to improve repair perforamnce with multiple DC.
Message-Id: <c6a345b6e8ed2b59f485e53c865241e463b44507.1498490831.git.asias@scylladb.com>
Fixes#2528.
* tag 'asias/gossip_talk_to_more_nodes/v3' of github.com:cloudius-systems/seastar-dev:
gossip: Use vector for _live_endpoints
gossip: Talk to more live nodes in each gossip round
In large clusters with multiple DC deployment, it is observed that it
takes long delay for gossip update to disseminate in the cluster.
To speed up, talk to more live nodes in each gossip round.
Fixes#2528
This patch does the same thing to column_family::make_sstable_reader() as
commit 186f031 did to sstable::as_mutation_source().
Although usually one can fast_forward_to() on the result of a
column_family::make_sstable_reader(), earlier we had an optimization
where if a single partition was specified, it was read exactly,
and fast_forward_to() was *NOT* allowed.
With the mutation_reader::forwarding flag patch, when this flag
was on - requesting fast_forward_to() - we disabled this optimization.
This makes sense, but is not backward compatible with the code which
previously assumes this optimization exists. In particular,
column_family::data_query() does a single partition read but does not
specify forwarding::no explicitly.
So this patch returns this optimization, despite this meaning that we
blatently ignore the fwd_mr flag in that case.
Fixes#2524.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170626141121.30322-1-nyh@scylladb.com>
"This series enables cache to keep partial partitions.
Reads no longer have to read whole partition from sstables
in order to cache the result.
The 10MB threshold for partition size in cache is lifted.
Known issues:
- There is no partial eviction yet, whole partitions are still evicted,
and partition snapshots held by active reads are not evictable at all
- Information about range continuity is not recorded if that
would require inserting a dummy entry, or if previous entry
doesn't belong to the latest snapshot
- Cache update after memtable flush happening concurrently with reads
may inhibit that reads' ability to populate cache (new issue)
- Cache update from flushed memtables has partition granularity,
so may cause latency problems with large partition
- Schema is still tracked per-partition, so after schema changes
reads may induce high latency due to whole partition needing
to be converted atomically
- Range tombstones are repeated in the stream for every range between
cache entries they cover (new issue)
- Populating scans for both small and large partitions (perf_fast_forward)
experienced a 40% reduction of throughput, CPU bound
How was this tested:
- test.py --mode release
- row_cache_stress_test -c1 -m1G
- perf_fast_forward, passes except for the test case checking range continuity population
which would require inserting a dummy entry (mentioned above)
- perf_simple_query (-c1 -m1G --duration 32):
before: 90k [ops/s] stdev: 4k [ops/s]
after: 94k [ops/s] stdev: 2k [ops/s]"
* tag 'tgrabiec/introduce-partial-cache-v8' of github.com:cloudius-systems/seastar-dev: (130 commits)
tests: row_cache: Add test_tombstone_merging_in_partial_partition test case
tests: Introduce row_cache_stress_test
utils: Add helpers for dealing with nonwrapping_range<int>
tests: simple_schema: Allow passing the tombstone to make_range_tombstone()
tests: simple_schema: Accept value by reference
tests: simple_schema: Make add_row() accept optional timestamp
tests: simple_schema: Make new_timestamp() public
tests: simple_schema: Introduce make_ckeys()
tests: simple_schema: Introduce get_value(const clustered_row&) helper
tests: simple_schema: Fix comment
tests: simple_schema: Add missing include
row_cache: Introduce evict()
tests: Add cache_streamed_mutation_test
tests: mutation_assertions: Allow expecting fragments
mutation_fragment: Implement equality check
tests: row_cache: Add test for population of random partitions
tests: row_cache: Add test for partition tombstone population
tests: row_cache: Test reading randomly populated partition
tests: row_cache: Add test_single_partition_update()
tests: row_cache: Add test_scan_with_partial_partitions
...
Runs readers, updates and eviction concurrently and verifies the
following property of reads:
- reads see all past writes
- reads see no partial writes within a single partition
[tgrabiec:
- extracted from a larger commit
- removed coupling with how cache_streamed_mutation is created (the
code went out of sync), used more stable make_reader(). it's simpler too.
- replaced false/true literals with is_continuous/is_dummy where appropraite
- dropped tests for cache::underlying (class is gone)
- reused streamed_mutation_assertions, it has better error messages
- fixed the tests to not create tombstones with missing timestamps
- relaxed range tombstone assertions to only check information relevant for the query range
- print cache on failure for improved debuggability
]
Those methods first create a neutral mutation_partition, and left-fold
it with the versions. The problem is that there is no neutral element
for static row continuity, the flag from the first addend always
wins. We have to copy the flag from the first version to preserve
the logical value.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
[tgrabiec:
- extracted from a larger commit
- fix heap comparator in apply_incomplete_target to order versions properly
- extracted partition_version detaching into
partition_entry::with_detached_versions()
- dropped unnecessary rows_iterator::_version field
- dropped unnecessary allocation of rows_entry and key copies
in rows_iterator
- dropped row_pointer
- replaced apply_reversibly() with weaker and faster apply()
- added handling of dummy entries at any position
- fixed exception safety issue in apply_to_incomplete() which may
result in data loss. We cannot move data out of applied versions
into a new synthetic row and then apply it, because if exception
happens in the middle, the data which was moved from the source
will be lost. To fix that, row_iterator::consume_row() is
introduced which allows in-place consumption of data without
construction of temporary deletable_row.
]
This streamed mutation populates cache with
the rows requested by the read. It takes whatever
it can find in the cache and fetches the remainings
from underlying source.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
[tgrabiec:
- fixed maybe_add_to_cache_and_update_continuity() leaking entries if
the key already exists in the snapshot
- fixed a problem where population race could result in a read
missing some rows, because cache_streamed_mutation was advancing
the cursor, then deferring, and then checking continuity. We
should check continuity atomically with advancing.
- fixed rows_handle.maybe_refresh() being accessed outside of update
section in read_from_underlying() (undefined behavior)
- fixed a problem in start_reading_from_underlying() where we would
use incorrect start if lower_bound ended with a range tombstone
starting before a key.
- range tombstone trimming in add_to_buffer() could create a
tombstone which has too low start bound if last_rt.end was a
prefix and had inclusive end. invert_kind(end_kind) should be used
instead of unconditional inc_start.
- range tombstone trimming incorrectly assumed it is fine to trim
the tombstone from underlying to the previous fragment's end and
emit such tombstone. That would mean the stream can't emit any
fragments which start before previous tombstone's end. Solve with
range_tombstone_stream.
- split add_to_buffer() into overloads for clustering_row, and
range_tombstone. Better than wrapping into mutation_fragment
before the call and having add_to_buffer() rediscover the
information.
- changed maybe_add_to_cache_and_update_continuity() to not set
continuity to false for existing entries, it's not necessary
- moved range tombstone trimming to range_tombstone class
- moved range tombstone slicing code to range_tombstone_list and partition_snapshot
- can_populate::can_use_cache was unused, dropped
- dropped assumption that dummy entries are only at the end
- renamed maybe_add_to_cache_and_update_continuity() to maybe_add_to_cache()
- dropped no longer needed lower_bound class
- extracted row_handle to a seaparate patch
- made the copy-from-cache loop preemptable
- split maybe_add_next_to_buffer_and_update_continuity(bool)
- dropped cache_populator
- replaced "underlying" class with use of read_context
- replaced can_populate class with a function
- simplified lsa_manager methods to avoid moves
]
The interaction will be as follows:
- Before creating cache_streamed_mutation for given partition, cache
mutation reader sets up read_context for current partition (in one
of two ways) so that the matching underlying streamed_mutation can
be accessed at any time by cached_stream_mutation.
- cache_streamed_mutation assumes that read_context is set up for
current partition and invokes fast_forward_to() and
get_next_fragment() to access the underlying
streamed_mutation.
When reading from incomplete partition entry, we may discover we need
to read something from the underlying mutation source. In such case we
will fast forward this reader to that partition. But we must do it
using a specific snapshot, the one we obtained when entering the
partition, not the latest one.
We will need to use this information later in yet another place, when
creating a reader for incomplete cache entry. This refactors the code
so that there is a single place which determines this fact.
Currently scanning_and_populating_reader asks
just_cache_scanning_reader for the next partition from cache, together
with information if the range is continuous. If it's not, it saves the
partition it got from it and moves on to reading from the underlying
reader up to that partition. When that's done, it emits the stored
partition.
This approach won't work well with upcoming changes for storing
partial partitions. We won't have whole partitions any more, so
streamed_mutation returned for the entry needs to be prepared for
reading from the underlying mutation source. We want to reuse the same
underlying reader as much as possible, so all streamed_mutations for
given read (read_context) will share the state of the underlying
reader. Construction of a streamed_mutation will depend on the fact
that the shared state is set up for it, so we cannot have two
streamed_mutations prepared at the same time (one for entry from
primary, and one for the earlier entry being populated). This change
defers the creation of a streamed_mutation for the entry present in
cache until the whole reader reaches it to avoid this problem.
This will also have antoher potentially beneficial effect. Since we
defer the decision about which snapshot to use until we reach the
entry, there is a higher chance that the current snapshot of the entry
will match the one used last by the populating read, and that we will
be able to reuse the reader.
It's implemented by utilizing a stable partition cursor which tracks
its current position so that it's possible to revisit the cache entry
(if it's still there) after population ends. The functionality of
just_cache_scanning_reader was inlined into
scanning_and_populating_reader.
dht::token needs to be stored as a pointer now and not a reference so
that validity of old pointers is not impacted by in-place object
construction which would occur in the copy-assignment operator.
[1] says that old pointers can be used to access the new object only
if the type "does not contain any non-static data member whose type is
const-qualified or a reference type".
[1] http://en.cppreference.com/w/cpp/language/lifetime#Storage_reuse
It's needed before switching cache_entry ordering to rely solely on
cache_entry::position() so that invalidate_unwrapped() never removes
the dummy entry at the end. Currently if the range has upper bound
like this:
{ ring_position::max(), inclusive=true }
The code which selects entries for removal would include the dummy row
at the end. It uses upper_bound() to get the end iterator, and the
dummy entry has a position which is equal to the position in the
bound.
ring_position_view ranges are end-exclusive, so it's impossible to
create a partition range which would include a dummy entry.
The code is also simpler.
This will allow conversion from streamed_mutation that
supports fast forwarding to streamed_mutation that does not.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Currently readers are always using the latest snapshot. This is fine
for respecting write atomicity if partitions are fully continuous in
cache (now), but will break write atomicity once partial population is
allowed.
Consider the following case:
flush write(ck=1), write(ck=2) -> snapshot_1
cache reader 1 reads and inserts ck=1 @snapshot_1
flush write(ck=1), write(ck=2) -> snapshot_2
cache reader 2 reads and inserts ck=2 @snapshot_2
Because cache update is not atomic, it can happen that reader 2 will
complete while the partition hasn't been updated yet for snapshot_2.
In such case, after read 2 the partition would contain ck=1 from
snapshot_1 and ck=2 from snapshot_2. It will match neither of the
snapshots, and this could violate write atomicity.
To solve this problem we conceptually assign each partition key in the
ring to its current snapshot which it reflects. The update process
gradually converts entries in ring order to the new snapshot. Reads
will not be using the latest snapshot, but rather the current snapshot
for the position in the ring they are at.
There is a race between the update process and populating reads. Since
after the update all entries must reflect the new snapshot, reads
using the old snapshot cannot be allowed to insert data which can no
longer be reached by the update process. Before this patch this race
was prevented by the use of a phased_barrier, where readers would keep
phased_barrier::operation alive between starting a read of a partition
and inserting it into cache. Cache update was waiting for all prior
operations before starting the update. Any later read which was not
waited for would use the latest snapshot for reads, so the update
process didn't have to fix anything up for such reads.
After this change, later reads cannot always use the latest snapshot,
they have to use the snapshot corresponding to given entry. So it's
not enough for update() to wait for prior reads in order to prevent
stale populations. The (simple) solution implemented in this patch is
to detect the conflict and abandon population of given sub-range. In
general, reads are allowed to populate given range only if it belongs
to a single snapshot.
Note that the range here is not the whole query range. For population
of continuity, it is the range starting after the previous key and
ending after the key being inserted. When populating a partition
entry, the range is a singular range containing only the partition
key. Readers switch to new snapshots automatically as they move across
the ring. It's possible that the insertion of the partition doesn't
conflict, but continuity does. In such case the entry will be inserted
but continuity will not be set.
Currently every time cache needs to create reader for missing data it
obtains a reader which is most up to date. That reader includes writes
from later populate phases, for which update() was not yet
called. This will be problematic once we allow partitions to be
partially populated, because different parts of the partition could be
partially populated using readers using different sets of writes, and break
write atomicity.
The solution will be to always populate given partition using the same
set of writes, using reader created from the current snapshot. The
snapshot changes only on update(), with update() gradually converting
each partition to the new snapshot.
This violation of the contract is currently benign, because there are
no reads from those tables before they are populated. If there were,
the cache would mark the whole (empty) range as continuous and the
table would appear empty.
It will cause similar problem once cache starts using snapshots of the
underlying mutation source. Then this lack of invalidate() will also
result in cache thinking that the table is still empty.
1) Reduce duplication by delegating to more general overloads
2) Improve documentation to not mention effects in terms of
population (detail) but rather write visibiliy
3) Rename clear() to invalidate() and merge with the range variant,
it has the same semantics
This object stores all read relevant context required all
over the place. This leads to a cleaner code.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
[tgrabiec:
- made read_context shareable to allow storing shared
mutable state later
- added range and cache getters
]
This is an abstraction that represents a reader
to the underlying source and auto updates itself
to make sure the reader reflects the latest state
of the underlying source.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
[tgrabiec: Add range getter to avoid friendships]
[tgrabiec:
- Extracted from a different patch
- Renamed concept names to more familiar Map and Reduce
- Renamed aggregate() to squashed() to match the existing nomenclature
- Uncommented the concepts
]
Currently mutation sources are free to return range tombstones
covering range which is larger than the query range. The cache
mutation source will soon become more eager about trimming such
tombstones. To cover up for such differences, allow telling the
restrictions to only care about differences relevant for given
clustering ranges.
This will be used by partial cache in later patches.
[tgrabiec:
- changed title,
- documented meaning of the variable,
- renamed the variable,
- introduced open_version(),
- fixed continuity of the static row not being preserved in case
a new version is created]
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
mutation source sometimes ignore fast forwarding parameter so
this change adds assertion to check that this parameter
can be safely ignored.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
By default make_reader_returning creates a reader that does not
support fast forwarding but the second parameter can be used to
make it support fast forwarding.
[tgrabiec: Improve title]
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This will allow expressing lack of information about certain ranges of
rows (including the static row), which will be used in cache to
determine if information in cache is complete or not.
Continuity is represented internally using flags on row entries. The
key range between two consecutive entries is continuous iff
rows_entry::continuous() is true for the later entry. The range
starting after the last entry is assumed to be continuous. The range
corresponding to the key of the entry is continuous iff
rows_entry::dummy() is false.
[tgrabiec:
- based on the following commits:
4a5bf75 - Piotr Jastrzebski : mutation_partition: introduce dummy rows_entry
773070e - Piotr Jastrzebski : mutation_partition: add continuity flag to rows_entry
- documented that partition tombstone is always complete
- require specifying the partition tombstone when creating an incomplete entry
- replaced rows_entry(dummy_tag, ...) constructor with more general
rows_entry(position_in_partition, ...)
- documented continuity semantics on mutation_partition
- fixed _static_row_cached being lost by mutation_partition copy constructors
- fixed conversion to streamed_mutation to ignore dummy entries
- fixed mutation_partition serializer to drop dummy entries
- documented semantics of continuity on mutation_partition level
- dropped assumptions that dummy entries can be only at the last position
- changed equality to ignore continuity completely, rather than
partially (it was not ignoring dummy entries, but ignoring
continuity flag)
- added printout of continuity information in mutation_partition
- fixed handling of empty entries in apply_reversibly() with regards
to continuity; we no longer can remove empty entries before
merging, since that may affect continuity of the right-hand
mutation. Added _erased flag.
- fixed mutation_partition::clustered_row() with dummy==true to not ignore the key
- fixed partition_builder to not ignore continuity
- renamed dummy_tag_t to dummy_tag. _t suffix is reserved.
- standardized all APIs on is_dummy and is_continuous bool_class:es
- replaced add_dummy_entry() with ensure_last_dummy() with safer semantics
- dropped unused remove_dummy_entry()
- simplified and inlined cache_entry::add_dummy_entry()
- fixed mutation_partition(incomplete_tag) constructor to mark all row ranges as discontinuous
]
If some row entries may have to be skipped by the reader then it could
be that _clustering_rows is not empty, but read_next() will return a
disengaged optional because there are no more rows in the current
range. The code assumed that it's never the case, and if read_next()
returns a disengaged optional then we exhousted all ranges. Before
introducing dummy entries this needs to be refactored.
key() will not be valid for dummy entries, but position() is always
valid.
[tgrabiec: Extracted from other commits]
[tgrabiec: Added missing change to range_tombstone_stream::get_next]
This is a position that's always in the end after any
other position. It will be used for dummy rows_entry.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
It can happen that touch() will trigger eviction on entry to
allocating section, and drop in occupancy around insertion
will not happen. As a result, we may evict a lot without detecting that.
Extend the check to include touch() and use more reliable eviction counters.
We support situations where `seastar.pc` is available, and when it isn't.
When `seastar.pc` is available, we grab the compilation flags from it in
addition to the defaults.
Some DPDK files are statically available in the source repository. We prefer
those to files placed during compilation in case modifications are made during
development (that would be lost during a build).
We always disable GCC6 concepts for the IDE, even if they are enabled while
configuring the project for compilation. CLion's parser doesn't understand them.
One final benefit to this revision is that now only target-specific flags are
modified rather than global flags.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <75457652201c5ed89d05081ec4b56b4340721cf5.1498237756.git.jhaberku@scylladb.com>
"This patch series makes two noteworthy changes:
- `gc_clock` now uses `seastar::lowres_system_clock` for better performance.
- Code common to all clocks is properly refactored and shared.
Otherwise, there are some small improvements that should have no functional impact."
* 'jhk/lowres_gc_clock/v1' of https://github.com/hakuch/scylla:
Seal clock definitions
`timestamp_clock::now()` is not `noexcept`
Make `db_clock` `time_t` conversions `constexpr`
Move common clock implementation helpers
Simplify clock implementations
db_clock.hh: Clean preprocessor directives
Make `gc_clock` a model of `Clock`
Use `lowres_system_clock` to back `gc_clock`
Add `time_t` conversions for `gc_clock`
The problem is that `std::chrono::duration_cast` is not `noexcept`. As a result,
`timestamp_clock` is actually a model of `Clock` and not `TrivialClock`.
This change fixes the dependencies between the clock implementation headers. All
the clocks share the common clock offset, but are otherwise independent (though
the `db_clock` does depend on `gc_clock` for time point conversions).
`gc_clock` reports system time, and these conversion functions allow for
manipulating time points produced by the clock without making assumptions about
its epoch.
"Fix minor issues found while building with clang trunk."
* 'clang' of https://github.com/avikivity/scylla:
seastarx: don't make seastar namespace inline
seastarx: add missing make_shared forward declaration
tests: fix call to seastar::sleep()
dht: fix bad to_sstring() call
* 'lcs_improvements_v1' of github.com:raphaelsc/scylla:
lcs: remove useless code for choosing L0 candidates
lcs: remove some dead code
lcs: make logger static
lcs: actually prefer oldest sstables of L0 when it falls behind
lcs: remove useless expensive check for overlapping L1 sstables
so now user can look at nodetool compactionstats and determine
whether or not resharding is running, for example:
$ ./bin/nodetool compactionstats
pending tasks: 3
id compaction type keyspace table completed total unit progress
<none> RESHARD system compaction_history 11 256 keys 4.30%
<none> RESHARD system compaction_history 2 256 keys 0.78%
<none> RESHARD system compaction_history 10 256 keys 3.91%
<none> RESHARD system compaction_history 8 256 keys 3.12%
<none> RESHARD system compaction_history 7 256 keys 2.73%
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170620175733.25882-1-raphaelsc@scylladb.com>
The code filters CFs by name to not include system keyspace, but v3
schema added yet another system namespace. Better filter according to
replication strategy to accommodate for schema v4 adding even more
system keyspaces.
Fixes: #2516
Message-Id: <20170621073816.GB3944@scylladb.com>
This version optionally reads the include paths for Seastar from pkg-config and
uses file globbing to register all source and header files.
In comparison to the previous version, I see all files in the project explorer
view are "active" (rather than just .cc files). I believe there are also fewer
errors reported by the editor.
Add support Debian new stable release.
Also including following changes:
- update libthrift due to unable to compile on Debian 9
- drop dist/debian/supported_release since distribution check code moved to pbuilderrc
- add libssl-dev for build-depends
- add sudo for pbuilder extra packages (Debian doesn't have it by default install)
Signed-off-by: syuu <syuu@dokukino.com>
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1498047515-19972-1-git-send-email-syuu@scylladb.com>
The memtable destructor can take a long time if the memtable is full; use
clear_gently() to clear it without impacting latency.
Fixes#2477.
Message-Id: <20170620093550.16121-1-avi@scylladb.com>
Although usually one can fast_forward_to() on the result of a
sstable::as_mutation_source(), earlier we had an optimization
where if a single partition was specified, it was read exactly,
and fast_forward_to() was *NOT* allowed.
With the mutation_reader::forwarding flag patch, when this flag
was on - requesting fast_forward_to() - we disabled this optimization.
This makes sense, but is not backward compatible with the code which
previously assumes this optimization exists... So this patch returns
this optimization, despite this meaning that we blatently ignore
the fwd_mr flag in that case.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170620081107.14335-1-nyh@scylladb.com>
* seastar 621b7ed...9e2b7ec (8):
> Merge "Low-resolution clocks" from Jesse
> build: disable -Wattributes when gcc -fvisibility=hidden bug strikes
> build: work around ragel 7 generated code bug
> rpc: make unmarshall_exception() inline
> future-utils: make functions global
> fix reactor stall detector rate limiting on an mostly idle system
> prometheus: Add ability to add /metrics to any http_server
> prometheus: fix memory leak in http_server_control
multi_range_mutation_reader uses fast_forward_to() to skip between
ranges, so we always need to create the underlying reader with with
mutation_reader::forwarding::yes if there is more than one range,
irrespective of whether multi_range_mutation_reader itself will be
forwarded or not.
Fixes#2510.
Introduced in commit 3018df1.
Message-Id: <1497943032-18696-1-git-send-email-tgrabiec@scylladb.com>
The code being removed could be used if parallel compaction were
allowed for LCS, but the current code isn't even allowing that.
At the moment, it's only wasting cycles.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Strategy prefers promoting oldest sstables in L0. Because sort
procedure is incorrectly sorting elements in descending order,
newest sstables will be promoted first *if and only if* L0 falls
behind (more than 32 sstables). If L0 doesn't fall behind, we'll
have all L0 sstables compacted with overlapping ones in L1.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
there's no way a L1 sstable will be in candidates set which was
previously built from list of L0 sstables.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Enable pbuilder for Ubuntu/Debian to prevent build enviroment dependent issues.
Also support cross building by pbuilder.
(cross-building from Fedora 25 and Ubuntu 16.04 are tested)
closes#629
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1497895661-26376-1-git-send-email-syuu@scylladb.com>
As Avi noticed, the "forwarding_tag" which was meant to be local in
streamed_mutation, became global. If another class copied the same trick,
it would share the same type instead of being distinct types as intended.
The problem is that in:
using forwarding = bool_class<class forwarding_tag>;
Apparently, the "class forwarding_tag" forward-declares a global type - it
does not create a local-scope type as intended, which the following apparently
does (even though no actual definition is given for that class):
class forwarding_tag;
using forwarding = bool_class<forwarding_tag>;
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170619153933.13116-1-nyh@scylladb.com>
In commit c63e88d556, support was added for
fast_forward_to() in data_consume_rows(). Because an input stream's end
cannot be changed after creation, that patch ignores the specified end
byte, and uses the end of file as the end position of the stream.
As result of this, even when we want to read a specific byte range (e.g.,
in the repair code to checksum the partitions in a given range), the code
reads an entire 128K buffer around the end byte, or significantly more, with
read-ahead enabled. This causes repair to do more than 10 times the amount
of I/O it really has to do in the checksumming phase (which in the current
implementation, reads small ranges of partitions at a time).
This patch has two levels:
1. In the lower level, sstable::data_consume_rows(), which reads all
partitions in a given disk byte range, now gets another byte position,
"last_end". That can be the range's end, the end of the file, or anything
in between the two. It opens the disk stream until last_end, which means
1. we will never read-ahead beyond last_end, and 2. fast_fordward_to() is
not allowed beyond last_end.
2. In the upper level, we add to the various layers of sstable readers,
mutation readers, etc., a boolean flag mutation_reader::forwarding, which
says whether fast_forward_to() is allowed on the stream of mutations to
move the stream to a different partition range.
Note that this flag is separate from the existing boolean flag
streamed_mutation::fowarding - that one talks about skipping inside a
single partition, while the flag we are adding is about switching the
partition range being read. Most of the functions that previously
accepted streamed_mutation::forwarding now accept *also* the option
mutation_reader::forwarding. The exception are functions which are known
to read only a single partition, and not support fast_forward_to() a
different partition range.
We note that if mutation_reader::forwarding::no is requested, and
fast_forward_to() is forbidden, there is no point in reading anything
beyond the range's end, so data_consume_rows() is called with last_end as
the range's end. But if forwarding::yes is requested, we use the end of the
file as last_end, exactly like the code before this patch did.
Importantly, we note that the repair's partition reading code,
column_family::make_streaming_reader, uses mutation_reader::forwarding::no,
while the other existing reading code will use the default forwarding::yes.
In the future, we can further optimize the amount of bytes read from disk
by replacing forwarding::yes by an actual last partition that may ever be
read, and use its byte position as the last_end passed to data_consume_rows.
But we don't do this yet, and it's not a regression from the existing code,
which also opened the file input stream until the end of the file, and not
until the end of the range query. Moreover, such an improvement will not
improve of anything if the overall range is always very large, in which
case not over-reading at its end will not improve performance.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170619152629.11703-1-nyh@scylladb.com>
We mistakenly placed ./configure.py to dh_auto_build, but it's should place at
dh_auto_configure.
This bug causes issue #2505, since we haven't defined dh_auto_configure task yet(It seems running cmake on top of the dir is one of default behavior of dh_auto_configure).
Fixes#2505
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1497866068-32097-1-git-send-email-syuu@scylladb.com>
This patch add the CMakeLists.txt file for IDEs based on cmake, like
CLion.
This file assumes the existence of a build/release/gen directory,
containing generated files.
Refs #867
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170618151333.94714-1-duarte@scylladb.com>
"This patch set ensures we quote the name of a UDT when it
contains characters that may cause parsing by the CQL parser
to fail.
Fixes#2491"
* 'cql3-quote-type/v1' of https://github.com/duarten/scylla:
cql3/util: Make maybe_quote() take argument by const reference
cql3/cql3_type: Quote UDT name if needed
schema: Lift maybe_quote() into cql3/util
Boost 1.55 (ubuntu 14) fails to compile because an iterator produce by
boost::adaptors::transformed() when std::ref to lambda is passed to
it do not match iterator concept. It cannot be default constructed
because std::reference_wrapper is not default constructable.
boost::range::min_element() never actually default construct it, but
concept is checked anyway. The patch fixes it by providing an explicit
functor that is default constructable.
Message-Id: <20170618131836.GD3944@scylladb.com>
When making the schema mutations for a view update, we should only
include the base table schema mutations (in case the target node
doesn't contain them) when the view is being directly updated. When it
is being updated as a side effect of updating the base table, then
including the base schema mutations will hide the actual changes being
performed on the base.
Fixes#2500
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1497782822-2711-1-git-send-email-duarte@scylladb.com>
This reverts commit 317d7fc253 (and also the
related 2c57ab84b2). It causes crashes
during range scans, reported by Gleb:
"To reproduce I run SELECT * FROM keyspace1.standard1; on typical c-s
dataset and 3 node cluster.
Backtrace:
at /home/gleb/work/seastar/seastar/core/apply.hh:36
rvalue=<unknown type in /home/gleb/work/seastar/build/release/scylla, CU 0x54cf307, DIE 0x55ebf2a>) at /home/gleb/work/seastar/seastar/core/do_with.hh:57
range=std::vector of length 6, capacity 8 = {...}) at /home/gleb/work/seastar/seastar/core/future-util.hh:142
at ./seastar/core/future.hh:890
at /home/gleb/work/seastar/seastar/core/future-util.hh:119
at /home/gleb/work/seastar/seastar/core/future-util.hh:142
Add a performance test for deletion in addition to the existing update
and query tests. The deletion performance test is executed using the
'--delete' argument to perf_simple_query.
Fixes#2417.
Signed-off-by: Etienne Kruger <el@loadavg.io>
Message-Id: <20170615232500.26987-1-el@loadavg.io>
This patch ensures we properly quote a UDT name, which may contain
characters like ".", which can lead the name to be interpreted as a
keyspace qualified name when parsed by the CQL parser.
Fixes#2491
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The typo went unnoticed since the compiler picked up the global scope's
forwarding_tag. The bug made streamed_mutation::forwarding and
mutation_reader::forwarding the same type, but fortunately there were
no type mixups due to this.
In commit c63e88d556, support was added for
fast_forward_to() in data_consume_rows(). Because an input stream's end
cannot be changed after creation, that patch ignores the specified end
byte, and uses the end of file as the end position of the stream.
As result of this, even when we want to read a specific byte range (e.g.,
in the repair code to checksum the partitions in a given range), the code
reads an entire 128K buffer around the end byte, or significantly more, with
read-ahead enabled. This causes repair to do more than 10 times the amount
of I/O it really has to do in the checksumming phase (which in the current
implementation, reads small ranges of partitions at a time).
This patch has two levels:
1. In the lower level, sstable::data_consume_rows(), which reads all
partitions in a given disk byte range, now gets another byte position,
"last_end". That can be the range's end, the end of the file, or anything
in between the two. It opens the disk stream until last_end, which means
1. we will never read-ahead beyond last_end, and 2. fast_fordward_to() is
not allowed beyond last_end.
2. In the upper level, we add to the various layers of sstable readers,
mutation readers, etc., a boolean flag mutation_reader::forwarding, which
says whether fast_forward_to() is allowed on the stream of mutations to
move the stream to a different partition range.
Note that this flag is separate from the existing boolean flag
streamed_mutation::fowarding - that one talks about skipping inside a
single partition, while the flag we are adding is about switching the
partition range being read. Most of the functions that previously
accepted streamed_mutation::forwarding now accept *also* the option
mutation_reader::forwarding. The exception are functions which are known
to read only a single partition, and not support fast_forward_to() a
different partition range.
We note that if mutation_reader::forwarding::no is requested, and
fast_forward_to() is forbidden, there is no point in reading anything
beyond the range's end, so data_consume_rows() is called with last_end as
the range's end. But if forwarding::yes is requested, we use the end of the
file as last_end, exactly like the code before this patch did.
Importantly, we note that the repair's partition reading code,
column_family::make_streaming_reader, uses mutation_reader::forwarding::no,
while the other existing reading code will use the default forwarding::yes.
In the future, we can further optimize the amount of bytes read from disk
by replacing forwarding::yes by an actual last partition that may ever be
read, and use its byte position as the last_end passed to data_consume_rows.
But we don't do this yet, and it's not a regression from the existing code,
which also opened the file input stream until the end of the file, and not
until the end of the range query. Moreover, such an improvement will not
improve of anything if the overall range is always very large, in which
case not over-reading at its end will not improve performance.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170614072122.13473-1-nyh@scylladb.com>
"During read query with CL<ALL not all replicas are contacted. It is
possible for some replicas to cache less data for some CF's (for instance
because of node restart), so the replica choice may have a big impact
on request's completion latency and on amount of work it generates in
a cluster.
This patch series keep track of per CF cached hit ratio and uses this
information to choose best replicas for a request. Nodes with lower
hit ratios are still contacted in order to populate their cache, but
less frequently."
* 'gleb/cache-hitrate' of github.com:cloudius-systems/seastar-dev:
storage_proxy: load balance read requests according to cache hit rates
choose extra replica for speculation in filter_for_query()
consistency_level: drop filter_for_query_dc_local function
database: reset node's hit rate information on connection drop
messaging_service: connection drop notifier
Store cluster wide cache hit statistics in CF
messaging_service: return cache hit ratio as part of data read
Distribute cache temperature over gossiper.
periodically calculate avg cache hit rate between all shards
database: introduce cache_temperature class
Rename load_broadcaster.cc to misc_services.cc
storage_proxy: use db::count_local_endpoints function instead open code it
"This series reduces repair memory usage and improves repair speed."
* tag 'asias/fix-repair-2430-branch-master-v4.1' of github.com:cloudius-systems/seastar-dev:
repair: Repair on all shards
repair: Allow one stream plan in flight
Range tombstones were introduced in version 1.3 and there exists
no direct upgrade from 1.2 to vnext, so we can retire the code
enforcing backwards compatibility.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170614211654.82501-1-duarte@scylladb.com>
Currently commitlog_entry_writer constructor calculates serialized size
before it is knows if a schema should be included into the entry. The
result is never used since it is recalculated when schema information is
supplied. The patch removes needless calculation.
Message-Id: <20170614114607.GA21915@scylladb.com>
Currently each time UDT or tuple is parsed new object is created. If
those objects are used to create container type repeatedly it will cause
memory leak since container types are interned, but lookup in the
cache is done using pointer to a contained type (which will be always
different for UDT and tuples). This patches interns also UDT and tuple,
so each type the same object is parsed same pointer is also returned.
Refs #2469Fixes#2487
Message-Id: <20170612142942.GO21915@scylladb.com>
Currently, shard zero is the coordinator of the repair. All the work of
checksuming of the local node and sending of the repair checksum rpc
verb is done on shard zero only. This causes other shards being
underutilized.
With this patch, we split the ranges need to be repaired into at least
smp::count ranges, so sizeof(ranges) / smp::count will be assigned to
each shard. For exmaple, we have 8 shards and 256 ragnes, each shard
will repair 32 ranges. Each shard will repair the 32 ranges
sequencially. There will be at most 8 (smp::count) ranges of repair in
parallel.
In "repair: Use more stream_plan" (commit 2043ffc064), we
switched to do stream while doing checksum instead of do stream only
after checksum pahse is completed. We take a parallelism_semaphore
before we do checksum, if there are more than sub_ranges_to_stream
(1024) ranges, we start a stream_plan and wait for the streaming to
complete (still under the parallelism_semaphore). So at most
parallelism_semaphore (100) stream_plans can be in parallel.
The parallelism_semaphore limits the parallelism of both checksum and the
streaming plan. However, it is not necessary to have the same
parallelism for both checksum and streaming, because 1) a streaming
operation itself runs in parallel (handling ranges on all shards in
prallel, sending mutaitons in parallel) , 2) and with more streaming plan
(in worse case 100) means we can write to 100 memtables at the same time
and flush 100 memtables to disk at the same time which can take a lot of
memory.
With this patch, we only allow one stream plan in flight.
If we do two truncates in a row, the second will have neither memtable
nor sstable data. Thus we will not write/remove sstables, and thus
get no resulting truncation replay position.
Message-Id: <1497378469-6063-1-git-send-email-calle@scylladb.com>
This patch makes storage proxy to choose replicas to read from base on
their cache hit rates. Replicas with higher cache hit rates will see
more requests while replicas with lower hit rates will see less. Local
node has a special bonus and will get more requests even if another node
has slightly higher cache hit rate (same goes for local vs remote DC),
but after the patch it is no longer guarantied that a coordinator node
will be chosen as a replica for the read (if the feature is enabled).
Currently storage proxy has to loop over remaining replicas to search
for suitable extra replica, but doing it in filter_for_query() is
extremely easy, so do it there instead.
Merge filter_for_query_dc_local() functionality into filter_for_query().
This is more efficient since filter_for_query_dc_local() partitions
endpoints into 'local' and 'remote' set but filter_for_query() already
does it for CL=LOCAL so for such queries we needlessly do it twice.
Node may go down, so after it restarts cache hit rate info will be
incorrect and it can be overwhelmed with traffic until new and
up-to-date cache hit rate arrives. Solve this by dropping node's
information on connection reset, it is more accurate than relying on
gossip which may be slow and miss reboot of a node.
When a node start it does not have any information about cache temperature
of other nodes in the cluster and it is hard (if not impossible) to make
right guess. During cluster startup all nodes have cold caches, so there
is no point to redirect reads to other nodes even though local cache it
cold, but if only that node restarted than other nodes have populated
cache and reads should be redirected.
The node will get up-to-date information about other nodes caches,
but only after receiving first reply, until then it does not have the
information to make right decisions which may cause unwanted spikes
immediately after restart. Having cache temperature in gossiper helps
to solve the problem.
This patch adds new class cache_hitrate_calculator whose responsibility
is to periodically calculate average cache hit rates between all shards
for each CF.
end_bound() returns temporary object (end_bound_ref), so it cannot be
taken by reference here and used later. Copy instead.
Message-Id: <20170612132328.GJ21915@scylladb.com>
Apparently some GDB versions (7.11.1-86.fc24) don't parse double '>'
in a type name, so this:
std::pair<utils::UUID const, seastar::lw_shared_ptr<column_family>>
should be this:
std::pair<utils::UUID const, seastar::lw_shared_ptr<column_family> >
Message-Id: <1497256644-4335-1-git-send-email-tgrabiec@scylladb.com>
The problem is that 'key' is a 'bytes' object now, which doesn't have __format__.
Fixes the following error:
Traceback (most recent call last):
File "~/src/scylla/scylla-gdb.py", line 184, in invoke
TypeError: non-empty format string passed to object.__format__
Error occurred in Python command: non-empty format string passed to object.__format__
Message-Id: <1497253433-374-2-git-send-email-tgrabiec@scylladb.com>
"This series switches repair to use more stream plans to stream the mismatched
sub ranges and use a range generator to produce sub ranges.
Test shows no huge memory is used for repair with large data set.
In addition, we now have a progress reporter in the log how many ranges are processed.
Jun 06 14:18:22 [shard 0] repair - Repair 512 out of 529 ranges, id=1, keyspace=myks, cf=mytable, range=(8526136029525195375, 8549482295083869942]
Jun 06 14:19:55 [shard 0] repair - Repair 513 out of 529 ranges, id=1, keyspace=myks, cf=mytable, range=(8526136029525195375, 8549482295083869942]
Fixes #2430."
* tag 'asias/fix-repair-2430-branch-master-v1' of github.com:cloudius-systems/seastar-dev:
repair: Remove unused sub_ranges_max
repair: Reduce parallelism in repair_ranges
repair: Tweak the log a bit
repair: Use more stream_plan
repair: iterator over subranges instead of list
From seastar-dev.git calle/concorde
Normally, we require that all mutations applied to a column family
have replay positions higher than all previously flushed.
The main reason for this is to be able to determine when to drop a
commit log segment, i.e. determine that all replay positions less
than X are now in sstables.
This patch series, small as it is, relaxes this by instead of just
keeping track of high rp applied, keep a reference count to each
segment per CF in memtables, and on flush, release this very count.
The only case where we need to keep a water mark for RP is then
for table truncation, for which we simply say that the highest RP
applied to the column family is the lowest allowed henceforth,
and use the old reordering logic for this instead. I.e. very rare.
There is of course one (big?) downside to all this, and this is
"normal" commit log replay on startup after crash/shutdown.
Since we relax RP ordering, we cannot use RP:s in sstables as
low marks for replay start, since it is now allowed to exist
non-persisted mutations in commitlog with lower RP:s than
previously flushed. I.e. we more or less always have to replay
the full commit log.
It is worth noting though that due to compaction and the non-
propagation of RP marks to new sstables, we end up often
doing this anyway, so it is hard to say how much of a regression
this is.
truncate
With commitlog keeping use-count per CF id, we can ease the ordering
restriction on replay positiontion. Previously we required that all
added mutations have a position > previously flushed. However, if
we accept that replay must now be all data, by keeping track instead
per CF of highest RP ever entered, we can instead just set a
low mark on truncation, since this is the only remaining hard
RP divider.
Use per CF-id reference count instead, and use handles as result of
add operations. These must either be explicitly released or stored
(rp_set), or they will release the corresponding replay_position
upon destruction.
Note: this does _not_ remove the replay positioning ordering requirement
for mutations. It just removes it as a means to track segment liveness.
Test should
a.) Wait for the flush semaphore
b.) Only compare segement sets between start and end, not start,
end and inbetwen. I.e. the test sort of assumed we started
with < 2 (or so) segments. Not always the case (timing)
Message-Id: <1496828317-14375-1-git-send-email-calle@scylladb.com>
We currently repair all the ranges in parallel.
1) All the ranges will contend for parallelism_semaphore, instead of
processing multiple ranges in parallel and calculating the sub ranges
(which take memory) for each range in parallel, we can handle the ranges
one bye one.
We could have enough parallelism because the checksum are calucated on
all the shards.
2) If for some reason the repair failed, if we handle ranges 1 by 1, we
can log which range of repair is successful. Next time, we can ignore
them. If we start ranges in parallel, it has a high chance, no single
range is completed because all the ranges are on going.
Refs #1912
- Count n out m ranges the repair is running for (kind of progress report)
- Make the 'Found differing range' log debug because it can be millions
of such entries
- Print the failed ranges
In the very beginning, we use a stream_plan for each checksum range.
Later, we changed to use a single stream_plan for all the checksum
ranges. It pushes memory presure to streaming, e.g., millinons of ranges
in a vector to send over RPC.
To fix, we do checksum and streaming in parallel, limit the number of
checksum ranges stored in memory.
Fixes#2430
When starting repair, we divided the large token ranges (vnodes) linto small
subranges of a desired length (around 100 partition), and built a huge list
of those subranges - to iterate over them later and compare checksums of
those chunks.
However, building this list up-front is completely unnecessary, and wastes
a lot of memory: In a test with 1 TB of data, as much as 3 gigabytes was
spent on this list. Instead, what we do in this patch is to find the next
chunk in a DFS-like splitting algorithm, using only the token range
midpoint() function (as before). The amount of memory needed for this is
O(logN), instead of O(N) in the previous implementation.
Refs #2430.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
After change in boot, read_filter is called by distributed loader,
so its update to _filter_file_size is lost. The load variant
which receives foreign components that must do it. We were also
not updating it for newly created sstables.
Fixes#2449.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170606151129.5477-1-raphaelsc@scylladb.com>
At least on Debian8, mk-build-deps -i silently finishes with return code 0
even it fails to install dependencies.
To prevent this, we should manually install the metapackage generated by
mk-build-deps using gdebi.
Fixes#2445
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1496737502-10737-2-git-send-email-syuu@scylladb.com>
Installing openjdk-8-jre-headless from jessie-backports breaks texlive on
jessie main repo.
It causes 'Unmet build dependencies' error when building gdb package.
To prevent this, force insatlling texlive from jessie-backports before start
building gdb.
Fixes#2444
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1496737502-10737-1-git-send-email-syuu@scylladb.com>
"This series fixes some issues with the thrift_server, namely
ensuring that streams and sockets are properly closed.
Fixes#499Fixes#2437"
* 'thrift-server-fixes/v1' of github.com:duarten/scylla:
thrift/server: Close connections when stopping server
thrift/server: Move connection class to header
thrift/server: Shutdown connection
thrift/server: Close output_stream when connection is done
This patch adds the shutdown() function to thrif_server::connection,
and calls it after a connection is done.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The column mask identifies the kind of atom in a row in an sstable. Two
definitions of these values were present: one as a C-style enumeration and one
as a C++11-style enumeration.
The C++11-style definition is used elsewhere in `sstables.cc`. It also offers
additional type-safety.
Therefore, this commit removes the inlined C-style enumeration.
Fixes#2214.
Signed-off-by: Jesse Haber-Kucharsky <jhaberku@scylladb.com>
Message-Id: <c525b4ae7fad3b54480e133921aa4ffe0dd5d9ce.1496352711.git.jhaberku@scylladb.com>
To reduce unwanted dependencies, we need to replace dependency from collectd to
collectd-core.
However, collectd provides /etc/collectd/collectd.conf, so without this package
we need to install the configuration file by our self.
So install the file on .postinst script.
Fixes#2426
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1496231743-7828-1-git-send-email-syuu@scylladb.com>
In the write path we don't wait for view updates, as they happen in
the background.
The view schema tests can fail when running with more than one cpu due
to this inherent race condition: the write to the base table returns
while the view updates are still being processed, after which we issue
a query to the view table. The shard handling the view data is not
guaranteed to finish processing the mutation before handling the query.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170531165726.9212-1-duarte@scylladb.com>
invariant is broken if size of L0 candidates is equal to max
sstable size because the overlapping L1 sstables will not be
added to compacting set, and they will be promoted.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170530143708.3775-1-raphaelsc@scylladb.com>
* seastar 68dbf60...b1f69cc (10):
> metrics: fix namespace in documentation
> add special logger for memory allocation failures
> xen: remove
> Merge "sanity checks, fixes and extensions in the perftune.py" from Vlad
> tutorial: more "seastar" namespace
> execution_stages: fix build errors in comments
> tutorial: more "seastar" namespace additions
> tutorial: more minor changes
> tutorial: minor changes to the introduction
> tutorial: start overhauling the examples to use "seastar" namespace
Mimic origin behaviour, iff TLS encryption is enabled, and
native_transport_port_ssl is set and different from
native_transport_port, start both tls- and non-tls
listeners.
Message-Id: <1496061600-24454-2-git-send-email-calle@scylladb.com>
Removing non-murmur3 partitioners will allow us to reduce memory footprint
and speed up some code by utilizing the properties of the murmur3 partitioner
token.
Message-Id: <20170528172536.16079-1-avi@scylladb.com>
"
- Introduce a parent span IP and span ID paradigm.
- Introduce time series tables to simplify traces processing.
- Add the "How to get traces?" chapter to the tracing.md.
"
* 'tracing-span-ids-and-time-series-helpers-v4' of github.com:cloudius-systems/seastar-dev:
docs: tracing.md: add a "how to get traces" chapter
tracing::trace_keyspace_helper: introduce a time series helper tables
tracing: cleanup: use nullptr instead of trace_state_ptr()
tracing: introduce a span ID and parent span ID
Currently, start and end size of compaction are calculated using the
uncompressed size of data component. bytes_on_disk() returns size
used by all components.
NOTE: start and end size are written to compaction history, so users
who monitor it should be aware of this change.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170525212129.6758-1-raphaelsc@scylladb.com>
sstable::data_size() is used by rebuild_statistics() which only
returns uncompressed data size, and the function called by it
expects actual disk space used by all components.
Boot uses add_sstable() which correctly updates the stat with
sstable::bytes_on_disk(). That's what needs to be used by
r__s() too.
Fixes#1592
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170525210055.6391-1-raphaelsc@scylladb.com>
We can make the dependency more abstract by using mutation_source
instead of an sstable.
Will be useful in some stress tests which want to avoid the disk, but
is also good for the sake of decoupling.
Message-Id: <1495729508-30081-2-git-send-email-tgrabiec@scylladb.com>
Information about interrupts is invaluable when debugging performance
problems with Scylla in the field. node_exporter doesn't include that in
the list of collectors enabled by default, so we suggest we do it here.
The list that goes into this file is the default list as shown by
node_exporter, with "interrupts" added to it.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170525153008.26720-1-glauber@scylladb.com>
"lcs' current behavior will make it hard to reduce number of
levels by increasing sstable size because it uses uncompressed
length when deciding whether or not a level needs promotion.
Demotion process is slower because of that."
* 'lcs_promotion_improvement_2' of github.com:raphaelsc/scylla:
lcs: use sstable compressed length when computing level size
sstables: introduce sstable::ondisk_data_size
sstables: unconditionally set sstable data file size
partial sstable files aren't being removed after each failed attempt
to flush memtable, which happens periodically. If the cause of the
failure is ENOSPC, memtable flush will be attempted forever, and
as a result, column family may be left with a huge amount of partial
files which will overwhelm subsequent boot when removing temporary
TOC. In the past, it led to OOM because removal of temporary TOC
took place in parallel.
Fixes#2407.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170525015455.23776-1-raphaelsc@scylladb.com>
lcs uses uncompressed length of sstables when computing size of a
level, and that may result in unnecessary promotion when the goal
is to reduce the number of levels after an increase in sstable
size, for example.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
this new function is an alternative to data_size(), which will
return size of data component. data_size() returns uncompressed
size of data component if compression is enabled.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Streaming ususally takes long time to complete. Abort it on false
positive idle detection can be very wasteful.
Increase the abort timeout from 10 minutes to a very large timeout, 300
minutes. The real idle session will be aborted eventually if other
mechanisms, e.g., streaming manager has gossip callback for on_remove
and on_restart event to abort, do not abort the session.
Fixes#2197
Message-Id: <57f81bfebfdc6f42164de5a84733097c001b394e.1494552921.git.asias@scylladb.com>
Use the boost::intrusive containers in order to achieve a O(1) complexity
for both "LRU list" update and to minimize the memory overhead in the hash
table item to "LRU list" item connection:
- Make the timestamped_val be both a bi::list and a bi::unordered_set
item.
- Make a bi::unordered_set be a cache backend instead of the
std::unordered_map.
As a result dropping k LRU items becomes an O(k) operation instead of
O(N log N), where N is a total number of all cached items:
- Every time a value is read - move it to the front of the "LRU list"
(O(1)).
- When we need to remove k LRU items:
- Repeat k times:
- Take an element from the back of the "LRU list". (O(1)).
- Remove it from the bi::unordered_set and dispose. (O(1)).
We use an auto-unlink configuration for bi::list, therefore
disposing an item is going to auto unlink it from the list.
* 'permissions_cache_move_to_intrusive-v1' of github.com:scylladb/seastar-dev:
utils::loading_cache: cleanup
utils/loading_cache.hh: use intrusive list to store the lru entry
utils::loading_cache: implement automatic rehashing
utils::loading_cache: make the underlying map to be an intrusive unordered_set
"Enforces commutativity of addition:
m1 + m2 == m2 + m1
and consistency of difference and addition with equality:
m1 + (m2 - m1) == m1 + m2"
* tag 'tgrabiec/fix-range-tombstone-commutativity-v2' of github.com:cloudius-systems/seastar-dev:
mutation: Make compare_*_for_merge() consistent with equals()
tests: mutation: Improve assertion failure message
tests: Use default equality in test_mutation_diff_with_random_generator
mutation: Make counter cell difference consistent with apply
tests: range_tombstone_list_test: Improve error message
tests: range_tombstone_list: Check adjacent range merging
range_tombstone_list: Merge adjacent range tombstones in apply()
tests: mutation: Check commutativity of mutation addition
range_tombstone_list: Avoid violating set invariant
range_tombstone_list: Make tombstone merging commutative
range_tombstone_list: Add erase() operation to the reverter
range_tombstone_list: Make all undo operations ordered relative to each other
utils: Extract to_boost_visitor() to a separate header
allocating_strategy: Introduce alloc_strategy_unique_ptr<>
equals() considers expiring cells to be different form non-expiring cells,
but compare_row_marker_for_merge() considers them equal. Fix the latter to
pick expiring cells. The choice was arbitrary.
The case when both cells are dead was not handled properly, the diff
was always empty, whereas the cell with higher timestamp should win.
Caused test_mutation_diff_with_random_generator to fail.
The code was inserting an entry with the same key as its successor,
and only later adjusting the key of the old entry. This is violating
set's invariant of unique keys, and insertion may cause rebalancing. I
don't know if this violation actually causes problems currently, but
it's safer not to.
Fix by first updating the existing entry and then inserting the new
one.
Example of non-commutative case:
a = [1, 5]@t2
b = {[2, 3]@t1, [4, 5]@t1}
a + b = [1, 5]@t2
b + a = [1, 4)@t2, [4, 5]@t2
After this patch, both will yield [1, 5]@t2.
The patch also changes the logic to handle overlaps of tombstones with
equal timestamps to be handled symmetrically. They are now merged
instead of split on either of the boundary.
Refs #2158.
Later operation may depend on the result of previous operation. Same dependency
is present when reverting the operations.
Fixes assertion failure in update reverter.
Fix the shrink() O(n log n) complexity issue by constantly pushing the corresponding intrusive
list entry to the head of the list every time the values are read.
This will keep the list ordered by the last read time from the most recently read
to the least recently read entry.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
- Start the cache with 256 buckets - the minimum number of buckets.
- Limit the maximal number of buckets by 1M buckets.
- Keep the load factor between 0.25 and 1.0 as long as the number of buckets is
between the minimum and the maximum values mentioned above.
- Grow and shrink the hash every "refresh" period if needed.
- Enable bi::power_2_buckets and bi::compare_hash bi::unordered_set options.
- Enable bi::unordered_set_base_hook's bi::store_hash option.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Make the underlying map to be a boost::intrusive::unordered_set<timestamped_val>
instead of std::unordered_set<Key, timestamped_val>.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Metadata is read using default priority class, which can significantly
slow down the process under high load. Compaction class can be used,
and if it turns out to be a problem, we can switch to a special class
for it.
Fixes#1859.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170517184546.17497-1-raphaelsc@scylladb.com>
"fixes a problem in which memory requirement for loading in-memory
components of sstables is very high due to unlimited parallelism."
* 'mem_requirement_sstable_load_v2_2' of github.com:raphaelsc/scylla:
database: fix indentation of distributed_loader::open_sstable
database: reduce memory requirement to load sstables
sstables: loads components for a sstable in parallel
sstables: enable read ahead for read of in-memory components
sstables: make random_access_reader work with read ahead
SSTable load temporarily uses more space than needed to store metadata,
due to:
1) All components are read using read_simple() which uses 128k buffer.
file::dma_read_bulk() will allocate 128k, and may potentially allocate
another big buffer (128k - read) for file::read_maybe_eof().
2) read_filter() may use double the space it needs to.
Due to the fact that sstable loading parallelism is unlimited, Scylla
may require much more memory to load all sstables, and that may lead to
OOM. Higher the number of sstables higher the memory overhead.
To confirm this problem, I wrote a test[1] which loads 30k sstables in
parallel and reports the memory usage peak in the end.
When loading 30k sstables, each of which metadata is ~300kb, memory
usage peak was ~18G. When loading completed, only ~9GB were needed to
store all the metadata.
[1]: https://gist.github.com/raphaelsc/2db37b4fb34301833ab9eeed3b1a524d
To fix this problem, we need to set a limit on load parallelism (let's
start with a small number like 3 and adjust later if needed) and rely
on readahead so that the requirement drops considerably without
increasing boot time. Actually, boot time is improved by it.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Reviewed-by: Nadav Har'El <nyh@scylladb.com>
Read ahead 4 is used. Let's adjust it later if needed. File size is
used to prevent file_input_stream from issuing useless reads beyond
file size with read ahead enabled. We can switch to variant without
length once file_input_stream handles it properly.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Scylla crashes if read ahead is enabled by file_random_access_reader
because a call to seek() destroys the existing input stream without
closing it first.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
The bootstrapping node will be a gossip only member, until the streaming
finishes and the node becomes NORMAL state. If during this time, the
bootstrapping node is overwhelmed with streaming, it is possible the node will
delay the update the gossip heartbeat. Be forgiving for the bootstrapping node
and do not remove it from gossip too fast. Otherwise, streaming rpc verbs will
not be resent becasue the node is not in gossip membership anymore.
Fixes#2150
Message-Id: <286d7035d854f2a48abf4e1e2e3bfcb8b22b9ca2.1494553580.git.asias@scylladb.com>
We are currently suspecting that the bloom filter false positive ratio
is not being respected. While trying to debug that, I found out that we
have a more basic problem:
The numbers are all meaningless, because the stats are wrong. We are
accumulating by summing the ratios together. It's easy to see how this
doesn't work, if we look at an example where the ratio for some CFs is
zero:
SST1: false = 1, total = 2. ratio = 0.5
SST2: false = 0, total = 98 . ratio = 0.
The real ratio in this example is 1 / (98 + 2) = 1 %, but the displayed
ratio will be 0.5 + 0 = 0.5.
This patch will map reduce all the sstables together keeping both
numerator and denominator, yielding the right value at the end. To do
that, we'll reuse the existing ratio_holder class, which already does
exactly what we want.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170518222333.16307-1-glauber@scylladb.com>
- introcduced "seastarx.hh" header, which does a "using namespace seastar";
- 'net' namespace conflicts with seastar::net, renamed to 'netw'.
- 'transport' namespace conflicts with seastar::transport, renamed to
cql_transport.
- "logger" global variables now conflict with logger global type, renamed
to xlogger.
- other minor changes
* seastar 45b718b...f726938 (2):
> memory: add --mbind option to supress warning message when running Seastar apps on container
> Add support for Gentoo Linux irqbalance configuration detection.
"This series fixes bugs related to materialized views, most pertaining
to column filtering in the where clause."
* 'materialized-views/bug-fixes/v1' of https://github.com/duarten/scylla:
tests/view_schema_test: Add more test cases
tests/cql_assertions: Add assertion for row set equality
single_column_relation: Correctly print IN relation
statement_restrictions: Allow filtering regular columns for views
statement_restrictions: Relax clustering restrictions for views
statement_restrictions: Relax partition restrictions for views
cql3/statements: Prevent setting default ttl on view
cql3/restrictions: Complete implementation of is_satisfied_by()
db/view: Re-implement clustering_prefix_matches()
db/view: Re-implement partition_key_matches()
db/view: Generate regular tombstone for base deletions
db/view: Consider cell liveness when generating updates
db/view: Don't generate view updates for static rows
"There are numerous issues in the current implementation of permissions
cache starting from the logical errors and bugs and ending with the
suboptimal implementation described in the issue #2262."
* 'permissions_cache_fixes-v4' of github.com:scylladb/seastar-dev:
utils::loading_cache: avoid the reads storm when the key is not in the cache
utils::loading_cache: cleanup
utils::loading_cache: align the constrains in the constructor with the parameters description
utils::loading_cache: refresh in the background
auth::auth: add operator<<() for a permission_cache key
auth::auth::permissions_cache: use the values from the configuration - don't try to be smart
db::config: define a saner default value for permissions_validity_in_ms
mutation_partition has a slicing constructor which is supposed to copy
only the rows from the query range. The rows are located using
nonwrapping_range::lower_bound() and
nonwrapping_range::lower_bound(). Those two have two different
implementations chosen with SFINAE. One is using std::lower_bound(),
and one is using container's built in lower_bound() should it
exist. We're using intrusive tree in mutation_partition, so
container's lower_bound() is preferred. It's O(log N) whereas
std::lower_bound() is O(N), because tree's iterator is not random
access.
However, the current rule for picking container's lower_bound() never
triggers, because lower_bound() has two overloads in the container:
./range.hh:618:14: error: decltype cannot resolve address of overloaded function
typename = decltype(&std::remove_reference<Range>::type::upper_bound)>
^~~~~~~~
As a result, the overload which uses std::lower_bound() is used.
Spotted when running perf_fast_forward with wide partition limit in
cache lifted off. It's so slow that I timeouted waiting for the result
(> 16 min).
Fixes#2395.
Message-Id: <1495048614-9913-1-git-send-email-tgrabiec@scylladb.com>
"These patches add support to setup and operate ScyllaDB on Gentoo Linux.
* scylla_setup and related scripts
* node_health_check
I have kept them as simple as possible and tested them to setup and operate
succesfully a three nodes cluster running on Gentoo Linux."
* 'gentoo_linux_support' of github.com:ultrabug/scylla:
scylla_setup: add gentoo linux installation detection
prometheus node_exporter install: add support for gentoo linux
raid setup: add support for gentoo linux
ntp setup: add support for gentoo linux
kernel check: add support for gentoo linux
cpuscaling setup: add support for gentoo linux
coredump setup: add support for gentoo linux
detect gentoo linux on selinux setup
add gentoo_variant detection and SYSCONFIG setup
According to description of permissions_validity_in_ms the permissions_cache is enabled if this
value is set to a non-zero value. Otherwise the permissions_cache is disabled.
According to the permissions_update_interval_in_ms description it must have a non-zero value if permissions_cache
is enabled.
permissions_cache_max_entries description doesn't explicitly state it but it makes no sense to allow it to be zero
if permissions_cache is enabled.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This patch changes the way a loading_cache works.
Before this patch:
1) If a permissions key is not in the cache it's loaded in the foreground and the original
query is blocked till the permissions are loaded.
2) Every _period the timer does the following:
1) If a value was loaded more than _expiry time ago it is removed from the cache.
2) If the cache is too big - the less recently loaded values are removed till the cache
fits the requested size.
After this patch:
1) If a permissions key is not in the cache it's loaded in the foreground and the original
query is blocked till the permissions are loaded.
2) Every _period the timer does the following:
1) If a value in the cache was loaded or read for the last time more than _expiry time ago - it's removed from the cache.
2) If the cache is too big - the less recently read values are removed till the cache fits the requested size.
3) The values that were loaded more than _refresh time ago are re-read in the background.
The new implementation allows to minimize the amount of the foreground reads for a frequently used value to a single
event (when the value is loaded for the first time).
It also ensures we do not reload values we no longer need.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Our configuration already has the default values for for permission cache parameters.
Therefore if user decides to give some bad parameters we'd rather fail the load and inform him/her
about the bad parameters instead of trying to silently "fix" them.
In addition the original code wasn't passing the parameters correctly: it switched the "expiry" and "refresh" parameters in
the utils::loaded_cache constructor.
Add to this that the original code was doing really strange things in the permission_cache::expiry(cfg) method.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
It makes little sense to have the same value for permissions_update_interval_in_ms and permissions_validity_in_ms.
This may cause the values to be invalidated only because some minor delays in the timer scheduling.
It makes a lot more sense to make the permissions_update_interval_in_ms value smaller than permissions_validity_in_ms.
This way we would minimize the chances of "false invalidation" due to some small delays in the timer scheduling.
In addition, 2s seems to be a too small value for permissions_validity_in_ms since our default read_request_timeout_in_ms is 5s.
This means that a single system_auth read failure would guarantee that the following queries are going to read system_auth data
in the foreground.
Setting it to 10s would allow a second read attempt before we enforce the foreground read.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
"Notably:
- add validation of the results (e.g. fragment count, expectations about disk activity)
- add cache-specific tests"
* 'tgrabiec/add-cache-tests-to-perf-fast-forward' of github.com:cloudius-systems/seastar-dev:
tests: perf_fast_forward: Report cache stats
row_cache: Keep counters in a struct
tests: perf_fast_forward: Add cache-specific tests
tests: perf_fast_forward: Extract test_reading_all()
tests: perf_fast_forward: Add validation of the results
tests: perf_fast_forward: Fix partition scans to read the expected amount of fragments
tests: perf_fast_forward: Allow the test to be interrupted
tests: perf_fast_forward: Allow testing with cache enabled
row_cache: Implement mutation_reader::fast_forward_to() for cache scanner
Not 100% proper, but in line with how we still store the info.
Ensures (helps at least) to keep schema loaded from tables
and schema from builder comparable.
Fixes schema_changes_test error.
Message-Id: <1495030581-2138-2-git-send-email-calle@scylladb.com>
"When mutation reader enters the partition using index,
streamed_mutation object is returned to the user before the row start
fragment is processed. In that case, when we process the row start, we
should ignore it and not call setup_for_partition() again. That may
override user's fast_forward_to() request."
* 'tgrabiec/fix-initial-fast-forward-to-for-single-key-sstable-readers' of github.com:scylladb/seastar-dev:
tests: mutation_source_test: Test forwarding in single-key readers
sstables: Remove unused code
sstables: mutation_reader: Fix setup_for_partition() being called twice in some cases
sstables: Fix verify_end_state() to tolerate ATOM_START_2 state
"This series adds private repository support to scylla-housekeeping"
* 'amnon/housekeeping_private_repo_v3' of github.com:cloudius-systems/seastar-dev:
scylla-housekeeping service: Support private repositories
scylla-housekeeping-upstart: Use repository id, when checking for version
scylla-housekeeping: support private repositories
make_pkeys() needs to be invoked with n equal to the number of keys
which the table was populated with. Otherwise the extra keys, which
are missing in the table, may be placed anywhere in the vector due to
ring order sorting, and break the assumption that the table contains
all keys from the array up to index n. This resulted in the test
reading slighlty less fragments than it would follow from the desired
count.
Another problem is that we should not skip the fast_forward_to() call
for the inital range (workaround for a bug in sstable mutation
reader), otherwise we will read slightly less than expected as well.
From http://github.com/avikivity/scylla exponential-sharder/v3.
The sharder, which takes a range of tokens and splits it among shards, is
slow with large shard count and the default
murmur3_partitioner_ignore_msb_bits.
This patchset fixes excessive iteration in sstable sharding metadata writer and
nonsignular range scans.
Without this patchset, sealing a memtable takes > 60 ms on a 48-shard
system. With the patchset, it drops below the latency tracker threshold I
used (5 ms).
Nonsingular queries used exponential expansion of the token space to
avoid spending too much cpu time on near-empty tables, but the generation
of the search space was itself exponential. Switch to the exponential sharder
which has linear cost.
The "exponentiality" is not carried over from one range to another, because
we expect one or two ranges (two ranges result from a wrapped around thrift
token range).
Like the regular sharder, the exponential sharder divides a range into
subranges owned by individual ranges. Unlike the regular sharder, it
generates ever-increasing subranges, spanning more and more shards, and
eventually returns several subranges per shard. To avoid using
exponential cpu and memory, subranges belonging to a single shard are merged,
and a flag is set to indicate the subranges are not ordered wrt. each other.
Right now, next_token_for_shard() only allows iterating linearly in shard
order. Add the ability to select a specific shard to skip to (in case we're
only interested in a single shard), and to select larger ranges (so that
exponential increases are not implemented by iteration).
The partitioners now depend on smp::count to be initialized correctly,
but smp::count isn't available at static initialization time.
The scylla executable isn't affected because it calls set_global_partitioner()
after smp::count has been initialized.
Fix by deferring initialization to the first global_partitioner() call.
"This patch ensures we read the base table rows that an update
is modifying, in order to correctly calculate the set of
materialized view updates.
The read-before-write is performed on the shard applying the
update and attempts to do a precise read of the rows being modified,
which can be more than one in case of ranged deletions or a
batch update."
* 'materialized-views/read-existing/v2' of https://github.com/duarten/scylla:
database: Read existing base mutations
db/view: Calculate clustering ranges for MV read-before-write query
db/view: Replace entry if cells don't match
view_info: Store base regular col in the view's PK as column_id
compound_view_wrapper: Add tri_compare
bound_view: Build range bound from bound_view
clustering_bounds_comparator: Enable Range concept
range: Add lvalue version of transform()
tests: Add test case for nonwrapping_range::intersection()
nonwrapping_range: Add intersection() function
When generating updates for a materialized view we need to read the
existing base row, to be able to determine the primary key of the view
row the new base update will supplant, in case the view includes a
base non-primary key column in its own primary key. That old view row
will be tombstoned or updated, if it exists, depending on the difference
between the new base row and the existing one, if any.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
For row set equality, the order of the actual rows doesn't need to
match the order of the expected rows.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Introduce the calculate_affected_clustering_ranges() function to
calculate the smallest subject of affected clustering ranges that we
need to query for.
The update_requires_read_before_write() function checks whether
a view is potentially affected by the base update.
The patch also cleans up the may_be_affected_by() function.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
So that the output of a set of relations can be fed back into the CQL
parser; useful for materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
If a base table regular columns is part of the view's pk, and if that
column changes, we should replace the entry, by deleting the row(s)
with the old value and inserting a new one.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
In process_clustering_columns_restrictions(), don't require all
clustering columns to be restricted if we're dealing with a
materialized view's where clause.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
In process_partition_key_restrictions(), don't require all partition
key columns to be restricted if we're dealing with a materialized
view's where clause.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch implements the is_satisfied_by() function for the remaining
types of restrictions, lifting the function declaration to
abstract_restrictions.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch implements clustering_prefix_matches() in terms of
abstract_restriction::is_satisfied_by() instead of ranges, which
supports filtering just a subset of the clustering columns.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch implements partition_key_matches() in terms of
abstract_restriction::is_satisfied_by() instead of ranges, which
supports filtering just a component of a compound partition key.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures we take into account the liveness of the base's
regular column in the view's pk when generating view updates.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch stores the base_non_pk_column_in_view column as column_id,
which is more convenient, and it also stores a two-level optional to
encode both lazy initialization and the absence of such a column.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We introduce the bound_view::to_range_bound() function, which builds a
wrapping_range or nonwrapping_range bound from a bound_view.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
intersection() returns an optional range with the intersection of the
this range and the other, specified range.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"Defines origin v3-format for system/schema tables, and use them for
schema storage/retrival.
Includes a legacy_schema_migrator implementation/port from origin. Note
that since we don't support features like triggers, functions and
aggregates, it will bail if encountering such a feature used.
Note also that this patch set does not convert the "hints" and
"backlog" tables, even though these have changed in v3 as well.
That will be a separate patch set.
Tested against dtests. Note that patches for dtest + ccm
will follow."
* 'calle/systemtables' of github.com:cloudius-systems/seastar-dev: (36 commits)
legacy_schema_migrator: Actually truncate legacy schema tables on finish
database: Extract "remove" from "drop_columnfamily"
v3 schema test fixes
thrift: Update CQL mapping of static CFs
schema_tables: Use v3 schema tables and formats
type_parser: Origin expects empty string -> bytes_type
cf_prop_defs: Add crc_check_chance as recognized (even if we don't use)
types_test: v3 style schemas enforce explicit "frozen" in tupes/ut:s
cql3_type: v3 to_string
cql_types: Introduce cql3_type::empty and associate with empty data_type
schema: rename column accessors to be in line with origin
schema: Add "is_static_compact_table"
schema_builder: Add helper to generate unique column names akin origin
schema: Add utility functions for static columns
schema: Use heterogeneous comparator for columns bounds
cql3_type_parser: Resolve from cql3 names/expressions
cql3_type: Add "prepare_interal" and "references_user_type"
cql3::cql3_type: Add prepare_internal path using only "local" holders
cql3_type: Add virtual destructor.
database/main: encapsulate system CF dir touching
...
* seastar 4a3118c...45b718b (7):
> tests: make connect_test use a random port
> log: Introduce log.info0
> configure.py: link to DPDK PMD drivers which are already built on build/dpdk and enabled by default on DPDK config
> Update fmt submodule
> perftune: fix perftune.py IndexError when NIC uses less IRQs than requested.
> build: Add more required build dependencies to the Dockerfile
> Prometheus: Reserve in protobuf object before iterating
"Tested with:
- test.py --mode relase
- debug/test-serialization
- c-s with both debug and relase compiled scylla with authentication enabled:
cassandra-stress write n=10000 no-warmup -rate threads=10 -mode native unprepared cql3 user='cassandra' password='cassandra'
Tested with:
- test.py --mode relase
- debug/test-serialization
- c-s with both debug and relase compiled scylla with authentication enabled:
cassandra-stress write n=10000 no-warmup -rate threads=10 -mode native unprepared cql3 user='cassandra' password='cassandra'"
* 'compress_tracing_session_id-v6' of github.com:cloudius-systems/seastar-dev:
cql_server::response: rework the tracing session ID insertion
utils::UUID: align the UUID serialization API with the similar API of other classes in the project
utils: serialization: unify the variety of serialize_XXX(...)
cql_server::response: rework the compress(...) method
cql_server::response: store the frame flags inside the class
"This series fixes a set of regressions introduced by
f7bc88734a, resulting in two failed
tests:
testDenseNonCompositeTable(org.apache.cassandra.cql3.validation.operations.CreateTest)
and
testStaticColumnsWithDistinct(org.apache.cassandra.cql3.validation.entities.StaticColumnsTest)"
* 'cql-fixes/v1' of github.com:duarten/scylla:
update_statement: Reject empty values for dense clustering key
modification_statement: Fix detection of clustering keys
cql3/restrictions/statement_restrictions: Consider statement type
cql3/statements/modification_statement: Extract statement_type
Insert the tracing session ID into the response body in the cql_server::response constructor.
Fixes#2356
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
The standard serialization API (e.g. in data_value) includes the following methods:
size_t serialized_size() const;
void serialize(bytes::iterator& it) const;
bytes serialize() const;
Align the utils::UUID API with the pattern above.
The only addition is that we are going to make an output iterator parameter of a second method above
a template so that we may serialize into different output sources.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Use the same templated implementation for all different serialize_XXX(...).
The chosen implementation is based on the std::copy_n(char*, size, OutputIterator),
which is heavily optimized and will be using memcpy/memmove where possible.
This patch also removes the not needed specializations that accept signed integer
values since we were casting them to unsigned value anyway.
The std::ostream based specifications are also removed since they are not used
anywhere except for a test-serialization.cc and adjusting the ostream to the iterator
is a single-liner.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Cleanup the compress(...) method interface:
- Encapsulate the technical details inside the method:
- Re-write the _body inside the method instead of returning it.
- Set the response::_flags inside the method.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
It makes a lot more sense to keep the flags mask inside the response and update it each time
the corresponding feature is set instead of holding the separate components like tracing state
pointer.
This patch adds this ability to set the flags.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
The code that removes each sstable runs in a thread. Parallel
removing of a lot of sstables may start a lot of threads each of which
is taking 128k for its stack. There is no much benefit in running
deletion in parallel anyway, so fix it by deleting sstables sequentially.
Fixes#2384
Message-Id: <20170516103018.GQ3874@scylladb.com>
When mutation reads enters the partition using index,
streamed_mutation object is returned to the user before the row start
fragment is processed. In that case, when we process the row start, we
should ignore it and not call setup_for_partition() again. That may
override user's fast_forward_to() request.
We would be in that state if consume_row_start() returns porceed::yes
and the stream ends after that. This can happen if slicing using
promoted index determined that there are no cells in the partition in
the range.
Newer setuptools parse_version() don't like dashed version strings,
so we should trim it to avoid false negative version_compare() checks.
Signed-off-by: Alexys Jacob <ultrabug@gentoo.org>
Message-Id: <20170511162646.22129-1-ultrabug@gentoo.org>
Prevent the accidental dropping of system_auth and system_traces objects (keyspaces and tables)
but allow their modification (including tables).
We need to be able to modify keyspases in order to set/modify the replication strategy and its parameters.
We need to be able to ALTER the tables in order to allow rolling upgrades when some of the tables has changed.
Fixes#2346Fixes#2338
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1494363335-20424-1-git-send-email-vladz@scylladb.com>
_underlying is created with _range, which is captured by
reference. But range_and_underlyig_reader is moved after being
constructed by do_with(), so _range reference is invalidated.
Fixes#2377.
Message-Id: <1494492025-18091-1-git-send-email-tgrabiec@scylladb.com>
Now that update_statement uses statement_restrictions, we need our
validation logic to take the statement type into account, in
particular to deal with insertion statements which only set static
columns but specify clustering values.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts the statement_type into its own file. The type
will be later passed to statement_restrictions for validation
purposes.
Further along, we could add methods to it that currently live in other
statements so we can move more validation into statement_restrictions.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch updates the mapping of static CFs so that their CQL
representation is a non-compound, non-dense schema with static
columns, instead of regular ones. This matches the representation os
static CFs in Cassandra 3.x.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
More pointedly: Expose columns as is (currently
all_columns_in_select_order), expose name->column mapping more
appropriately named.
Renaming like this is not strictly neccesary, but there is a point to
trying to keep nomenclature similar-ish with origin, esp. when select
order column need to become filtered (spoiler alert).
Cassandra 3 uses cql names for column/field types, thus
we need to parse these out-of-line, and resolve more akin
to the cql parser.
Also wrap building user types similarly to origin, using
a "builder" wrapper, and usage graph resolving.
Regular single key query will never reconcile with CL=ONE since there
will be no digest mismatch, but range queries do not have digest stage,
so always goes through reconcile code. For CL=ONE there will be only one
result though, so no need to run complicated reconciliation logic and the
only result can be returned directly.
Message-Id: <20170509100334.GQ28272@scylladb.com>
The "_specs" array contains column specifications that have the bind
marker name if there is one. That results in
get_partition_key_bind_indices() not being able to look up a column
definition for such columns. Fix the issue by keeping track of the
actual column specifications passed to add() like Cassandra does.
Fixes#2369
Message-Id: <1494397358-24795-1-git-send-email-penberg@scylladb.com>
Having a varadic parameter being used in implicit sprint is not
very readable + makes it less intuitive when suddenly system keyspace
becomes more than one -> multiple sprints in the chain -> more confusion
or more execution paths.
Its not that horrible with some spread out sprint:s
Reduces processing overhead and network traffic.
We can't use the NO_METADATA flag in the metadata object, because this
is a request attribute; different executions of the same prepared statement
can have different settings for skip_metadata.
Message-Id: <20170419175145.19766-1-avi@scylladb.com>
"This patch series adds CQL front-end support for secondary indices. You
can now execute CREATE INDEX and DROP INDEX statements, which will
update the newly added "Indexes" system table. However, the indexes are
not actually backed up by anything nor are they available for CQL
queries. The feature is hidden behind a new cluster feature flag and
enabled only with the "--experimental" flag."
* 'penberg/cql-2i/v2' of github.com:cloudius-systems/seastar-dev: (34 commits)
schema: Kill index_type enum
schema: Kill index_info class
cql3/statements/create_index_statement: Use database::existing_index_names() in validation
cql3/statements: Use secondary index manager in alter_table_statement class
index: Add secondary_index_manager
thrift/handler: Use index_metadata
db/schema_tables: Index persistence
schema: Add all_indices() to schema class
schema: Remove add_default_index_names() from schema_builder class
db/schema_tables: Add system table for indices
cql3/Cgl.g: DROP INDEX
cql3/statements: Add drop_index_statement class
database: Add find_indexed_table() to database class
cql3: Return change event from announce_migration()
cql3/statements: Multiple index targets for CREATE INDEX
cql3/statements: Use index_metadata in create_index_statement class
cql3/statements: Use feature flag in create_index_statement class
service/storage_service: Add feature flag for secondary indices
database: Add get_available_index_name() to database class
schema: Add get_default_index_name() to index_metadata class
...
Fix the CQL front-end to populate the partition key bind index array in
result message prepared metadata, which is needed for CQL binary
protocol v4 to function correctly.
Fixes#2355.
Message-Id: <1494247871-3148-1-git-send-email-penberg@scylladb.com>
"This series introduces partial support for range deletions. This allows
deletion operations such as
delete from cf where p=1 and c > 0 and c <= 3.
This series only adds support for single-column range restrictions.
We enforce that both range bounds be specified, because we can't represent
infinite bounds in the current sstable format. Such bounds are represented
as a prefix with no components, with the bound_kind informing whether they
are a bottom of top bound.
We're currently unable to serialize an infinite bound in such a way that it
would be correctly interpreted by Cassandra 2.2.x. A serialized bound is a
composite with a (<length><value><EOC>)+ format. While we could technically
represent the bottom bound, the top bound, if written as a single component
with 0 bytes in size and some EOC, would always sort before other values.
The same would happen if represented as an empty (no components) composite,
because in Cassandra 2.2.x those always have EOC = NONE.
This limitation should stay in place until we can properly represent range
tombstones in the storage format."
* 'range-deletions/v2' of https://github.com/duarten/scylla:
mutation: Set cell using clustering_key_prefix
mutation_partition: Harmonize apply_delete overloads
prefix_compound_view_wrapper: Add is_full and is_empty functions
tests/cql_query_test: Add range deletion tests
cql3: Partially support ranged deletions
single_column_primary_key_restrictions: Implement has_bound()
modification_statement: Use statement_restrictions for where clause
statement_restrictions: Expose primary key restrictions
to_string: Add missing include
Current script has a bug that overwrites symlink on
/etc/sysconfig/selinux by real file, the script not able to disable
SELinux because of it.
So keep symlink after modifying the file.
Fixes#2279
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1493663263-10573-1-git-send-email-syuu@scylladb.com>
This is a folded version of the following rework on the node health
check script:
- Added support for non-default cql + nodetool ports
- Script will not exit if either Scylla-server / Scylla-jmx / Both
services are not up and running. It will alert the user about it and
which output cannot be collected, but continue collecting everything
else.
- Removed lshw installation and non-needed use in commands
- Script supports RHEL/CentOS/Ubuntu14/Ubuntu16/Debian (tested on all
beside Debian, should behave the same as Ubuntu14/16)
- All Indentation issues fixed -> using only tab (no spaces)
consistently.
- >> vs. > was fixed as well in the needed places.
- Changes the ${VAR_NAME} instances to $VAR_NAME, and kept the {} only
where needed.
- Check Scylla service as Vlad recommended using 'ps -C'
- Fixed the CQL not listening error message.
- Added Sanity check if script is attempted to run on non-Fedora and
non-Debian OS -> alert the user and exit.
- Removed the MANUAL CHECK LIST section (moved to Google Forms)
- Added date in head of the report.
- Removed text from Report's "PURPOSE" section, which was referring to
the "MANUAL CHECK LIST" (not needed anymore).
[ penberg: Fold into a single commit and add proper license. ]
Acked-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1493900076-29170-1-git-send-email-penberg@scylladb.com>
"gcc 7 doesn't like some of our code, so adjust to make it happy."
* 'gcc7' of http://github.com/avikivity/scylla:
Remove exception specifications
commitlog: handle noexcept conflict between unlink and function object
thrift: change generated code namespace
org::apache::cassandra (the generated namespace name) gets confused with
apache::cassandra (the thrift runtime library namespace), either due to
changes in gcc 7 or in thrift 0.10. Either way, the problem is fixed
by changing the generated namespace to plain cassandra.
"Fixes abort which happens when making a single-key query for a key which
is after all keys present in the sstable."
* 'tgrabiec/fix-abort-in-index-reader' of github.com:scylladb/seastar-dev:
tests: mutation_source_test: Add test cases for single-key out of range reads
sstables: index_reader: Remove redundant function
sstables: index_reader: Fix abort in advance_and_check_if_present()
Change the clustering key argument in mutation::set_cell from
exploded_clustering_prefix to clustering_key_prefix, which allows for
some overall code simplification and fewer copies. This mostly affects
the cql3 layer.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures the different mutation_partition::apply_delete()
overloads behave similarly, so that, for example, an empty clustering
key is treated the same way as an empty
exploded_clustering_key_prefix.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces partial support for range deletions. This allows
deletion operations such as
delete from cf where p=1 and c > 0 and c <= 3.
We enforce that both range bounds be specified, because we can't represent
infinite bounds in the current sstable format. Such bounds are represented
as a prefix with no components, with the bound_kind informing whether they
are a bottom of top bound.
We're currently unable to serialize an infinite bound in such a way that it
would be correctly interpreted by Cassandra 2.2.x. A serialized bound is a
composite with a (<length><value><EOC>)+ format. While we could technically
represent the bottom bound, the top bound, if written as a single component
with 0 bytes in size and some EOC, would always sort before other values.
The same would happen if represented as an empty (no components) composite,
because in Cassandra 2.2.x those always have EOC = NONE.
This limitation should stay in place until we can properly represent range
tombstones in the storage format.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch replaces the custom where clause processing by adding and
using a statement_restrictions field to modification_statement.
This improves code reuse and also moves some checks to prepare-time.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This changes announce_migration() to return a change event directory in
schema_altering_statement base class. It's needed for drop index
statement, which does not know the keyspace or column family until it
looks up them based on the index. Two stage approach of announcing a
migration and then creating the change event won't work because in the
latter stage, the lookup will fail. The same change in
announce_migration() has been applied to Apache Cassandra.
This adds a find_index_nomame() helper to the schema class, which
searches for index that is otherwise equal but ignores the name of the
index in comparison. This is needed to for CREATE INDEX to reject
duplicate index creation.
After compaction revamp, compaction type set by cleanup at its ctor
is being overwritten at compaction::setup(). Consequently, cleanup
would not be stopped by 'nodetool stop cleanup' and cleanup would
be listed as regular compaction in 'nodetool compactionstats'.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170504013046.23522-1-raphaelsc@scylladb.com>
We estimate number of partitions for a given range of a column familiy
and split the range into sub ranges contains fewer partitions as a
checksum unit.
The estimation is wrong, because we need to count the partitions on all
the shards, instead of only counting the local shard.
Fixes#2299
Message-Id: <7876285bd26cfaf65563d6e03ec541626814118a.1493817339.git.asias@scylladb.com>
"This series fixes the user after free issue in gossip and elimates the
duplicated / unnecessary mark alive operations.
Fixes#2341"
* tag 'asias/gossip_fix_mark_alive/v1' of github.com:cloudius-systems/seastar-dev:
gossip: Ignore callbacks and mark alive operation in shadow round
gossip: Ingore the duplicated mark alive operation
gossip: Fix user after free in mark_alive
In shadow round, we only interested in the peer's endpoint_state, e.g., gossip
features, host_id, tokens. No need to call the on_restart or on_join callbacks
or to go through the mark alive procedure with EchoMessage gossip message. We
will do them during normal gossip runs anyway.
After sending echo message, the Node might not be in the
endpoint_state_map anymore, use the reference of local_state
might cause user after free.
Fixes#2341
These will as of-late always be ready. Removing the future usage
in preparation for changing api signature to void(*)() (i.e. prevent
breakage on seastar update)
While blob_storage is marked as an unaligned type, the back references also
point to an unaligned type (a pointer to blob_storage), since a back
reference can live in a blob_storage. This triggers errors from zapcc/clang 4.
Fix by creating a type for the reference, which is marked as unaligned.
Message-Id: <20170502071404.507-1-avi@scylladb.com>
after 'compaction: make major compaction go through compaction manager',
the test fails because task is preempted in debug mode before it reaches
intruction to increase stat.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170501183255.6191-1-raphaelsc@scylladb.com>
During the changes in the way the housekeeping check for newer version
and warn about it in the installation the UUID part was removed but kept
in the sarounding if.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170426075724.7132-1-amnon@scylladb.com>
Some fields that belong to regular and cleanup aren't needed for
resharding_compaction, such as incremental selector (which is used
for determining max purgeable timestamp for a given decorated key)
Better move those fields to regular and make cleanup inherit from
regular compaction.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170428195611.9196-1-raphaelsc@scylladb.com>
Currently, fully expired sstable[1] is unconditionally chosen for compaction
by DTCS, but that may lead to a compaction loop under certain conditions.
Let's consider that an almost expired sstable is compacted, and it's not
deleted yet, and that the new sstable becomes expired before its ancestor is
deleted.
Because this new sstable is expired, it will be chosen by DTCS, but it will
not be purged because 'compacted undeleted' sstables are taken into account
by calculation of max purgeable timestamp and prevents expired data from
being purged. The problem is that this sequence of events can keep happening
forever as reported by issue #2260.
NOTE: This problem was easier to reproduce before improvement on compaction
of expired cells, because fully expired sstable was being converted into a
sstable full of tombstones, which is also considered fully expired.
Fixes#2260.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170428233554.13744-1-raphaelsc@scylladb.com>
"The logic responsible for converting counter updates to counter shards was
not covered by unit tests and didn't transform counter cells inside static
rows.
This series fixes the problem and makes sure that the tests cover both
static rows and transformation logic."
* tag 'pdziepak/static-counter-updates/v1' of github.com:cloudius-systems/seastar-dev:
tests/counter: test transform_counter_updates_to_shards
tests/counter: test static columns
counters: transform static rows from updates to shards
"Fixes #2326."
* 'tgrabiec/fix-range-tombstones-missing-when-slicing' of github.com:cloudius-systems/seastar-dev:
tests: mutation_source_test: Cover single-ranged queries in test_streamed_mutation_slicing_returns_only_relevant_tombstones()
tests: mutation_source_test: Add test for slicing of clustered rows
tests: mutation_reader_assertions: Log expectations
tests: mutation_reader_assertions: Add produces_eos_or_empty_mutation()
tests: sstables: Use read_row() for single-key reads
tests: sstables: Test more configutaions of sstable writer in test_sstable_conforms_to_mutation_source()
sstables: Improve logging
sstables: index_reader: Fix advance_to() to include relevant range tombstones
Attempting to create huge zones may introduce significant latency. This
patch introduces the maximum allowed zone size so that the time spent
trying to allocate and initialising zone is bounded.
Fixes#2335.
Message-Id: <20170428145916.28093-1-pdziepak@scylladb.com>
So that as_mutation_reader() will create the same kind of reader which
database::make_sstable_reader() does.
Before this change, all readers were range readers.
Test different versions of the format, and different promoted index
block sizes. The size of 1 is especially important, it will put each
fragment in a separate block, exposing various issues with promoted
index handling.
We set the scheduler wakeup granularity to 500usec, because that is the
difference in runtime we want to see from a waking task before it
preempts the running task (which will usually be Scylla). Scheduling
other processes less often is usually good for Scylla, but in this case,
one of the "other processes" is also a Scylla thread, the one we have
been using for marking ticks after we have abandoned signals.
However, there is an artifact from the Linux scheduler that causes those
preemption to be missed if the wakeup granularity is exactly twice as
small as the sched_latency. Our sched_latency is set to 1ms, which
represents the maximum time period in which we will run all runnable
tasks.
We want to keep the sched_latency at 1ms, so we will reduce the wakeup
granularity so to something slightly lower than 500usec, to make sure
that such artifact won't affect the scheduler calculations. 499.99usec
will do - according to my tests, but we will reduce it to a round
number.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170427135039.8350-1-glauber@scylladb.com>
Fix issues found by PVS-Studio as reported by Phillip Khandeliants.
Merge branch 'pvs_analyzer_errors-v1' of github.com:cloudius-systems/seastar-dev
* 'pvs_analyzer_errors-v1' of github.com:cloudius-systems/seastar-dev:
type_parser: catch exceptions by reference and not by value
token_metadata::get_host_id(ep): add a missing 'throw'
Found by PVS-Studio static analyzer:
Type slicing. An exception should be caught by reference rather than by value.
Fixes#2288
Reported-by: Phillip Khandeliants
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Caught by PVS-Studio static analyzer:
The object was created but it is not being used. The 'throw' keyword could be missing: throw runtime_error(FOO);
Reported-by: Phillip Khandeliants
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
A column family which was truncated will remove itself from compaction
manager. Any task running a compaction should be interrupted and a
task waiting to run should bail out when it wakes up.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170425224350.15965-3-raphaelsc@scylladb.com>
Introduce two time series helper tables that will simplify the querying
of traces.
One for querying regular traces:
CREATE TABLE system_traces.sessions_time_idx (
minute timestamp,
started_at timestamp,
session_id uuid,
PRIMARY KEY (minute, started_at, session_id))
and one for querying slow query records:
CREATE TABLE system_traces.node_slow_log_time_idx (
minute timestamp,
started_at timestamp,
session_id uuid,
start_time timeuuid,
node_ip inet,
shard int,
PRIMARY KEY (minute, started_at, session_id))
With these tables one may get the relevant traces like in an example below:
SELECT * from system_traces.sessions_time_idx where minutes in ('2016-09-07 16:56:00-0700') and started_at > '2016-09-07 16:56:30-0700'
Fixes#2243
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This patch makes the tracing framework follow the general idea of Google's
Dapper paper: traces generated in a context of the same query are forming
a single-rooted acyclic tree where in a ScyllaDB case vertexes are spans running
on each involved replica Node and edges are RPCs sent from one Node to another.
- Each vertex in the tree above has an ID - "span ID".
- In order to be able to build the tree from the sessions traces we need
to know the parent "span ID" - the ID of a span that sent an RPC that created
the current span.
- Each span of a tracing session is given a 64-bit random span ID.
- The root span has a span_id::illegal_id value.
This patch adds:
- The described above parent span ID and a span ID to the one_session_records
object.
- The current span ID is passed in the trace_info struct to the remote replica.
- Add parent_id and span_id columns to system_traces.events table for the parent
ID and span ID.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
After 4742008b70, _read_partial_row is
never set, and we will fail here in case the consumer will exhoust the
range. That would be the case if the end bound of the slice aligns
with the end of the index page.
Fix by assuming that if we're out of range in the middle of partition,
we sliced.
Message-Id: <1493121249-18847-1-git-send-email-tgrabiec@scylladb.com>
"This series introduces the row_tombstone class, which represents a
tombstone applied to a clustering row. It distinguishes itself from a
normal tombstone by the fact that it contains a regular tombstone and
a shadowable one, which can be erased by a row marker.
The intent of the series is thus to reify the idea of shadowable
tombstones, that up until now we considered all materialized view row
tombstones to be, leading to incorrect results."
* 'materialized-views/shadowable/v5' of https://github.com/duarten/scylla:
sstables: Read and write shadowable tombstones
mutation_partion: Use row_tombstone
mutation_partion: Introduce row_tombstone
mutation_partition: Introduce shadowable tombstones
idl-compiler: Support optional fields in views
tombstone: Extract out relational operators
row_marker: Mark constructors explicit
This patch replaces the current row tombstone representation by a
row_tombstone.
The intent of the patch is thus to reify the idea of shadowable
tombstones, that up until now we considered all materialized view row
tombstones to be.
We need to distinguish shadowable from non-shadowable row tombstones
to support scenarios such as, when inserting to a table with a
materialzied view:
1. insert into base (p, v1, v2) values (3, 1, 3) using timestamp 1
2. delete from base using timestamp 2 where p = 3
3. insert into base (p, v1) values (3, 1) using timestamp 3
These should yield a view row where v2 is definitely null, but with
the current implementation, v2 will pop back with its value v2=3@TS=1,
even though its dead in the base row. This is because the row
tombstone inserted at 2) is a shadowable one.
This patch only addresses the memory representation of such
row_tombstones.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the row_tombstone class, which represents a
tombstone made up of a regular tombstone and a shadowable one.
The rules for row_tombstones are as follows:
- The shadowable tombstone is always >= than the regular one;
- The regular tombstone works as expected;
- The shadowable tombstone doesn't erase or compact away the regular
row tombstone, nor dead cells;
- The shadowable tombstone can erase live cells, but only provided
they can be recovered (e.g., by including all cells in a MV update,
both updated cells and pre-existing ones);
- The shadowable tombstone can be erased or compacted away by a newer
row marker.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
A shadowable tombstone is a tombstone that can be replaced by a
smaller one if provided a row_marker with a bigger timestamp than the
shadowable tombstone.
In the context of a row, it is only valid as long as no newer insert
is done (thus setting a live row marker; note that if the row
timestamp set is lower than the tombstone's, then the tombstone
remains in effect as usual). If a row has a shadowable tombstone with
timestamp Ti and that row is updated with a timestamp Tj, such that
Tj > Ti (and that update sets the row marker), then the shadowable
tombstone is shadowed by that update. A concrete consequence is that
if the update has cells with timestamp lower than Ti, then those cells
are preserved (since the deletion is removed), and this is contrary to
a regular, non-shadowable row tombstone where the tombstone is
preserved and such cells are removed.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts out the relational operators in struct tombstone
to a class capable of generating them from a tri-compare function.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Some gcc versions incorrectly complain:
tests/log_histogram_test.cc:87:22: error: ‘opts1’ is not a valid template argument for type ‘const log_histogram_options&’ because object ‘opts1’ has not external linkage
size_t hist_key<node<opts1>>(const node<opts1>& n) { return n.v; }
Apparently this is a bug in gcc:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52036Fixes#2307.
Message-Id: <1493108791-11247-1-git-send-email-tgrabiec@scylladb.com>
"This series fixes some more errors found by clang, with the aim of enabling
clang/zapcc as a supported compiler. A single issue remains, but it's
probably in std::experimental::optional::swap(); not in our code."
* tag 'clang/2/v1' of https://github.com/avikivity/scylla:
sstable_test: avoid passing negative non-type template arguments to unsigned parameters
UUID: add more comparison operators
sstable_datafile_test: avoid string_view user-defined literal conversion operator
mutation_source_test: avoid template function without template keyword
cql_query_test: define static variable
cql_query_test: add braces for single-item collection initializers
storage_service: don't use typeid(temporary)
logalloc: remove unused max_occupancy_for_compaction
storage_proxy: drop overzealous use of __int128_t in recently-modified-no-read-repair logic
storage_proxy: drop unused member access from return value
storage_proxy: fix reference bound to temporary in data_read_resolver::less_compare
read_repair_decision: fix operator<<(std::ostream&, ...)
Fixes the following error in "scylla segment-descs" and a similar one in "scylla lsa-segment":
Traceback (most recent call last):
File "scylla-gdb.py", line 530, in invoke
gdb.write('0x%x: lsa free=%d region=0x%x zone=0x%x\n' % (addr, desc['_free_space'], desc['_region'], desc['_zone']))
TypeError: %x format: an integer is required, not gdb.Value
Message-Id: <1493029465-6482-1-git-send-email-tgrabiec@scylladb.com>
Every lsa-allocated object is prefixed by a header that contains information
needed to free or migrate it. This includes its size (for freeing) and
an 8-byte migrator (for migrating). Together with some flags, the overhead
is 14 bytes (16 bytes if the default alignment is used).
This patch reduces the header size to 1 byte (8 bytes if the default alignment
is used). It uses the following techniques:
- ULEB128-like encoding (actually more like ULEB64) so a live object's header
can typically be stored using 1 byte
- indirection, so that migrators can be encoded in a small index pointing
to a migrator table, rather than using an 8-byte pointer; this exploits
the fact that only a small number of types are stored in LSA
- moving the responsibility for determining an object's size to its
migrator, rather than storing it in the header; this exploits the fact
that the migrator stores type information, and object size is in fact
information about the type
The patch improves the results of memory_footprint_test as following:
Before:
- in cache: 976
- in memtable: 947
After:
mutation footprint:
- in cache: 880
- in memtable: 858
A reduction of about 10%. Further reductions are possible by reducing the
alignment of lsa objects.
logalloc_test was adjusted to free more objects, since with the lower
footprint, rounding errors (to full segments) are different and caused
false errors to be detected.
Missing: adjustments to scylla-gdb.py; will be done after we agree on the
new descriptor's format.
We both move names_ to its destination, and call names_.size() in the same
expression; this has undefined evaluation order, and fails with clang.
With this patch as well as the clang build fixes, Scylla starts and is
able to serve requests (light cassandra-stress load).
Message-Id: <20170423121727.1948-1-avi@scylladb.com>
This patch fixes a failure of virtual_reader_test, where both the test
itself and the cql_test_env initialize the messaging_service to listen
on the same address and port, triggering an assert in
posix_ap_server_socket_impl::accept().
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170423104240.21275-1-duarte@scylladb.com>
Clang warns that the expression will be evaluated (doh). While the warning
seems dubious, keep it and change the code to call the function outside
typeid(), in case it does help someone one day.
Clang's std::abs() doesn't support __int128_t, so use __int64_t instead. With
this change, it's possible that a read repair 252,700 years after a write
will be interpreted as a recent write and the read repair will incorrectly
be skipped; hopefully by that time __int128_t will be standardized.
Argument-dependent lookup requires that the operator be declared in the
same namespace as the class; move it there.
While at it, de-static it, it only causes bloat.
"Currently, a shared sstable is rewritten at all shards it belongs to, and only
after that, it's deleted.
This new algorithm adds the ability to reshard a set of sstables together at a
single shard and produce unshared sstable for all shards involved.
That's important for the leveled compaction strategy issue, in which the number
of sstables growing considerably after resharding. What happened is that every
sstable was being split into N ones, so we could end up with tons of small
sstables. Now, we will reshard together a set of adjacent sstables."
* 'sstable_resharding_revamp_v9' of github.com:raphaelsc/scylla:
tests: add test for new sstable resharding
database: kill column_family::start_rewrite
database: wire up new resharding algorithm
database: implement new sstable resharding algorithm
database: introduce function to replace new sstables by their ancestors
prevent regular compaction from choosing shared sstables
compaction_strategy: implement resharding strategy for compaction strategies
sstables: store more info in foreign_sstable_open_info
sstables: make it possible to get open info from loaded sstable
database: export column family dir
database: inform if column family has shared tables
sstables: add method to export ancestors
lcs: implement get_level_count
compaction_manager: introduce method to check if manager stopped
lcs: restore invariant instead of sending overlapping sst to L0
sstables: extend compaction for new resharding
sstables: allow shard A to correctly create sstable for shard B
compaction: rework compacting_sstable_writer to work with multiple writers
compaction: prepare compacting_sstable_writer to work with writers
sstables: rework compaction to make it easy to extend
NOTE: it's not wired yet.
Currently, a shared sstable is rewritten at all shards it belongs
to and only after that, it's deleted. With this new algorithm, a
shared sstable will be read only once and N unshared sstables
will be created, each of them with 1/N of the data. After it's
done, each owner shard will receive its new unshared sstable
replacing its ancestors.
Another benefit is that we'll no longer have resharding resulting
in number of sstables growing considerably after resharding.
A full-sized leveled sstable is usually 160MB, so after resharding,
we could have N files of 160MB/N. Now, leveled strategy will help
resharding. N adjacent sstables of same level will be resharded
together, so we'll end up with N files of N*160MB/N.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
When resharding, we're working with sstables from all shards. So let's say
we're done with resharding of sstable A that belongs to shard 0 and 1 and
sstable B that belongs to shard 1 and 2. SStables were generated for
shards 0, 1, and 2. So shards 0, 1, and 2 need to load the new sstables
and remove the ancestors. Shard 1 for example will remove sstables A and
B (ancestors) and add the new one. Then it comes this new function.
We'll forward new sstables to their target shards using foreign sstable
open info.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
For new resharding, it's important to exclude resharding sstables
from the list of candidates for regular compaction. That's doesn't
affect current resharding because it marks the sstables as
compacting. That won't work with new resharding which will work
with sstables from multiple shards.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Strategies other than leveled will reshard one shared sstable at
a time, and the target shard, shard at which job will run, for each
job will be chosen in a round-robin fashion.
For leveled strategy, we will reshard together smp::count adjacent
sstables that belong to same level.
The reason for that is because resharding one sstable at a time
may result in creation of file for each shard, meaning after
resharding we could end up with NO_SSTABLES*NO_SHARDS.
These resharding strategies will be used for our new resharding
algorithm.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
We need that info for opening a sstable at different shard, unlike
sstable loader which has everything in entry_descriptor, obtained
from components in sstable filename.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
It will be useful for resharding which will need to move a
sstable across shards, and to do that without reloading the
sstable at target shard, we need to be able to get the open
info and move it to the target shard instead.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
That's gonna be useful to quickly determine if it's worth resharding
a column family.
Reviewed-by: Nadav Har'El <nyh@scylladb.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
A large token span sstable may find its way into high level due to resharding,
which means the strategy invariant is broken. The invariant is restored by
compacting first set of overlapping sstables, meaning that the restoration
is done incrementally for multiple overlapping sets.
Invariant is restored by regular compaction after resharding puts new unshared
sstables into their original level, where level > 0.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Extends compaction for new resharding algorithm. Not wired yet.
New resharding will compact shared sstable(s) and create one
sstable for each owner. It's up to the caller to open these
new unshared sstables at their respective column families.
This new approach will save a lot of bandwidth because we'll
no longer read the entire shared sstable #smp::count times.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
That's possible by shard A explicitly saying that sstable is created
for shard B. If we don't do that, sharding metadata isn't correct,
and consequently sstable will report wrong owners.
We'll need this for resharding which will create sstables for all
shards that own the shared sstable.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
compacting_sstable_writer only allowed one writer so far, but we will need
multiple ones for resharding.
It's done by moving writer management to compaction.
finish_sstable_writer() is added for compaction impl to stop all writers,
whereas stop_sstable_writer() will only stop current writer (needed when
current sstable reaches max limit size for example).
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
No need for compacting_sstable_writer to store items that are available
in compaction class. Also, that's a step towards supporting multiple
writers for compaction.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
compact_sstables() supported both regular and cleanup compaction,
but with lots of conditions that made it ugly and hard to extend.
In the future, we want to introduce a new type of compaction for
resharding that will create one sstable for every shard owning
the sstable(s) given as input. That will be easier now.
Reviewed-by: Nadav Har'El <nyh@scylladb.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
* seastar 2eec212...194d80f (4):
> removing the collectd tests
> fix fstream metrics reporting.
> do_for_each: Make it check for need preempt
> core/sharded: introduce copy method to foreign_ptr
"Currently eviction is performed until occupancy of the whole region
drops below the 85% threshold. This may take a while if region had
high occupancy and is large. We could improve the situation by only
evicting until occupancy of the sparsest segment drops below the
threshold, as is done by this change.
I tested this using a c-s read workload in which the condition
triggers in the cache region, with 1G per shard:
lsa-timing - Reclamation cycle took 12.934 us.
lsa-timing - Reclamation cycle took 47.771 us.
lsa-timing - Reclamation cycle took 125.946 us.
lsa-timing - Reclamation cycle took 144356 us.
lsa-timing - Reclamation cycle took 655.765 us.
lsa-timing - Reclamation cycle took 693.418 us.
lsa-timing - Reclamation cycle took 509.869 us.
lsa-timing - Reclamation cycle took 1139.15 us.
The 144ms pause is when large eviction is necessary.
Statistics for reclamation pauses for a read workload over
larger-than-memory data set:
Before:
avg = 865.796362
stdev = 10253.498038
min = 93.891000
max = 264078.000000
sum = 574022.988000
samples = 663
After:
avg = 513.685650
stdev = 275.270157
min = 212.286000
max = 1089.670000
sum = 340573.586000
samples = 663
Refs #1634."
* tag 'tgrabiec/lsa-reduce-reclaim-latency-v3' of github.com:cloudius-systems/seastar-dev:
lsa: Reduce reclamation latency
tests: Add test for log_histogram
log_histogram: Allow non-power-of-two minimum values
lsa: Use regular compaction threshold in on-idle compaction
tests: row_cache_test: Induce update failure more reliably
lsa: Add getter for region's eviction function
Currently eviction is performed until occupancy of the whole region
drops below the 85% threshold. This may take a while if region had
high occupancy and is large. We could improve the situation by only
evicting until occupancy of the sparsest segment drops below the
threshold, as is done by this change.
I tested this using a c-s read workload in which the condition
triggers in the cache region, with 1G per shard:
lsa-timing - Reclamation cycle took 12.934 us.
lsa-timing - Reclamation cycle took 47.771 us.
lsa-timing - Reclamation cycle took 125.946 us.
lsa-timing - Reclamation cycle took 144356 us.
lsa-timing - Reclamation cycle took 655.765 us.
lsa-timing - Reclamation cycle took 693.418 us.
lsa-timing - Reclamation cycle took 509.869 us.
lsa-timing - Reclamation cycle took 1139.15 us.
The 144ms pause is when large eviction is necessary.
Statistics for reclamation pauses for a read workload over
larger-than-memory data set:
Before:
avg = 865.796362
stdev = 10253.498038
min = 93.891000
max = 264078.000000
sum = 574022.988000
samples = 663
After:
avg = 513.685650
stdev = 275.270157
min = 212.286000
max = 1089.670000
sum = 340573.586000
samples = 663
Refs #1634.
Message-Id: <1484730859-11969-1-git-send-email-tgrabiec@scylladb.com>
We will want to reuse the min_size mechanism for the whole compaction
threshold, including the occupancy threshold. That threshold is close
to the segment size and we cannot pick a power of two which would be
close enough to what we need.
Therefore, change log_histogram to support arbitrary minimum base.
bucket_of() was moved into log_histogram_options so that it can be used
in number_of_buckets(), which makes for a simple and much less
error-prone implementation.
Idle-time compaction should not produce not-compactible segments
becuase that means we would have to evict a lot when we finally need
to reclaim some memory, so that occupancy falls below the regular
compaction threshold. This may cause latency spikes.
Refs #1634.
After changing region evicitability condition to be less strict, cache
update stopped failing because reclamation was able to compact dense
region. Induce failure by installing evictor which refuses to evict
from cache beyond few elements.
"This series makes several optimizations to sstable mutation reader relevant
for large partitions.
Some highlights:
One optimization is to use the index for skipping across clustering restrictions.
Currently we read whole partition in such cases. That includes the case when
we need to read a static row and then jump to some clustering row in the
middle of the partition. Another case is having more than one clustering
restriction, e.g. selecting multiple single rows from the same partition.
Another optimization is using information from the index for creation of
streamed_mutation. That can save us the cost of reading the partition header
form the data file in case we would not continue reading, but skip to the
middle of that partition. Or we may not even attempt to read anything from
that partition, if after we determine the key that reader will be put behind
other readers, which will exhaust the query limit first.
Another optimization is switching single-partition queries to use the
index_reader infrastructure. Index lookups via index_reader are faster than
find_disk_ranges(). This is also a cleanup, a step towards converting all code
to use the index_reader."
* tag 'tgrabiec/optimize-sstable-reads-with-restrictions-v2' of github.com:cloudius-systems/seastar-dev: (44 commits)
sstables: Remove unused code
sstables: mutation_reader: Use index_reader::advance_to_next_partition() to skip to next partition
sstables: mutation_reader: Use index_reader for single-partition reads
sstables: mutation_reader: Add trace-level logging
sstables: mutation_reader: Move partition reading code to sstable_data_source
sstables: mutation_reader: Move definitions out of the class body
sstables: Move binary_search() to a header
database: Pass partition_range to single_key_sstable_reader to avoid copies and decorating
sstables: index_reader: Introduce advance_to_next_partition()
sstables: index_reader: Introduce advance_and_check_if_present()
sstables: index_reader: Introduce advance_past()
sstables: index_reader: Make copyable
sstables: index_reader: Optimize advancing to extreme positions
sstables: index_reader: Keep two last pages alive
dht: ring_position_view: Add key getter
dht: ring_position_view: Add constructor and factory from ring_position_view
sstables: mutation_reader: Advance to next partition using index in some cases
sstables: index_reader: Expose access to partition key and tombstone
sstables: index_reader: Introduce promoted_index_view
sstables: mutation_reader: Move _index_in_current to sstable_data_source
...
This switches single-partition query to use the index_reader
infrastructure. Index lookups via index_reader are faster than
find_disk_ranges().
perf_fast_forward, rows: 1000000, value size: 100
Before:
Testing forwarding with clustering restriction in a large partition:
pk-scan time [s] frags frag/s aio [KiB] blocked dropped idx hit idx miss idx blk cpu
no 0.002182 2 916 3 152 2 0 0 1 1 88.1%
After:
Testing forwarding with clustering restriction in a large partition:
pk-scan time [s] frags frag/s aio [KiB] blocked dropped idx hit idx miss idx blk cpu
no 0.000758 2 2639 3 152 2 0 0 1 1 48.6%
This is also a cleanup, a step towards converting all code to use the
index_reader.
There are instantiations of binary_search() used in sstables.cc, but
defined in partition.cc. The instantiations are explicitly declared in
partition.cc, but the types changed and they became obsolete. The
thing worked because partition.cc also instantiated it with the right
type. But after that code will be removed, it no longer would, and we
would get a linker error. To avoid such problems, define
binary_search() in a header.
The idea behind caching is that when we have two index readers where
one is catching up with the other, each page will be read only
once. Currently that's not always the case. There is a case when
advance_to() may need to read two pages. That's when the target
position is not found in the first page as determined by the summary
index. The second reader which catches up would have to read the first
page as well, but it would not be in cache any more. To avoid this
extra I/O let's keep a reference to the two last pages touched by the
index.
To produce a streamed_mutation for the next partition, we need to read
its key and the tombstone. Currently we always do that by consuming
the partition header from the data file. In some cases that may cause
unnecessary IO.
It's better to obtain partition information from the index if we
already have it. We can save on IO if the user will skip past the
front of partition immediately after.
It is also better to pay the cost of reading the index if we know that
we will need to use the index anyway soon. This patch predicts that by
checking if there are any clustering restrictions. If there are any,
we will almost surely need_skip() and use the index anyway.
This change also lays the ground for unification of multi and single
partiton queries without loss of performance.
sstable_data_source holds a shared state between mutation_reader and
streamed_mutation for sstables. The information whether index is in
current partition will have to be accessed by both in the following
patches.
Before the change, the following scenario was happening:
1) we try to skip based on clustering restrictions
2) we find the page and fast forward to it, recording walker's
lower bound counter
3) we read the first fragment, it's not a tombstone, so we reset the walker,
and its lower bound counter too
4) the fragment is not in range (the range starts in the middle of the page)
5) needs_skip() is true, we redo the index lookup, which wastes some CPU
This change fixes the problem by avoiding resetting the walker. We can
do that because leading tombstones are checked with a non-mutable
contains_tombstone()
Allows detecting changes of lower_bound().
Result of advance_to() is not enough. When we get false from
advance_to() twice in a row, lower bound may or may not have changed.
Simplifies the code a bit, but also will make it easier to calculate
the next position we should skip to after forwarding, taking into
consideration both the position forwarded to as well as clustering
ranges of the query. That will be just calling
_ck_ranges_walker->lower_bound() after it is trimmed.
In general mp_row_consumer has better information about the next
position to read. It could be after the position we forward to if
there are clustering restrictions. This will be exploited in the
following patches.
From now on, major compaction will go through compaction manager.
Major compaction is serialized to reduce disk space requirement.
Each column family will be running either minor and major compaction
at a given time. The only issue is number of small sstables growing
while major compaction is running, but major compaction itself will
reduce the number of tables considerably. If this turns out to be
an issue, we can allow minor to start in parallel to major, but not
the other way around.
Fixes#1156.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170417233125.14092-1-raphaelsc@scylladb.com>
Problem is that column family field of task wasn't being set for resharding,
so column family wasn't being properly removed from compaction manager.
In addition to fixing this issue, we'll also interrupt ongoing compactions
when dropping a column family, exactly like we do with shutdown.
Fixes#2291.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170418125807.7712-1-raphaelsc@scylladb.com>
streaming generates lots of small sstables with large token range,
which triggers O(N^2) in space in interval map.
level 0 sstables will now be stored in a structure that has O(N)
in space complexity and which will be included for every read.
Fixes#2287.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170417185509.6633-1-raphaelsc@scylladb.com>
"This series fixes some errors found by clang, with the aim of enabling
clang/zapcc as a supported compiler. A few more fixes are needed to produce
a binary."
* tag 'clang/1/v1' of https://github.com/avikivity/scylla:
logalloc: avoid auto in function argument declaration
thrift: avoid auto in function argument declaration
streamed_mutation: fix non-POD argument to C-style variadic function
mutation_partition_serializer: avoid auto in function argument declaration
date: use correct casts for years
streaming: avoid auto in function argument declaration
repair: avoid auto in function argument declaration
gms: expose gms::inet_address streaming operator
murmur3_partitioner: fix build on clang
i_partitioner: remove unused function
byte_ordered_partitioner: fix bad operator precedence
result_set: pass comparator by reference to std::sort()
to_string: move standard container overloads of to_string to std:: namespace
cql_type: fix bad enum syntax on clang
build: disable more warnings for clang
build: fix detection of unsupported warnings on clang
'auto' in a non-lambda function argument is not legal C++, and is hard
to read besides. Replace with the right type.
Since the right type is private, add some friendship.
Clang warns that passing a non-POD to a C-style variadic function will
result in an abort(). That happens to be exactly what we want, but to
silence the warning, use a template instead. Since templates aren't
allowed in local classes, move the containing class to namespace scope.
The standard says, and clang enforces, that declaring a function via
a friend declaration is not sufficient for ADL to kick in. Add a namespace
level declaration so ADL works.
Clang complains about some error without it, I could not understand it, but
I'm not going to argue with it.
Since std::sort() will copy the comparator, it's better to pass using an
std::ref(), and everyone is happy.
Argument-dependent lookup will not find to_string() overloads in the global
namespace if the argument and the caller are in other namespaces.
Move these to_string() overloads to std:: so ADL will find them.
Found by clang.
The diagnostic that clang spits out when it sees an unrecognized warning
is itself a warning, so the test compilation succeeds and we don't notice
the warning is not supported.
Adding -Werror turns the warning about the unrecognized warning into an
error, allowing the detection machinery to work.
The blocked task detector introduced in
113ed9e963 was seeing
the initialization phase of perf_ssttable as a blocked
task.
Tranform this part of the code in a futurized loop
to make to blocked task detector happy.
Signed-off-by: Benoît Canet <benoit@scylladb.com>
Message-Id: <20170413132506.17806-1-benoit@scylladb.com>
This patch allows the check version to support private repositories.
If a repository file is passed as a parameter, the repository id will be
passed passed as a parameter when checking version.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
"The version of create_index_statement class that was translated to C++
is pretty old by now. This series of cleanups brings it closer to Apache
Cassandra trunk to make it easier to bring over more secondary index
code to Scylla."
* 'penberg/create-index-stmt-cleanup/v1' of github.com:cloudius-systems/seastar-dev:
cql3/statements/create_index_statement: Move target validation
cql3/statements/create_index_statement: Remove static column validation
cql3/statements/create_index_statement: Extract validations
cql3/statements/create_index_statement: Kill bogus custom validation
cql3/statements/create_index_statement: Add materialized view to validate()
cql3/statements/create_index_statement: Remove validation
We take a reference of endpoint_state entry in endpoint_state_map. We
access it again after code which defers, the reference can be invalid
after the defer if someone deletes the entry during the defer.
Fix this by checking take the reference again after the defering code.
I also audited the code to remove unsafe reference to endpoint_state_map entry
as much as possible.
Fixes the following SIGSEGV:
Core was generated by `/usr/bin/scylla --log-to-syslog 1 --log-to-stdout
0 --default-log-level info --'.
Program terminated with signal SIGSEGV, Segmentation fault.
(this=<optimized out>) at /usr/include/c++/5/bits/stl_pair.h:127
127 in /usr/include/c++/5/bits/stl_pair.h
[Current thread is 1 (Thread 0x7f1448f39bc0 (LWP 107308))]
Fixes#2271
Message-Id: <529ec8ede6da884e844bc81d408b93044610afd2.1491960061.git.asias@scylladb.com>
This introduce offline installer generator.
It will generate self-extractable archive witch contains Scylla packages
and dependency packages.
Package installation automatically starts when the archive executed.
Limitation: Only supported CentOS at this point.
Fixes#2268
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1491997091-15323-1-git-send-email-syuu@scylladb.com>
"This series was initially meant to only transition the keyspace based backend to work
on top of prepared statements but there were a few potential issues found on the way.
In addition the original Tracing series has been expanded with a few patches in the cql3 layer
that are improving the generic clq3 layer but are not obvious without the context of the
following Tracing patches.
The "main" patch contains a heavy rework of trace_keyspace_helper:
- Use prepared statements for updating tables instead of manually constructing mutations:
- We intentionally decrease the level of code robustness from "paranoid" to "normal".
- The code gets a lot more simple, e.g. we don't need to cache columns definitions any more.
- We are loosing some performance here but:
- Tracing write is not in the fast path.
- Tracing write events should be rare.
- Currently the performance loss (for the actual write time of all trace records) for a "SELECT" query with a specific key is
about 45%: 144us vs 99us."
* 'tracing_rework_using_prepared-v6' of github.com:cloudius-systems/seastar-dev:
tracing: use prepared statment for updating tables
tracing::trace_keyspace_helper: add a bad_column_family constructor that accepts an std::exception parameter
tracing::trace_keyspace_helper: introduce a table_helper class
tracing::trace_keyspace_helper: add static qualifier to make_monotonic_UUID_tp() and elapsed_to_micros() methods
tracing::tracing: allow slow query TTL only in the signed 32-bit integer range
cql3::query_processor::prepare(): futurize the error case
cql3::query_options: add a factory method for creation of options for a BATCH statement
cql3::statements::batch_statement: add a constructor that doesn't receive the "bound_terms" value
cql3::query_processor: use weak_ptr for passing the prepared statements around
When compacting a fully expired sstable, we're not allowing that sstable
to be purged because expired cell is *unconditionally* converted into a
dead cell. Why not check if the expired cell can be purged instead using
gc before and max purgeable timestamp?
Currently, we need two compactions to get rid of a fully expired sstable
which cells could have always been purged.
look at this sstable with expired cell:
{
"partition" : {
"key" : [ "2" ],
"position" : 0
},
"rows" : [
{
"type" : "row",
"position" : 120,
"liveness_info" : { "tstamp" : "2017-04-09T17:07:12.702597Z",
"ttl" : 20, "expires_at" : "2017-04-09T17:07:32Z", "expired" : true },
"cells" : [
{ "name" : "country", "value" : "1" },
]
now this sstable data after first compaction:
[shard 0] compaction - Compacted 1 sstables to [...]. 120 bytes to 79
(~65% of original) in 229ms = 0.000328997MB/s.
{
...
"rows" : [
{
"type" : "row",
"position" : 79,
"cells" : [
{ "name" : "country", "deletion_info" :
{ "local_delete_time" : "2017-04-09T17:07:12Z" },
"tstamp" : "2017-04-09T17:07:12.702597Z"
},
]
now another compaction will actually get rid of data:
compaction - Compacted 1 sstables to []. 79 bytes to 0 (~0% of original)
in 1ms = 0MB/s. ~2 total partitions merged to 0
NOTE:
It's a waste of time to wait for second compaction because the expired
cell could have been purged at first compaction because it satisfied
gc_before and max purgeable timestamp.
Fixes#2249, #2253
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170413001049.9663-1-raphaelsc@scylladb.com>
In addition to actually moving to using the prepared statements the changes also include:
- Kill the cache_xxx() methods - the schema is going to be checked during the
prepared statement creation and during its execution.
- Move the caching of table ID and the prepared statement to the get_schema_ptr_or_create().
- Rename: get_schema_ptr_or_create() -> cache_table_info().
After these changes we are less strict in our demands to system_traces tables schemas, e.g.
if some column's type is not exactly as we expect but rather only "compatible" in the CQL sense
we will tolerate this and will continue to write into that table.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
An object built with this constructor will use the what() message from the
given exception in the final error message.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This class contains a general table info and implements standard operations on this table:
- Creation.
- Info caching.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Extract specific validations to separate functions to preserve the same
structure as Apache Cassandra code and make it easier to add support for
multiple index targets.
Rejecting custom indices is bogus because it's just a configuration
mechanism like replication strategy, for example. Furthermore, it's
needed for SASI indices, which we likely need to be compatible with.
The validation was removed in Apacha Cassandra commit 0626be8 ("New 2i
API and implementations for built in indexes"). Let's also remove it
from our code so that we remove one dependency to the obsolete db/index/
code.
Any TTL is eventually converted into the gc_clock::duration value, which
is based on int32_t type. Limit the node_slow_log TTL user configurable value
to the same values range for consistency.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Make sure that errors are reported in a form of an exceptional future and
not by a direct exception throwing.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
This constructor should be used when we know that there are no bound terms in the current
batch statement.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Use seastar::checked_ptr<weak_ptr<pepared_statement>> instead of shared_ptr for passing prepared statements around.
This allows an easy tracking and handling of statements invalidation.
This implementation will throw an exception every time an invalidated
statement reference is dereferenced.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
segment_zone::migrate_all_segments() was trying to migrate all segments
inside a zone to the other one hoping that the original one could be
completely freed. This was an attempt to optimise for throughput.
However, this may unnecesairly hurt latency if the zone is large, but
only few segments are required to satisfy reclaimer's demands.
Message-Id: <20170410171912.26821-1-pdziepak@scylladb.com>
While cpuset.conf is supposed to be set before this is used, not having
a cpuset.conf at all is a valid configuration. The current code will
raise an exception in this case, but it shouldn't.
Also, as noted by Amos, atoi() is not available as a global symbol. Most
invocations were safe, calling string.atoi(), but one of them wasn't.
This patch replaces all usages of atoi() with int(), which is more
portable anyway.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170407172937.17562-1-glauber@scylladb.com>
We have recently fixed the ami init scripts to mark i3 as a supported
instance. However, the code to detect whether or not the instance is
supported is duplicated, and called from multiple locations. That means
that when the user logs in, it will see the instance as not supported -
as the test is coming from a different source.
This patch moves it to the scylla_lib.sh utilities script, so we can
share it, and make sure it is right for all locations.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170406201137.8921-1-glauber@scylladb.com>
Older versions of packer do not support ENA, and when faced
with the option
"enhanced_networking": true,
will only actually enable it for the older 82599 VF instances.
Fortunately, packer 1.0 already supports it, and all we have to do is
update it.
While we are at it, let's check if the file is legit before using a
random file we have downloaded from the internet, to avoid breaching our
building process.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170406123952.14708-1-glauber@scylladb.com>
For the debian package, the files don't have to be listed individually,
but for RPMs they do.
The rpm builds are currently failing with:
error: Installed (but unpackaged) file(s) found:
/usr/lib/scylla/scylla_util.py
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <20170406012336.21081-1-glauber@scylladb.com>
Following C* API there are two APIs for getting the load from
storage_service:
/storage_service/metrics/load
/storage_service/load
This patch adds the implementation for
/storage_service/metrics/load
The alternative, is to drop on of the API and modify the JMX
implementation to use the same API.
Fixes#2245
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170401181520.19506-1-amnon@scylladb.com>
Currently we return proceed::no after every mutation_fragment which is
to be consumed. This froces parser to save and reload its state
often. This can be avoided if we pushed the fragments directly from
mp_row_consumer, then we would return proceed::no only when the buffer
fills up.
tests/perf/perf_fast_forward shows 15% increase in throughput of a large partition scan,
from 1.34M frag/s to 1.55M frag/s.
Message-Id: <1490882700-22684-1-git-send-email-tgrabiec@scylladb.com>
rpmbuild tries to compile any *.py by default, but it causes compilation
error on python3 code when python2 is system default (CentOS, RHEL).
So skip compiling it, drop .pyc / .pyo from the package.
Fixes#2235
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1490859768-18900-1-git-send-email-syuu@scylladb.com>
"Also, this new version supports i3.
Number of requests for i3 is obtained similarly as to i2: I have run tests
for a single disk, and then we'll take the amount of disks into account. Other
possible limits are also taken into account, like the max per-shard seastar limit
of 128 in-flight request, and the per-disk limit obtained by sysfs."
* 'python-io-setup' of https://github.com/glommer/scylla:
rewrite scylla_io_setup in python
scripts: add python module with common utilities
As we convert more stuff to python, we'll have more opportunities for
sharing code between them. We already do that for the bash scripts with
a file "scylla_lib.sh". We'll do the same for python.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We recently introduced fstrim cronjob / systemd timer unit, but some of
distributions already has their own fstrim cronjob / systemd timer unit.
So let's use them when it's posssible.
Fixes#2233
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1491226727-10507-1-git-send-email-syuu@scylladb.com>
After RAID devices mounted to /var/lib/scylla, scylla-server doesn't able to
find ./conf/ directory since we haven't created symlinks on the volume.
Instead of creating symlink on scylla_raid_setup, let's specify scylla.yaml
path on program argument.
Fixes#2236
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1491056110-1078-1-git-send-email-syuu@scylladb.com>
Running nodetool status on each login to Scylla AMI helps in three ways:
- give the user a quick view of the node and cluster status beyond the current "scylla is active"
- hint to the user about the nodetool and how to use it
- move the first, slow, run of nodetool to the login phase, making the second interactive run much faster
on the down side, it does slow the login in a few sec
Signed-off-by: Tzach Livyatan <tzach@scylladb.com>
[ penberg: fix formatting ]
Message-Id: <20170330081711.22038-1-tzach@scylladb.com>
This script validates schemas of distributed system keyspaces:
system_traces and system_auth.
It tries to add the missing columns and checks that the existing columns
have expected types.
In case of any problem the corresponding info message is printed and non-zero exit
status is returned.
Extra columns are ignored.
The validation function may also be called from another Python program.
It returns True in case of success and False otherwise.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1490195645-29905-1-git-send-email-vladz@scylladb.com>
"sstable_streamed_mutation::fast_forward_to() is changed to use promoted index
(via index_reader) to optimize skipping in large partitions.
In addition to that, sstable mutation_reader is changed to use the index
to skip to the next partition.
Performance impact was evaluated using newly added tests/perf/perf_fast_forward
What's beyond this series:
- Using index_reader for single-partition reads as well
- Using index_reader for skipping across ranges in clustering restrictions"
* tag 'tgrabiec/skip-within-partition-using-index-v2' of github.com:cloudius-systems/seastar-dev: (47 commits)
tests: Add performance test for fast forwarding of sstable readers
tests: Allow starting cql_test_env on pre-existing data
config: Allow specifying source when setting value
tests: sstable: Add test for fast forwarding within partition using index
sstables: sstable_streamed_mutation: use index in fast_forward_to()
sstables: Store parsed promoted index in index_entry
sstables: Add trace-level logging for sstable consumption
sstables: Define deletion_time earlier
sstables: Make parsing throw exception on malformed promoted index
tests: Add tests for ordering of position_in_partition relative to composites
position_range: Introduce all_clustered_rows() factory method
position_in_partition: Introduce for_key()/after_key() factory methods
position_in_partition: Add factory methods for positions around all rows
position_in_partition: Introduce for_range_start()/for_range_end()
position_in_partition: Fix friendship declaration
keys: Introduce is_empty() for prefixes
position_in_partition: Make comparable with composites
types: Enhance lexicographical comparators
compound_compat: Accept marker value in serialize_value()
compound_compat: Add trichotomic comparator
...
quick introduction to level starvation:
high levels may be left uncompacted (thus starved) for a long time if user
makes something that make they contain little data, such as cleanup or change
of max sstable size (default 160M). Leveled strategy handles this problem as
follow: consider we're compacting L1 to L2. If L3 is starved, we look for one
of its sstable that is fully contained in token range of candidates L1->L2,
so that we won't end up with an overlapping in L2.
now the problem:
the functionality isn't working properly now because range of candidates is
being incorrectly calculated due to an accident when converting the code to
C++. It won't cause an overlap because it's actually being more restrictive
about which sstable from starved level can be used.
A test case was added to confirm the problem.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170328223753.15398-1-raphaelsc@scylladb.com>
So that is_set() will be true for that option. Needed in tests which
set some config options in higher layer and then lower layers detects
if option was set or not before applying its default.
Will be easier to propagate failure to upper layers once parsing is
reused in the index_reader.
The old behavior of ignoring parsing failures is preserved, but the
error is logged now.
They now accept optional lexicographical_relation which can be used
to alter position of the element relative to elements prefixed by it.
Example. Let's consider lexicographical ordering on strings. The
position of "bc" in a sample sequence is affected by
lexicographical_realtion as follows:
aa
aaa
b
ba
--> before_all_prefixed
bc
--> before_all_strictly_prefixed
bca
bcd
--> after_all_prefixed
bd
bda
c
ca
The iterator doesn't really allow modifying unserlying component.
This change enables using the iterator with boost::make_iterator_range() and
boost::range::join(), which get utterly confused otherwise.
Direct motivation for this is to be able to use two index readers from
a single mutation reader, one for lower bound of the range and one for
the upper bound of the range, without sacrificing optimization of
avoiding index reads when forwarding to partition ranges which are
close by. After the change, all index readers of given sstable will
share index buffers, so lower bound reader can reuse the page read by
the upper bound reader.
The reason for using two readers will be so that we are able to skip
inside the partition range, not only outside of it. This is not
possible if we use the same index reader to locate the upper bound of
the range, because we may only advance the cursor.
Needed to make the index_list copyable, which is going to be needed to
implement legacy get_index_entries() which returns by value, after
index sharing is implemented.
Index reader already can be queried only with monotonic positions, so
the concept of a cursor is ingrained. Making it explicit will make it easier
to define behavior for forwarding withing the partition.
After the change:
- lower_bound() is renamed to advance_to() and doesn't return
the position, only advances the cursor
- data file position for partition under cursor can be obtained
at any time with data_file_position()
The previous fix removed the additional insertion of "min rp" per source
shard based on whether we had processed existing CF:s or not (i.e. if
a CF does not exist as sstable at all, we must tag it as zero-rp, and
make whole shard for it start at same zero.
This is bad in itself, because it can cause data loss. It does not cause
crashing however. But it did uncover another, old old lingering bug,
namely the commitlog reader initiating its stream wrongly when reading
from an actual offset (i.e. not processing the whole file).
We opened the file stream from the file offset, then tried
to read the file header and magic number from there -> boom, error.
Also, rp-to-file mapping was potentially suboptimal due to using
bucket iterator instead of actual range.
I.e. three fixes:
* Reinstate min position guarding for unencoutered CF:s
* Fix stream creating in CL reader
* Fix segment map iterator use.
v2:
* Fix typo
Message-Id: <1490611637-12220-1-git-send-email-calle@scylladb.com>
This patch ensures we generate UUIDs using the same randomness source
as all the other values we randomly generator, so that we can get a
deterministic run from the seeds we print.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170327161149.8938-2-duarte@scylladb.com>
* seastar 6b21197...2ebe842 (6):
> Merge "Various improvements to execution stages" from Paweł
> app-template: allow apps to specify a name for help message
> bool_class: avoid initializing object of incomplete type
> app-template: make sure we can still get help with required options
> prometheus: Http handler that returns prometheus 0.4 protobuf or text format
> Update DPDK to 17.02
Includes patch from Pawel to adjust to updated execution_stage interface.
Users sometimes need to run their own yaml configuration files, and it
is currently annoying to deploy modified files on docker.
One possible solution is to bind mount the file into the docker
container using the -v switch, just like we already do for for the data
volume.
The problem with the aforementioned approach is that we have to change
the yaml file to insert the addresses, and that will change the file in
the host (or fail to happen, if we bind mount it read-only).
The solution I am proposing is to avoid touching the yaml file inside
the container altogether. Instead, we can deploy the address-related
arguments that we currently write to the yaml file as Scylla options.
Fixes#2113
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1490195141-19940-1-git-send-email-glauber@scylladb.com>
If early startup fails in docker-entrypoint.py, the container does not
start. It's therefore not very helpful to log to a file _within_ the
container...
Message-Id: <1490275943-23590-1-git-send-email-penberg@scylladb.com>
Fixes#2173
Per-shard min positions can be unset if we never collected any
sstable/truncation info for it, yet replay segments of that id.
Wrap the lookups to handle "missing data -> default", which should have been
there in the first place.
Message-Id: <1490185101-12482-1-git-send-email-calle@scylladb.com>
Don't report a Tracing session ID unless the current query had a Tracing bit in its
flags.
Although the current master's behaviour is legal it's suboptimal and some Clients are sensitive to that.
Let's fix that.
Fixes#2179
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1490063752-8915-1-git-send-email-vladz@scylladb.com>
column_family constructor uses delegation, as such, only the actual
constructor implementation should contain a call to register the
metrics.
Current implementation ends up with re registration of the metrics.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170320140817.22214-1-amnon@scylladb.com>
This patch exposes Scylla's Prometheus port by default. You can now use
the Scylla Monitoring project with the Docker image:
https://github.com/scylladb/scylla-grafana-monitoring
To configure the IP addresses, use the 'docker inspect' command to
determine Scylla's IP address (assuming your running container is called
'some-scylla'):
docker inspect --format='{{ .NetworkSettings.IPAddress }}' some-scylla
and then use that IP address in the prometheus/scylla_servers.yml
configuration file.
Fixes#1827
Message-Id: <1490008357-19627-1-git-send-email-penberg@scylladb.com>
On Ubuntu 14.04, the lsblk doesn't have '-p' option, but
`scylla_setup` try to get block list by `lsblk -pnr` and
trigger error.
Current simple pattern will match all help content, it might
match wrong options.
scylla-test@amos-ubuntu-1404:~$ lsblk --help | grep -e -p
-m, --perms output info about permissions
-P, --pairs use key="value" output format
Let's use strict pattern to only match option at the head. Example:
scylla-test@amos-ubuntu-1404:~$ lsblk --help | grep -e '^\s*-D'
-D, --discard print discard capabilities
Signed-off-by: Amos Kong <amos@scylladb.com>
Message-Id: <4f0f318353a43664e27da8a66855f5831457f061.1489712867.git.amos@scylladb.com>
We're cleaning up sstables in parallel. That means cleanup may need
almost twice the disk space used by all sstables being cleaned up,
if almost all sstables need cleanup and every one will discard an
insignificant portion of its whole data.
Given that cleanup is frequently issued when node is running out of
disk space, we should serialize cleanups in every shard to decrease
the disk space requirement.
Fixes#192.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20170317022911.10306-1-raphaelsc@scylladb.com>
The mapping between a base table update and a view update is schema
dependent, so we need to ensure the view schema versions match the
base schema version. For example, we match base columns to view
columns by name, so we need to ensure the base and view schemas we're
using for writting are isolated with respect to a previous alter
table statement.
We thus need to match base schema versions with view schema versions,
and we need to so atomically to ensure that when one fiber sees a
schema, it also sees the complete set of corresponding view schemas.
This series ensures the schemas modified as a result of an alter
table statement are published atomically, under the schema lock. This
way, all the schemas referenced by the database are consistent with
each other when they are observed by other fibers.
Finally, we upgrade the mutation schema before generating the view
updates, to ensure it matches the most recent view schemas the base
replica knows about, registered in the database.
The db::view::view class was replaced by a set of non-member
functions, with its state, which used to reflect only the most recent
schema version, being moved to a new view_info class.
One of the goals of can_allocate_more_memory() is to prevent depleting
seastar's free memory close to its minimum, leaving a head room above
that minimum so that standard allocations will not cause reclamation
immediately. Currently the function doesn't take into accoutn actual
threshold used by the seastar allocator, so there could be no gap or
even could go below the minimum.
Fix that by ensuring there's always a gap above min_free_memory().
min_gap was reduced to 1 MiB so that low memory setups are not
impacted significantly by the change.
Message-Id: <1489667863-15099-1-git-send-email-tgrabiec@scylladb.com>
* seastar 4d25b85...6b21197 (3):
> core: memory: Expose control of the free memory low water mark
> scripts: add perftune.py
> tutorial: make network examples work on multi-core
"The test allocates objects in batches (allocation is always under a reclaim
lock) of ~3MiB and assumes that it will always succeed because if we cross the
low water mark for free memory (20MiB) in seastar, reclamation will be
performed between the batches, asynchronously.
Unfortunately that's prevented by can_allocate_more_memory(), which fails
segment allocation when we're below the low water mark. LSA currently doesn't
allow allocating below the low water mark.
The solution which is employed across the code base is to use allocating_section,
so use it here as well.
Exposed by recent consistent failures on branch-1.7."
* 'tgrabiec/fix-lsa-async-eviction-test' of github.com:cloudius-systems/seastar-dev:
tests: lsa_async_eviction_test: Allocate objects under allocating section
lsa: Allow adjusting reserves in allocating_section
This patch ensures we upgrade the mutation to the current schema when
generating and pushing view updates, so that the it matches the most
up to date views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures that the schema merging atomically publishes
schema changes. In particular, it ensures that when a base schema
and a subset of its views are modified together (i.e., upon an alter
table or alter type statement), then they are published together as
well, without any deferring in-between.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the migration path for table updates such that the
base table mutations are sent and applied atomically with the view
schema mutations.
This ensures that after schema merging, we have a consistent mapping
of base table versions to view table versions, which will be used in
later patches.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The write path uses a base schema at a particular version, and we
want it to use the materialized views at the corresponding version.
To achieve this, we need to map the state currently in db::view::view
to a particular schema version, which this patch does by introducing
the view_info class to hold the state previously in db::view::view,
and by having a view schema directly point to it.
The changes in the patch are thus:
1) Introduce view_info to hold the extra view state;
2) Point to the view_info from the schema;
3) Make the functions in the now stateless db::view::view non-member;
4) Remove the db::view::view class.
All changes are structural and don't affect current behavior.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
In preparation for upcoming patches, which will deal with
moving the state in db::view::view to view_info.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"The current implementations of collection_type_impl::is_empty() and
collection_type_impl::difference() don't handle tombstoned collection
mutations correctly. In particular:
- is_empty() considers a collection mutation with a tombstone (and no
entries) as empty;
- difference() doesn't do set difference between the cells tombstones,
and always returns the highests.
Fixes#2152"
* 'collection-diff/v4' of github.com:duarten/scylla:
mutation_test: Add more test cases for difference()
mutation_source_test: Randomly generate collection cells
collection_type_impl: Use set difference for tombstones
collection_type_impl: A mutation with a tombstone is not empty
This patch fixes collection_type_impl::difference() so it does set
difference for tombstones instead of just returning the larger
one, as difference() is supposed to return only the information in
mutation A that supersedes that in B, given difference(A, B).
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the collection_type_impl::is_empty() function so
that it doesn't consider empty a collection_mutation which has a
tombstone.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Discarding blocks on large RAID volume takes too much time, user may suspects
the script doesn't works correctly, so it's better to skip, do discard directly on each volume instead.
Fixes#1896
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1489533460-30127-1-git-send-email-syuu@scylladb.com>
Fixes#2098
Replay previously did all segments in parallel on shard 0, which
caused heavy memory load. To reduce this and spread footprint
across shards, instead do X segments per shard, sequential per shard.
v2:
* Fixed whitespace errors
Message-Id: <1489503382-830-1-git-send-email-calle@scylladb.com>
Metrics name should be unique per type.
requests_blocked_memory was registered twice, one as a gauge and one as
derived.
This is not allowed.
Fixes#2165
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170314162826.25521-1-amnon@scylladb.com>
* seastar fd29fd0...4d25b85 (2):
> core/file: fix EOF detection for file with custom impl
> tutorial: fix echo server example
Includes patch from Raphael updating checked_file_impl:
"Now file_impl requires dma_read_bulk to be implemented, and for
checked_file_impl, it only's about calling dma_read_bulk from
the posix file it wraps."
Metrics should have their unique name. This patch changes
throttled_writes of the queu lenght to current_throttled_writes.
Without it, metrics will be reported twice under the same name, which
may cause errors in the prometheus server.
This could be related to scylladb/seastar#250
Fixes#2163.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <20170314081456.6392-1-amnon@scylladb.com>
* seastar 84a0b70...fd29fd0 (4):
> Fix smp::submit_to() with function reference
> execution_stage: add concept restraint for operator()
> core/temporary_buffer: Add operator==()
> map_reduce: allow reducer to take accumulated value by rref
"This replaces use of a generic forwarding wrapper in sstable reader with
specialized implentation. Forwarding doesn't yet utilize indexes in this
series, only integrates it with mp_row_consumer, which is a prerequisite.
It's still an optimization, since mp_row_consumer will not try to consume
past the range as it used to.
Sending early for easier consumption."
* tag 'tgrabiec/forwarding-of-mp-row-consumer-v2' of github.com:scylladb/seastar-dev:
sstables: Remove use of forwarding wrapper
sstables: Implement sstable_streamed_mutation::fast_forward_to()
sstables: Extract and use clustering_ranges_walker
tests: sstables: Add test for handling of repeated tombstones
sstables: Extract writer parameters into config objects
tests: Move as_mutation_source() helper to header
tests: Extract ensure_monotonic_positions() to streamed_mutation_assertions
streamed_mutation: Add streamed_mutation_returning() helper
tests: mutation_source_test: Add test case for forwarding to a full range
tests: simple_schema: Add fragment factories
tests: Extract simple_schema
sstables: Move workaround for out-of-order range tombstones to mp_row_consumer
sstables: Drop default mp_row_consumer constructor
sstables: Swap order of values in "proceed" so that "no" is assigned 0
util/optimized_optional: Make printable
position_in_partition: Add is_static_row() in the view
range_tombstone_stream: Add reset()
range_tombstone_stream: Add get_next(position_in_partition_view)
sstables: streamed_mutation: Stop reading when end of slice reached
sstables: Switch is_in_range() to position_in_partition
Handling of forwarding is done inside mp_row_consumer, because it
allows us to filter out irrelevant data sooner and thus more
efficiently.
Becuase static row can be now skipped as well, _skip_clustering_row
was renamed to more generic _skip_in_progress.
This is a preliminary step before adding support for fast-forwarding
to mp_row_consumer, so that range handling can be solely in
mp_row_consumer rather than split between it and
sstable_streamed_mutation.
This also alleviates #2080 by reading all tombstones only up to the
first row, after that range tombstones are treated like other
fragments.
As part of this change, skip detection detection is refactored. This
simplifies reasoning about mp_row_consumer's state a bit because now
is_mutation() is not reset externally and only depends on current
position of the reader.
It will prove useful when we extend mutation reader to decide if it
should skip to the next partition up front before calling
_context.read(), so that we can for instance skip using index instead.
Fixes#2088.
This patch changes a lambda argument type so the keyspace name is
passed by reference instead of copying it, in
read_schema_for_keyspaces().
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170309213134.10331-1-duarte@scylladb.com>
Change linkage of segment_descriptor_hist_options to external to keep
good old GCC5 happy, despite C++11 allowing static linkage of non-type
template arguments.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170309213206.10383-1-duarte@scylladb.com>
Add batch_size_fail_threshold_in_kb to prevent huge batch from been
applied and causing troubles. Also do not warn or fail if only one
partition is affected.
Fixes: #2128
Message-Id: <20170309111247.GE8197@scylladb.com>
"These patches introduce execution stages to Scylla in order to improve
icache friendliness. The places were stages are added are not chosen
very carefully and rather introduced at between different subsystems:
cql, storage proxy and database. This already results in a rather
significant improvement and can be tuned later if necessary.
Performance results:
perf_simple_query -c4 --duration 60
(medians)
before after diff
write 83017.75 242876.04 +192.6%
read 61709.16 168258.26 +172.7%
The real life improvements aren't as good because it is much harder
to collect sufficiently high number of operations in a batch."
Additional benchmarking from Paweł:
"I did some tests on my local setup.
* Latency at light loads
Scylla running on 16 logical CPUs (8 cores) with 64 GB of RAM.
cassandra-stress -rate threads=32
write latency
master seda
median 1.2 0.6
95th 1.6 0.8
99th 1.7 0.9
99.9th 2.5 1.3
max 26.4 24.2
Flags '--poll-mode' and '--defragment-memory-on-idle false' didn't improve situation for master.
See also attached graph write_99.svg and write_999.svg.
read latency
master seda
median 0.8 0.6
95th 1.0 0.9
99th 1.1 1.0
99.9th 1.4 1.2
max 18.5 18.0
See also attached graph read_99.svg and read_999.svg.
* Server 100% loaded, dataset fitting in memory (throughput)
Scylla running on 2 cores with 64 GB of RAM.
4x scylla-bench with the uniform workload
(concurrency of each s-b: 512 for writes, 256 for reads).
There were no cache misses during reads.
master seda diff
writes 107722.4 168482.26 +56.4%
reads 51049.48 76158.19 +49.2%
* Server 100% loaded, writes being flushed and compacted (throughput)
Scylla running on 2 cores with 4 GB of RAM.
4x scylla-bench with the uniform workload, concurrency 256 each.
master seda diff
writes 79575.77 114206.11 +43.5%
See attached graph: writes_with_flushes_and_compaction.png (first run: master, second: seda)."
* tag 'pdziepak/scylla-execution-stages/v1-rebased' of github.com:cloudius-systems/seastar-dev:
transport: make process_request_one() an execution stage
mutation_query: add an execution stage
db: make database::query() an execution stage
db: make apply an execution stage
storage_proxy: make mutate() an execution stage
cql3: make batch statement an execution stage
cql3: make modification statement an execution stage
cql3: make select statement an execution stage
mutation_reader: make mutation_source nothrow movable
There is no need to wait when starting the prometheus server. As it is
up to each of the modules to register its metrics when it is ready.
This is especially important when debuging boot issues.
This patch moves the prometheus initilization to be done at an early
stage of the boot sequencec.
Fixes#2144
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1489041986-28974-1-git-send-email-amnon@scylladb.com>
We have:
auto halves = range.split(midpoint, dht::token_comparator());
We saw a case where midpoint == range.start, as a result, range.split
will assert becasue the range.start is marked non-inclusive, so the
midpoint doesn't appear to be contain()ed in the range - hence the
assertion failure.
Fixes#2148
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Signed-off-by: Asias He <asias@scylladb.com>
Message-Id: <93af2697637c28fbca261ddfb8375a790824df65.1489023933.git.asias@scylladb.com>
* seastar 5861f99...84a0b70 (13):
> build: don't error out on [[deprecated]] APIs
> Merge "Introduce execution stages" from Paweł
> Remove unused include statement
> http: catch and count errors in read and respond
> Merge "Adding metrics configuration" from Amnon
> future: add concepts for map_reduce(), when_all_succeed()
> doxygen: exclude c-ares directory
> scripts/posix_net_conf.sh: add --use-cpu-mask option
> file: take flush into account when calculating size for truncate in optimize_queue()
> Fixing the prometheus cleanup patch
> Merge "posix_net_conf.sh: better distribute ingress processing" from Vlad
> prometheus: code clean up
> future: relax finally() constraints even more
"If a node is bootstrapped with auto_boostrap disabled, it will not
wait for schema sync before creating global keyspaces for auth and
tracing. When such schema changes are then reconciled with schema on
other nodes, they may overwrite changes made by the user before the
node was started, because they will have higher timestamp.
To prevent that, let's use minimum timestamp so that default schema
always looses with manual modifications. This is what Cassandra does.
Fixes #2129."
* tag 'tgrabiec/prevent-keyspace-metadata-loss-v1' of github.com:scylladb/seastar-dev:
db: Create default auth and tracing keyspaces using lowest timestamp
migration_manager: Append actual keyspace mutations with schema notifications
The skip() implementation for the compressed file input stream incorrectly
handled the case of skipping to the end of file: In that case we just need
to update the file pointer, but not skip anywhere in the compressed disk
file; In particular, we must NOT call locate() to find the relevant on-disk
compressed chunk, because there is none - locate() can only be called on
actual positions of bytes, not on the one-past-end-of-file position.
Fixes#2143
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170308100057.23316-1-nyh@scylladb.com>
If the node is bootstrapped with auto_boostrap disabled, it will not
wait for schema sync before creating global keyspaces for auth and
tracing. When such schema changes are then reconciled with schema on
other nodes, they may overwrite changes made by the user before the
node was started, because they will have higher timestamp.
To prevent that, let's use minimum timestamp so that default schema
always looses with manual modifications. This is what Cassandra does.
Fixes#2129.
There is a workaround for notification race, which attaches keyspace
mutations to other schema changes in case the target node missed the
keyspace creation. Currently that generated keyspace mutations on the
spot instead of using the ones stored in schema tables. Those
mutations would have current timestamp, as if the keyspace has been
just modified. This is problematic because this may generate an
overwrite of keyspace parameters with newer timestamp but with stale
values, if the node is not up to date with keyspace metadata.
That's especially the case when booting up a node without enabling
auto_bootstrap. In such case the node will not wait for schema sync
before creating auth tables. Such table creation will attach
potentially out of date mutations for keyspace metadata, which may
overwrite changes made to keyspace paramteters made earlier in the
cluster.
Refs #2129.
The current code is assymetric: the first N-1 shards to delete a set receive
a synthetic future to wait on, while the last deletion receives the result
of the delete operation (which also broadcasts completion to the first N-1
operations. This results, in case of an error, with the Nth future being
reported as an unhandled error.
Fix by making everything symmetric: all N callers receive a synthetic
future. Nobody waits for the deletion operation (which still broadcasts its
completion to all waiters, so errors are not lost).
Message-Id: <20170305151607.14264-1-avi@scylladb.com>
"This series adds various optimisations to counter implementation
(nothing extreme, mostly just avoiding unnecessary operations) as well
as some missing features such as tracing and dropping timed out queries.
Performance was tested using:
perf-simple-query -c4 --counters --duration 60
The following results are medians.
before after diff
write 18640.41 33156.81 +77.9%
read 58002.32 62733.93 +8.2%"
* tag 'pdziepak/optimise-counters/v3' of github.com:cloudius-systems/seastar-dev: (30 commits)
cell_locker: add metrics for lock acquisition
storage_proxy: count counter updates for which the node was a leader
storage_proxy: use counter-specific timeout for writes
storage_proxy: transform counter timeouts to mutation_write_timeout_exception
db: avoid allocations in do_apply_counter_update()
tests/counters: add test for apply reversability
counters: attempt to apply in place
atomic_cell: add COUNTER_IN_PLACE_REVERT flag
counters: add equality operators
counters: implement decrement operators for shard_iterator
counters: allow using both views and mutable_views
atomic_cell: introduce atomic_cell_mutable_view
managed_bytes: add cast to mutable_view
bytes: add bytes_mutable_view
utils: introduce mutable_view
db: add more tracing events for counter writes
db: propagate tracing state for counter writes
tests/cell_locker: add test for timing out lock acquisition
counter_cell_locker: allow setting timeouts
db: propagate timeout for counter writes
...
logalloc::reclaim_lock prevents reclaim from running which may cause
regular allocation to fail although there is enough of free memory.
To solve that there is an allocation_section which acquire reclaim_lock
and if allocation fails it run reclaimer outside of a lock and retries
the allocation. The patch make use of allocation_section instead of
direct use of reclaim_lock in memtable code.
Fixes#2138.
Message-Id: <20170306160050.GC5902@scylladb.com>
"Work on this series started with fixing the 'nodetool clearsnapshot'.
The current master code ignores the snapshots in deleted keyspaces (issue #2045).
I noticed that in many places our code has to build the path to some directory/file
it simply had the sstring(<path1>) + "/" + sstring(<path2>) constructs which may cause us issues
if somebody decides to complile/run scylla on not-Unix-based OS, like Microsoft Windows.
I understand that this is a long shot but if we can make it right now - why not to.
The answer is boost::filesystem::path class - its synchronous parts, of course.
I decided to take an initiative and fix the issues above and then use the fixed code for
fixing the issue #2045:
- Fix some minor issues in the existing code.
- Extend the lister class and move it into the separate files outside database.cc.
On the way I've found an issue in the existing code (issue #2071).
This series fixes this one too (PATCH2)."
The current test for whether or not the filesystem is mounted is weak
and will fail if multiple pieces of the hierarchy are mounted.
util-linux ships with a mountpoint command that does exactly that,
so we'll use that instead.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1488742801-4907-1-git-send-email-glauber@scylladb.com>
boost::split() return one empty string if called on an empty input.
Trying to cast an empty string to a token value results in a bad_lexical_cast
exception. Fix it by handling empty token list explicitly.
Message-Id: <20170302125405.GU11471@scylladb.com>
From Paweł:
These patches optimise commitlog_entry_writer so that it avoids copying
column mapping, which is a particularly expensive operation.
perf_simple_query -c4 --write --duration 60
(medians)
before after diff
write 79434.35 89247.54 +12.3%
Tested with:
commitlog_test.py:TestCommitLog.test_commitlog_replay_on_startup
commitlog_test.py:TestCommitLog.test_commitlog_replay_with_alter_table
commitlog_test.py:TestCommitLog.test_commitlog_replay_with_counters
The general algorithm for merging counter cells involves allocating a
new buffer for the shards. However, it is expected that most of the
applies are just updating the values of existing shards and not adding
new ones, therefore can be done in place.
However, reverting the general and in-place applies requires different
logic, hence the need for an additional flag to differentiate between
them.
Encountering tombstones while transforming counter update from deltas to
shards is expected to be rare due to the fact that counter cells cannot
be recreated once removed.
This assumption makes it unnecessary to care much about removed cells
during delta->shard transformation as it adds complexity to the code and
is not required to produce correct results.
This patch attempts to avoid excessive allocations and copies when
constructing counter cells using counter_cell_builder. That involves
adding serializer interface to atomic_cell so that the counter cell can
be directly serialized to the buffer allocated for atomic cell.
counter_cell_builder::from_single_shard() is added as well to avoid
std::vector<> overhead when creating a counter cell from a single shard.
By default behavior is kept the same. There are deployments in which we
would like to mount data and commitlog to different places - as much as
we have avoided this up until this moment.
One example is EC2, where users may want to have the commitlog mounted
in the SSD drives for faster writes but keep the data in larger, less
expensive and durable EBS volumes.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1488258215-2592-1-git-send-email-glauber@scylladb.com>
If query_time is time_point::min(), which is used by
to_data_query_result(), the result of subtraction of
gc_grace_seconds() from query_time will overflow.
I don't think this bug would currently have user-perceivable
effects. This affects which tombstones are dropped, but in case of
to_data_query_result() uses, tombstones are not present in the final
data query result, and mutation_partition::do_compact() takes
tombstones into consideration while compacting before expiring them.
Fixes the following UBSAN report:
/usr/include/c++/5.3.1/chrono:399:55: runtime error: signed integer overflow: -2147483648 - 604800 cannot be represented in type 'int'
Message-Id: <1488385429-14276-1-git-send-email-tgrabiec@scylladb.com>
In the later stages of counter write path a mutation is produced that
already has all cells transformed to counter shards and can be applied
to the memtable and written to the commitlog.
The current interface expectes a frozen mutation, which is suboptimal
for counters. The freeze itself is unaviodable -- it is required by
commitlog, but we can avoid later deserialization of frozen_mutation
when it is applied to the memtable if we pass the unfrozen mutation
along.
Counter write path involves read-modify-write. That read is guaranteed
to query only a single partition, does not care about dead cells and
expects to receive an unserialized mutation as a result.
Standard mutation queries can are able to produce results fit for
counter updates, but the logic involved is much more general (i.e.
slower), hence the addition of new, counter-specific kind of query.
Since gcc-5/stretch=5.4.1-2 removed from apt repository, we nolonger able to
build gcc-5.
To avoid dead link, use launchpad.net archives instead of using apt-get source.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1488189378-5607-1-git-send-email-syuu@scylladb.com>
Fixes the following UBSAN warning:
core/semaphore.hh:293:74: runtime error: reference binding to misaligned address 0x0000006c55d7 for type 'struct basic_semaphore', which requires 8 byte alignment
Since the field was not initialied properly, probably also fixes some
user-visible bug.
Message-Id: <1488368222-32009-1-git-send-email-tgrabiec@scylladb.com>
This patch replaces the current heap with a logarithmic histogram
to hold the closed segment descriptors.
This histogram stores elements in different buckets according to
their size. Values are mapped to a sequence of power-of-two ranges
that are split in N sub-buckets. Values less than a minimum value
are placed in bucket 0, whereas values bigger than a maximum value
are not admitted.
There is some loss of precision as segments are now not totally
ordered, and precision decreases the more sparse a segment is. This
allows to reduce the cost of the computations needed when freeing
from a closed segment.
Performance results for perf_simple_query -c4 --duration 60
before after diff
read 43954.27 45246.10 +2.9%
write 48911.54 52807.76 +7.9%
Fixes#1442
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170227235328.27937-1-duarte@scylladb.com>
* seastar 4d4a58d...5861f99 (9):
> future: adjust finally constraint to allow any future to be returned from the continuation
> build: allow specifying the C compiler
> socket: Change signature (and impls) of socket shutdown to void
> reactor: give names to OS threads
> Concepts support
> core/file: Fix short-read in read_maybe_eof()
> core/fstream: Avoid issuing read requests beyond _remain
> tests: Improve assertion failure message
> reactor: Expose IO stats in a public API
It is safe to copy column_mapping accros shards. Such guarantee comes at
the cost of performance.
This patch makes commitlog_entry_writer use IDL generated writer to
serialise commitlog_entry so that column_mapping is not copied. This
also simplifies commitlog_entry itself.
Performance difference tested with:
perf_simple_query -c4 --write --duration 60
(medians)
before after diff
write 79434.35 89247.54 +12.3%
"This introduces an API which allows forward navigation in a stream of mutation
fragments. It allows one to consume only a subset of the stream by iteratively
specifying sub-ranges from which fragments should be returned.
API outline:
When in forwarding mode, the stream does not return all fragments right away,
but only those belonging to the current range. Initially current range only
covers the static row. The stream can be forwarded, even before reaching end-
of-stream for current range, to a later range with fast_forward_to().
Forwarding doesn't change initial restrictions of the stream, it can only be
used to skip over data.
Monotonicity of positions is preserved by forwarding. That is fragments
emitted after forwarding will have greater positions than any fragments
emitted before forwarding.
For any range, all range tombstones relevant for that range which are present
in the original stream will be emitted. Range tombstones emitted before
forwarding which overlap with the new range are not necessarily re-emitted.
When not in forwarding mode, the stream acts as if the current range was equal
to the full range. This implies that fast_forward_to() cannot be
used.
Whether stream is in forwarding mode or not is specified when the stream
is created, typically via mutation_source interface.
What's left for later series:
Optimization by providing specialized implementations. This series implements
forwarding support in all mutation sources via generic wrapper which simply
drops fragments."
* tag 'tgrabiec/clustering-fast-forward-to-v2' of github.com:scylladb/seastar-dev:
tests: mutation_source_tests: Verify monotonicty of positions
tests: random_mutation_generator: Spread the keys more
tests: mutation_source_test: Make blobs more easily distinguishable
tests: streamed_mutation: Test that merged stream passes mutation source tests
tests: mutation_source_test: Add tests for forwarding of streamed_mutation
tests: streamed_mutation_assertions: Add methods for navigating the stream
tests: Add range generators to random_mutation_generator
partition_slice_builder: Add with_ranges()
query: Introduce full_clustering_range
streamed_mutation: Add non-owning variant of mutation_from_streamed_mutation()
db: Enable creating forwardable readers via mutation_source
mutation_source: Document liveness requirements
mutation_source: Cleanup
db: Replace virtual_reader_type with mutation_source_opt
partition_version: Refactor make_partition_snapshot_reader() overloads
database: Fix mutation_source created by as_mutation_source() to not ignore trace_state_ptr
memtable: Accept all mutation_source parameters
streamed_mutation: Implement fast_forward_to() in stream merger
streamed_mutation: Add generic implementation of forwardable streamed_mutation
streamed_mutation: Add fast_forward_to() API
position_in_partition: Introduce position_range
position_in_partition: Introduce position constructor for right after the static row
streamed_mutation: Make cast to view non-explicit
streamed_mutation: Make schema() getter non-copying
So that streamed_mutation is created in only one of the overloads and
others delegate to that one. Later there will be common logic added to
the construction and doing this will help avoid a duplication.
It was using the state passed via as_mutation_source() instead. Let's
respect mutation_source contract instead, and use the state passed via
mutation_source invocation.
Technically just a cleanup. Alse prerequisite for more cleanup.
Do not ignore a future<> retuned by cycle() since it will produce a
warning in case of an error. Log it instead.
Message-Id: <20170219151811.GN11471@scylladb.com>
Set murmur3_partitioner_ignore_msb_bits to 12 (enabling the new sharding
algorithm), but do this in scylla.yaml rather than the built-in defaults.
This avoids changing the configuration for existing clusters, as their
scylla.yaml file will not be updated during the upgrade.
Message-Id: <20170214123253.3933-1-avi@scylladb.com>
Adds yet another magic function "SCYLLA_COUNTER_SHARD_LIST", indicating that
argument value, which must be a list of tuples <int, UUID, long, long>,
should be inserted as an actual counter value, not update.
This of course to allow counters to be read from sstable loader.
Note that we also need to allow timestamps for counter mutations,
as well as convince the counter code itself to treat the data as
already baked. So ugly wormhole galore.
v2:
* Changed flag names
* More explicit wormholing, bypassing normal counter path, to
avoid read-before-write etc
* throw exceptions on unhandled shard types in marshalling
v3:
* Added counter id ordering check
* Added batch statement check for mixing normal and raw counter updates
Message-Id: <1487683665-23426-2-git-send-email-calle@scylladb.com>
* seastar 5088065...4d4a58d (3):
> reactor utilization should return the utilization in 0-1 range
> collectd should ignore type label in name creation
> fix append_challenged_posix_file_impl::process_queue() to handle recursion
On database stop, we do flush memtables and clean up commit log segment usage.
However, since we never actually destroy the distributed<database>, we
don't actually free the commitlog either, and thus never clear out
the remaining (clean) segments. Thus we leave perfectly clean segments
on disk.
This just adds a "release" method to commitlog, and calls it from
database::stop, after flushing CF:s.
Message-Id: <1485784950-17387-1-git-send-email-calle@scylladb.com>
Failing to close a file properly before destroying file's object causes
crashes.
[tgrabiec: fixed typo]
Message-Id: <20170221144858.GG11471@scylladb.com>
5a0955e89d "db: add operations for
applying counter updates" merged two column_family::apply() overloads
into do_apply() in order to reduce code duplication. Unfortunately,
a call to check_valid_rp() didn't survive that change.
Message-Id: <20170221133800.30411-1-pdziepak@scylladb.com>
When loading a schema asynchronously, we're leaving a strong
reference to the loaded schema in the entry's shared future. This
patch fixed this by storing a shared_promised, which is reset when the
schema is loaded.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170220193654.17439-1-duarte@scylladb.com>
This reverts commit 1e2c01ff49.
We do not detect repeated tombstone if it follows an in-range
tombstone following a skipped clustering row, because _in_progress
will be disengaged after such tombstone is emitted.
Message-Id: <1487596080-21480-1-git-send-email-tgrabiec@scylladb.com>
Move lister class away from database.cc.
This is a preparation for moving it to the seastar library.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
The current implementation of 'nodetool clearsnapshot' command only deletes the snapshots
of the keyspaces that are alive at the time the command is issued (issue #2045).
This, besides not implementing the spec, prevents users from being able to clear
the disk space occupied by snapshots of deleted keyspaces that are no longer needed
(e.g. snapshots created when KS is deleted).
This patch fixes this issue by making the database::clear_snapshot() scan the data directories
looking for the snapshots to be deleted instead of relying on in-memory data structures.
This patch makes column_family::clear_snapshot() method not needed any more.
Fixes#2045
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Filter should get all information that the caller has in hand that may be used for filtering.
directory_entry has the following information:
- Type of the entry
- Its name
For the code that used lister filters so far this would be enough, however it's not
hard to imagine a filter that may need the parent directory as well.
We will add the parent directory path in the follow up patches to make the interface
complete.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
If show_hidden parameter is set to show_hidden::yes - list hidden entries, otherwise skip them.
By default set to show_hidden::no.
This patch also completely removes default parameters in lister::scan_dir() and replaces them
with a few lister::scan_dir() overloads that ensure that lambdas are always going to
be the last parameter in the parameters list.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
There is a possibility that the type of the given entry may not be available
that would manifest in the ENOENT or ENOTDIR value set in the errno
by the fstat() call for this entry. In this case engine().file_type() will return
a not engaged optional<directory_entry_type> value.
Return the future with the std::runtime_error exception in this case.
This will prevent any further usage of the not engaged optional value by the code
in the normal flow.
The exception is going to be propagated to the caller and it's the caller's responsibility
to handle it.
Fixes#2071
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
* seastar 28a143a...5088065 (8):
> configure.py: switch cmake to build c-ares to do out-of-source-tree build
> iotune: make sure help is working
> collectd: send double correctly for gauge
> tls: make shutdown/close do "clean" handshake shutdown in background
> tls: Make sink/source (i.e. streams) first class channel owners
> native-stack: Make sink/source (i.e. streams) first class channel owners
> posix-stack: Make sink/source (i.e. streams) first class channel owners
> Merge "Detector for tasks blocked" from Glauber
Fixes#2085.
Packaging updated to require cmake, drop libtool and automake.
"This series contains some fixes and a unit test for the logic responsible
for locking counter cells."
* 'pdziepak/cell-locking-fixes/v1' of github.com:cloudius-systems/seastar-dev:
tests: add test for counter cell locker
cell_locking: fix schema upgrades
cell_locker: make locker non-movable
cell_locking: allow to be included by anyone
* cell_entry destructor may be called when the former is unlinked
* update pointer to schema in partition_entry on schema upgrade
* use correct bucket count when creating a new hash table
The test started to fail sporadically on jenkins after
7a00dd6985 due to quiesce() timing out. It's not
clear though if this is a regression because before the series such timeouts
would not cause test failure if the future resulves eventually, timeout was
only logged.
I was not able to reproduce it on my setup nor on jenkins, so let's add more
debugging output and trigger a coredump next time the test fails.
Message-Id: <1487089576-27147-1-git-send-email-tgrabiec@scylladb.com>
ninja-build-1.6.0-2.fc23.src.rpm on fedora web site deleted for some
reason, but there is ninja-build-1.7.2-2 on EPEL, so we don't need to
backport from Fedora anymore.
Fixes#2087
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1487155729-13257-1-git-send-email-syuu@scylladb.com>
"Immediate reason to do this is to ensure that forwarding of streamed_mutation
will give the same mutations as slicing would, and have unit tests which
verify that those two access methods are consistent with each other.
Secondary reason is performance, to avoid processing unnecessary data.
Note that this should not cause digest mismatch of data queries during rolling
upgrade, because data queries are checksumming only tombstones affecting rows
in the results, so only relevant tombstones.
Fixes #1254."
* tag 'tgrabiec/only-relevant-range-tombstones-v2' of github.com:scylladb/seastar-dev:
tests: mutation_source_test: Test that slicing returns only relevant range tombstones
tests: Pass all mutation source parameters
tests: mutation_source_tests: Ensure timestamps are strictly monotonic
tests: streamed_mutation_assertions: Add more expectation methods
tests: streamed_mutation_assertions: Make produces_end_of_stream() give better error messages
sstables: Simplify sstable_streamed_mutation::read_next()
sstables: Emit only relevant range tombstones
range_tombstone: Introduce end_position()
position_in_partition: Print position when printing fragment
position_in_partition: Make printable
position_in_partition: Add cast to view
position_in_partition: Generalize from-bound_view constructor
bound_view: Extract converters for range start and end bounds
mutation_partition: Drop unneeded range tombstones
mutation_partition: Simplify row removal
range_tombstone_list: Introduce erase()
partition_snapshot_reader: Emit only relevant tombstones
range_tombstone_stream: Add slicing apply() overload
range_tombstone_list: Introduce slice()
write_{live, counter, expiring, dead}_cell() take a const reference to
an atomic cell as argument. However, their caller (which is
write_row_cells) passes to them an atomic_cell_view.
There is an appropriate implicit constructor so instead of compiler
complaints we get atomic_cell objects being constructed from views which
involves an allocation and a copy.
Message-Id: <20170213100106.9071-1-pdziepak@scylladb.com>
* seastar 83a41c8...28a143a (5):
> prometheus: send one MetricFamily per unique metric name
> tests: Add test for circular_buffer::erase()
> circular_buffer: Introduce erase()
> protect against infinite do_until loop
> metrics: alternative metrics creation with labels
There are several problems with storage_proxy::send_to_endpoint right
now. It uses create_write_response_handler() overload that is specific
to read repair which is suboptimal and creates incorrect logs, it does
not process errors and it does not hold storage_proxy object until write
is complete. The patch fixes all of the problems.
Message-Id: <20170208101949.GA19474@scylladb.com>
Reviewed-by: Nadav Har'El <nyh@scylladb.com>
"This series changes buffering of mutation fragments in streamed mutations
so that the size of the fragments is taken into account.
The original implementation buffered up to 16 fragments which was pretty
much meaningless since it could be far too much if the fargments were
large or not nearly enough in case they were small
Fixes #2036.."
* 'pdziepak/buffer-mfs-by-size/v1' of github.com:cloudius-systems/seastar-dev:
streamed_mutation: size-based mutation_fragment buffer limit
mutation_fragment: cache size in memory
mutation_fragment: make write access more explicit
Currently, streamed mutations buffer up to 16 mutation fragments. This
may be too much, not enough or a perfect choice depending on the
mutation fragment size.
This patch makes streamed mutation choose how much mutation fragments to
keep in the buffer depending on their size, so that we avoid using too
much memory in case of large mutation fragments and are able to buffer a
lot of fragments if they are small.
mutation_fragments are going to be caching their size in memory. In
order to be able to invalidate that correctly, they need to know when
that size may change (but avoid invalidation when it is not necessary).
"This series uses the newly added histogram and label support to add metrics to
the storage_proxy and to the column_family.
This would add latency and histogram and the missing metrics from column family."
* 'amnon/histogram_metrics' of github.com:cloudius-systems/seastar-dev:
database: add metrics registration for the coloumn family
storage_proxy: add read and write latency histogram
estimated_histogram: returns a metrics histogram
"This series makes sure that schemas containing both counter and non-counter
regular or static columns are not allowed."
* 'pdziepak/disallow-mixed-schemas/v1' of github.com:cloudius-systems/seastar-dev:
schema: verify that there are no both counter and non-counter columns
test/mutation_source: specify whether to generate counter mutations
tests/canonical_mutation: don't try to upgrade incompatible schemas
Tests using random mutation generator should be provided with bot
counter and non-counter mutations to ensure that both cases are
sufficiently covered. However, mixed schemas (with both counter and
non-counter columns) are not allowed so the RMG has to be explicitly
told whether to use counter or non-counter schema.
Test case test_reading_with_different_schemas uses randomly generated
pairs of mutations and tries to upgrade one to the schema of the other.
However, there are cases when one schema cannot be upgraded to another,
for example, counter and non-counter schemas.
This patch adds a metrics registration to the column_family.
Using label each column metrics is label with its keyspace and column
family name.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
The metrics histogram is a struct that describe a histogram.
This patch adds a getter method that lets the estimated_histogram return
a metrics::histogram, this will allow to register it as a histogram
metrics.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
"This patchset adds bits of the MV write-path, enough to support
new entries to be added. Note that this is still limited, as only
adding new rows to a base table will work correctly."
* 'materialized-views/insert-path/v4' of https://github.com/duarten/scylla: (30 commits)
database: Apply mutation to views
column_family: Push view replica update
materialized views: partial mutate_MV
materialized views: function to send a mutation to endpoint
materialized views: add VIEW write type
database: Ensure new write_type is correctly printed
materialized views: match base and view replicas
column_family: Generate view updates
column_family: Adds affected_views() function
view: Add view_update_builder class
range_tombstone_accumulator: Expose current tombstone
range_tombstone_accumulator: apply() takes value
view_updates: Generate updates
view_updates: Adds function to replace row
view_updates: Update view entry
view_updates: Delete old view entry
mutation_partition: Introduce shadowable tombstone
view_updates: Create view entry
view_updates: Compute row marker
view: Introduce view_updates class
...
This patch changes the database apply path so that it also
generates the mutations for the column family's views and
sends them to the paired view replicas.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This adds a function mutate_MV() which takes view mutations and sends
them to the appropriate nodes (this may be the current node, or a
remote node).
This is only a partial implementation - we still don't do the local
batch log (to survive reboots and failures) and some other stuff which
is left commented out.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Add a function for sending one mutation to one remote replica owning
this mutation. This is needed for materialized views, where each
base replica sends each view mutation to one particular view replica.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
This adds to the "write_type" enum also the "VIEW" write type.
To be honest, I don't understand why the "write_type" distinction
is important.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
By removing the default case in the switch statement over a write_type
variable, we ensure the compiler warns us about lack of exhaustiveness
in case we add a value to the enum but forget to change the
corresponding operator<<().
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
A function to find the appropriate replica to send a view update to.
This patch creates a new source file db/view/view.cc. We should
eventually move a lot more of the materialized views code there.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
This patch adds the generate_view_updates() function to the
column_family class, which will use the view_update_builder to
generate updates to the column_family's materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the view_update_builder class, which is responsible
for calculating the mutations to apply to a column family's
materialized views, given a streamed_mutation representing an update
to the base table and a streamed_mutation representing the
pre-existing rows which the update covers.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
range_tombstone_accumulator::apply() now takes a value so the caller
can decide whether to move or copy the argument.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the view_updates::generate_update() function to
generate view updates given a base row update and the corresponding,
pre-existing row. This function will decide which of the previously
introduced functions to call based on whether there is a pre-existing
row and whether there exists a regular base column that's part of the
view's PK.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a function to replace a view row given a base
table update and the pre-existing row, which simply deletes the old
view entry and adds a new one.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the view_updates::update_entry function,
which creates the updates to apply to the existing view entry given
the base table row before and after the update.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the view_updates::delete_old_entry function,
which creates a view row mutation to delete an entry given an updated
base table row.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces shadowable row tombstones. A shadowable row
tombstone is valid only if the row has no live marker. In other words,
the row tombstone is only valid as long as no newer insert is done
(thus setting a live row marker; note that if the row timestamp set
is lower than the tombstone's, then the tombstone remains in effect
as usual).
If a row has a shadowable tombstone with timestamp Ti and that row
is updated with a timestamp Tj, such that Tj > Ti (and that update
sets the row marker), then the shadowable tombstone is shadowed by
that update. A concrete consequence is that if the update has cells
with timestamp lower than Ti, then those cells are preserved (since
the deletion is removed), and this is contrary to a regular,
non-shadowable row tombstone where the tombstone is preserved and
such cells are removed.
Currently, only Materialized Views require shadowable row tombstones,
which solve a problem with view row deletions. Consider a base row with
columns p, v1, v2, PRIMARY KEY (p) denormalized into a view row consisting
of columns p, v1, v2 PRIMARY KEY (p, v1), and the following operations:
1) INSERT INTO base (p, v1, v2) VALUES (0, 0, 1) USING TIMESTAMP 0;
2) UPDATE base SET v1 = 1 USING TIMESTAMP 1 WHERE p = 0;
3) UPDATE base SET v1 = 0 USING TIMESTAMP 2 WHERE p = 0;
Without shadowable tombstones, the view contains:
At 1), pk = (0, 0), row_marker@T0, v2=1@T0
At 2), pk = (0, 0), row_marker@T0, row_tombstone@T1, v2=1@T0
pk = (0, 1), row_marker@T1, v2=1@T0
At 3), pk = (0, 0), row_marker@T2, row_tombstone@T1, v2=1@T0
pk = (0, 1), row_marker@T1, row_tombstone@T2, v2=1@T0
Notice how, if we read row (0, 0), the value of v2 will be shadowed by
the row tombstone we previously inserted.
With a view's row tombstone becoming shadowable, at 3) the row (0, 0)
will look like pk = (0, 0), row_marker@T2, shadowable_tombstone@T1, v2=1@T0,
which is equivalent to pk = (0, 0), row_marker@T2, v2=1@T0.
Since the shadowable tombstone is shadowed by the new row marker (T0 <
T2), now v2 would be taken into account.
Finally, note that this patch doesn't generalize the idea of
shadowable tombstone, instead taking advantage of the fact that they
are only needed by Materialized Views. This saves changing the
tombstone representation to account for an extra flag, the bits such
representation would require, and also avoids changes to the storage
format.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the view_updates::create_entry function, which
creates a view row mutation given a new base table row.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a function to compute the row marker of a view row
given the base row. There are two cases to consider when building the
row marker: 1) there is a column C that is a regular base column but
is in the view PK; and 2) the columns for the base and the view PKs
are the same.
For 1), the view row marker timestamp will be the biggest between the
base's row marker and C. The TTL will be that of C. This means that if
C expires, the view row maker will expire as well (and the row, if no
other column is keeping it alive). Note that if the base row marker
expires but not C, then the base row will still be live due to C and
we shouldn't expire the view row.
For 2), the view row timestamp will be the same as the base row
timestamp. The TTL should be set in such a way that both base and view
rows live for the same time. We thus set the view row TTL to be the
max of any other TTL in the base row. This is particularly important
in the case where the base row marker has a TTL, but a column *absent*
from the view holds a greater one.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the view_updates class, which is responsible
for generating and storing updates to a particular materialized view.
The updates will be generated from an updated base row and the
pre-existing one (if any), in later patches.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the collection_type_impl::for_each_cell()
function, which allows the caller to iterate over the cells of a
particular collection_mutation_view.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
To help calculate the view mutations from a base update, we store in
the view class the column that's part of the view's primary key but
not part of the base's, if such column exists.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the matches_view_filter() function which specifies
whether a given base row matches the view filter. Unlike
may_be_affected_by(), this function has no false positives.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the is_satisfied_by() function to
single_column_restriction, which given a clustering row returns
whether the restrictions applies or not.
This is useful for secondary indexing such as materialized views,
where filters on regular columns precisely select which base table
rows to denormalize.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch exposes the non-primary key column restrictions in a given
select statement, exposing them as single_column_restrictions.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the is_set() and is_list() functions to
collection_type_impl, which identify the concrete collection
type.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch renames abstract_restriction::uses_function() to
term_uses_function(), as it was previously hiding a function with the
same name in the restriction base class.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the may_be_affected_by() function to the view class,
which is responsible to determine whether an update to a base class
affects one of its views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Instead of storing the view in the column_family's map of materialized
views, store a lw_shared_ptr so that the view can be removed while it
is being updated.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch fixes a use-after-free error in
rename_column_in_where_clause(), where we were creating a boost
adaptor on an rvalue.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"Fixes #1531
Adds lookup to gms::inet_address and uses it in (hopefully all) the
salient places where configured symbolic names are interpreted.
Removes the dummy dns modula in scylla in favour of the seastar one."
* 'calle/use-dns' of github.com:cloudius-systems/seastar-dev:
remove scylla dns code
service::storage_service: Remove depedency on scylla dns
main.cc: remove scylla dns dependency
main/init: Lookup inet addresses from config by dns lookup
db::system_keyspace: Find rpc_address by lookup
gms::inet_address: Add lookup functionality.
scylla tls: Add option support for client auth and tls opts
This helps achieve more repeatable runs that can then be compared via the
Linux perf tool. The option overrides duration-based testing and runs the
test for a specific number of iterations.
Message-Id: <20170204172937.8462-1-avi@scylladb.com>
Merge commit 45b6070832 used butchered version of storage_proxy
patch to adjust to rpc timer change instead the one I've sent. This
patch fixes the differences.
Message-Id: <20170206095237.GA7691@scylladb.com>
Refs #1813 (fixes scylla part)
Added require_client_auth and priority_string options to
server_encryption_options/client_encryption_options an process them.
Allows TLS method/algo specification. Also enabled enforcing known cert
authentication for both node-to-node and client communication.
scylla-housekeeping requires to run 'restart mode' for check the version during
scylla-server restart, which wasn't called on systemd timer so added it.
Existing scylla-housekeeping.timer renamed to scylla-housekeeping-daily.timer,
since it is running 'daily mode'.
Fixes#1953
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1486180031-18093-1-git-send-email-syuu@scylladb.com>
The comparator constructor took schema by value instead of const l-ref
and, consequently, later tried to access object that has been destroyed
long time ago.
Message-Id: <20170202135853.8190-1-pdziepak@scylladb.com>
"Before, the logic for releasing writes blocked on dirty worked like this:
1) When region group size changes and it is not under pressure and there
are some requests blocked, then schedule request releasing task
2) request releasing task, if no pressure, runs one request and if there are
still blocked requests, schedules next request releasing task
If requests don't change the size of the region group, then either some request
executes or there is a request releasing task scheduled. The amount of scheduled
tasks is at most 1, there is a single releasing thread.
However, if requests themselves would change the size of the group, then each
such change would schedule yet another request releasing thread, growing the task
queue size by one.
The group size can also change when memory is reclaimed from the groups (e.g.
when contains sparse segments). Compaction may start many request releasing
threads due to group size updates.
Such behavior is detrimental for performance and stability if there are a lot
of blocked requests. This can happen on 1.5 even with modest concurrency
because timed out requests stay in the queue. This is less likely on 1.6 where
they are dropped from the queue.
The releasing of tasks may start to dominate over other processes in the
system. When the amount of scheduled tasks reaches 1000, polling stops and
server becomes unresponsive until all of the released requests are done, which
is either when they start to block on dirty memory again or run out of blocked
requests. It may take a while to reach pressure condition after memtable flush
if it brings virtual dirty much below the threshold, which is currently the
case for workloads with overwrites producing sparse regions.
I saw this happening in a write workload from issue #2021 where the number of
request releasing threads grew into thousands.
Fix by ensuring there is at most one request releasing thread at a time. There
will be one releasing fiber per region group which is woken up when pressure is
lifted. It executes blocked requests until pressure occurs."
* tag 'tgrabiec/lsa-single-threaded-releasing-v2' of github.com:cloudius-systems/seastar-dev:
tests: lsa: Add test for reclaimer starting and stopping
tests: lsa: Add request releasing stress test
lsa: Avoid avalanche releasing of requests
lsa: Move definitions to .cc
lsa: Simplify hard pressure notification management
lsa: Do not start or stop reclaiming on hard pressure
tests: lsa: Adjust to take into account that reclaimers are run synchronously
lsa: Document and annotate reclaimer notification callbacks
tests: lsa: Use with_timeout() in quiesce()
"This series introduces support for counters. The implementation of
counters more or less follows the design described on our wiki page [1].
Counter cells contain many shards with replicas being able to modify
and announce new versions only of the shards that they own. Historically,
there were three types of shards: local, remote and global. In these
patches only support for the global ones is added.
[1] https://github.com/scylladb/scylla/wiki/Counters
Currently, counters are only enabled as experimental features as there
still several things that need to be done before they become production
ready. Namely, the performance is expected to be quite poor (especially
for writes), there is no proper tracing support and timed out counter
requests may not be recognized and dropped early. There are also no
counter-related metrics.
However, apart from these problems there are no other missing parts of
counter implementation and they are expected to work correctly.
Fixes #577."
* 'pdziepak/counters/v3-rebased' of github.com:cloudius-systems/seastar-dev: (38 commits)
perf_simple_query: add counter tables tests
thrift: add support for counter operations
cql3: allow counters in CREATE TABLE statements
cql3: selection: do not panic when seeing counters
storage_proxy: support counter updates
storage_proxy: add get_live_endpoints()
cql3: add counter increment and decrement operations
db: add operations for applying counter updates
counters: implement transforming counter deltas to shards
add infrastructure for locking counter cells
add fnv1a hasher
position_in_partition: add feed_hash()
position_in_partition: add functions for querying object type
types: make counter_type_impl report its cql3_type
transport: encode counters as long_type
mutation_partition: make for_each_cell() accessible outside source file
messaging_service: add COUNTER_MUTATION verb
storage_service: add COUNTERS feature
idl: add idl description of consistency level
schema: make is_counter() return correct value
...
The leader receives counter updates as deltas which have to be
transformed to counter shards. In order to do that, current local shard
of the modified counter cell needs to be read, logical clock incremented
and the value modified by the specified delta.
The leader receives counter update in a form of deltas which need to be
transformed to counter shards. In order to do that the node needs to
read its current state of the modified counter cells. Since this is
essentially a read-modify-write opertation an appropriate locking
mechanism is needed.
Counter cell locker introduced in this patch uses a hashtable of
partition entry each containing a hashtable of cell entries. Inside a
cell entry there is a semaphore used for synchronization. Once no longer
needed cell entries and partition entries are removed.
In order to avoid deadlocks cell entries are always locked in the same
order which is the lexicographical order of (clustering key, column id)
pairs. Note that schema changes are not a difficulty since they do not
make it possible to change ordering of such pairs.
Support for deletion of counters is limited in a way that once deleted
they cannot be used again (i.e. tombstone always wins, regardless of the
timestamp). Logic responsible for merging two counter cells already
makes sure that tombstones are handled properly, but it is also
necessary to ensure that higher level tombstones always cover counters.
Live counter cells are collections of shards, each one representing the
sum of all operations performed by a particular replica. This commits
introduces an in-memory representation of counters as well as basic
operations such as merge, difference and hashing.
* seastar c1dbd89...f07f8ed (3):
> Merge "Introduce when_all_succeed()" from Paweł
> tests: adjust collectd test for metric API change
> Merge "DNS query support" from Calle
Before, the logic for releasing writes blocked on dirty worked like
this:
1) When region group size changes and it is not under pressure and
there are some requests blocked, then schedule request releasing
task
2) request releasing task, if no pressure, runs one request and if
there are still blocked requests, schedules next request
releasing task
If requests don't change the size of the region group, then either
some request executes or there is a request releasing task
scheduled. The amount of scheduled tasks is at most 1, there is a
single thread of excution.
However, if requests themselves would change the size of the group,
then each such change would schedule yet another request releasing
thread, growing the task queue size by one.
The group size can also change when memory is reclaimed from the
groups (e.g. when contains sparse segments). Compaction may start
many request releasing threads due to group size updates.
Such behavior is detrimental for performance and stability if there
are a lot of blocked requests. This can happen on 1.5 even with modest
concurrency becuase timed out requests stay in the queue. This is less
likely on 1.6 where they are dropped from the queue.
The releasing of tasks may start to dominate over other processes in
the system. When the amount of scheduled tasks reaches 1000, polling
stops and server becomes unresponsive until all of the released
requests are done, which is either when they start to block on dirty
memory again or run out of blocked requests. It may take a while to
reach pressure condition after memtable flush if it brings virtual
dirty much below the threshold, which is currently the case for
workloads with overwrites producing sparse regions.
Refs #2021.
Fix by ensuring there is at most one request releasing thread at a
time. There will be one releasing fiber per region group which is
woken up when pressure is lifted. It executes blocked requests until
pressure occurs.
The logic for notification across hierachy was replaced by calling
region_group::notify_relief() from region_group::update() on the
broadest relieved group.
The hard pressure was only signalled on region group when
run_when_memory_available() was called after the pressure condition
was met.
So the following loop is always an infinite loop rather than stopping
when engouh is allocated to cause pressure:
while (!gr.under_pressure()) {
region.allocate(...);
}
It's cleaner if pressure notification works not only if
run_when_memory_available() is used but whenever conditino changes,
like we do for the soft pressure.
There is comment in run_when_memory_available() which gives reasons
why notifications are called from there, but I think those reasons no
longer hold:
- we already notify on soft pressure conditions from update(), and if
that is safe, notifying about hard pressure should also be safe. I
checked and it looks safe to me.
- avoiding notification in the rare case when we stopped writing
right after crossing the threshold doesn't seem benefitial. It's
unlikely in the first place, and one could argue it's better to
actually flush now so that when writes resume they will not block.
We already call these when crossing the soft threshold. We shouldn't
stop reclaiming when hard pressure is gone because soft pressure may
still be present. Calling start_reclaiming() on hard pressure is
unnecessary because soft pressure also starts it, and when there is
hard pressure there is also soft pressure.
* seastar 397685c...c1dbd89 (13):
> lowres_clock: drop cache-line alignment for _timer
> net/packet: add missing include
> Merge "Adding histogram and description support" from Amnon
> reactor: Fix the error: cannot bind 'std::unique_ptr' lvalue to 'std::unique_ptr&&'
> Set the option '--server' of tests/tcp_sctp_client to be required
> core/memory: Remove superfluous assignment
> core/memory: Remove dead code
> core/reactor: Use logger instead of cerr
> fix inverted logic in overprovision parameter
> rpc: fix timeout checking condition
> rpc: use lowres_clock instead of high resolution one
> semaphore: make semaphore's clock configurable
> rpc: detect timedout outgoing packets earlier
Includes treewide change to accomodate rpc changing its timeout clock
to lowres_clock.
Includes fixup from Amnon:
collectd api should use the metrics getters
As part of a preperation of the change in the metrics layer, this change
the way the collectd api uses the metrics value to use the getters
instead of calling the member directly.
This will be important when the internal implementation will changed
from union to variant.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1485457657-17634-1-git-send-email-amnon@scylladb.com>
need_preempt() is always true in debug mode. Because of that, this loop
will never be executed. Rewrite it as a do-while loop so we are sure
that it is executed at least once - or exactly once in debug mode.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <1485913079-1283-1-git-send-email-glauber@scylladb.com>
The per-node limit will be total memory divided by number of shards
instead of just total memory. For example, when Scylla is started with
-c16 -m16G, the commit log will induce flushes on given shard when
unflushed data exceeds on that shard 62MB instead of 1GB.
Fixes#2046.
Message-Id: <1485874534-10939-1-git-send-email-tgrabiec@scylladb.com>
Currently the test does not wait for cache update
to finish before carrying on with the checks.
This makes the test nondeterministic and purely wrong
because checks expect update to be finished.
This patch changes the test to wait for update to finish.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Message-Id: <2a99bba24b1628466d3495332b48ef3ccdb43c26.1485862389.git.piotr@scylladb.com>
An earlier patch of mine was using should_yield to do the same. That
is a better direction, but should_yield() was demonstrably more
expensive so for now we'll go with need_preempt() - since this is
hurting pretty much every latency-dependent workload.
I am also including the scripts that I have used to measure and
compare the various versions of this patch.
The data() method already returns a bytes_opt so there's no need to call to_bytes_opt() again.
Fixes compliation failure on CentOS:
In file included from ./cql3/query_options.hh:51:0,
from ./cql3/cql_statement.hh:47,
from ./cql3/statements/raw/select_statement.hh:45,
from build/release/gen/cql3/CqlParser.hpp:65,
from build/release/gen/cql3/CqlParser.cpp:44:
./cql3/values.hh: In function 'bytes_opt to_bytes_opt(const cql3::raw_value&)':
./cql3/values.hh:184:37: error: no matching function for call to 'to_bytes_opt(bytes_opt)'
return to_bytes_opt(value.data());
Message-Id: <1485761863-28236-1-git-send-email-penberg@scylladb.com>
The reason is the same as why foreground writes are reported instead of
total writes (049ae37d08): It is much easier to see what is going on
this way.
Also fixes a typo in a counter's description.
Fixes#1217
Message-Id: <20170129093412.GS11469@scylladb.com>
This patch series adds support for "unset values" that were introduced
in CQL binary protocol v4. They allow bound statements to skip updates
to some or all of the bound variables.
Unset values are specified using the BoundStatement.unset() method in
the Java driver:
http://docs.datastax.com/en/drivers/java/3.1/com/datastax/driver/core/BoundStatement.html#unset-int-
and using the UNSET_VALUE constant in the Python driver:
https://datastax.github.io/python-driver/api/cassandra/query.html#cassandra.query.UNSET_VALUEFixes#2039.
* 'penberg/cql-unset-values/v2' of github.com:cloudius-systems/seastar-dev:
transport/server: CQL unset value support
cql3/statements/select_statement: Unset value support
cql3/user_types: Unset value support
cql3/tuples: Unset value support
cql3/maps: Unset value support
cql3/sets: Unset value support
cql3/lists: Unset value support
cql3/constants: UNSET_VALUE constant
cql3/constants: Unset value support
cql3/attributes: Unset value support
types.hh: Add field_name_as_string() to user_type_impl type
cql3: Introduce raw_value and raw_value_view types
The quota check is quite old at the moment, and dates back to a time in
which the infrastructure in seastar threads was lacking a lot. It is a
bad check since it will not take into consideration the size of the
partition or the time it takes to merge them.
A better check would at least take need_preempt() into account, so that
we would respect the task quota. That check is now embedded into
should_yield(), so there would no need to check anything else.
Although should_yield() does the job, it is still currently quite
expensive. And because we are in a seastar thread with a computationally
intensive loop, it can hurt latency a lot.
So as a temporary measure, let's at least check for need_preempt() - as
it is hurting real users at the moment - and soon work on making
should_yield() cheaper.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
update is one of our biggest sources of performance issues as far as the
cache is concerned. systemtap can be useful in helping tracking some of
them down.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Currently, the code is using bytes_opt and bytes_view_opt to represent
CQL values, which can hold a value or null. In preparation for
supporting a third state, unset value introduced in CQL v4, introduce
new raw_value and raw_value_view types and use them instead.
The new types are based on boost::variant<> and are capable of holding
null, unset values, and blobs that represent a value.
This reverts commit d61002cc33.
Introduced a regression in row_cache_alloc_stress.
The problem is that reclaim_from_evictable() evicts way too much after
the refactor due to the stop condition not taking into account how
much data was evicted so far and only looking at occupancy of the
minimal segment. This may lead to eviction of the whole region.
parse_time() adds hourse, minutes, etc to a final value 'result'.
However, it is of type std::chrono::nanoseconds which means it is not
zeroed at initialization unless it is explicitly asked to do so.
Fixed debug mode failures in types_tyes and cql_query_test.
Message-Id: <20170125155239.1253-1-pdziepak@scylladb.com>
"This patchset properly implements range_tombstone_list::difference(),
which was very broken. We add unit tests for the function and ensure
we always randomly generate range_tombstones in other unit tests so
other problems aren't hidden."
This patch ensures the mutation_merger emits any deferred tombstones
that it still may be holding before closing the stream.
Together with the range_tombstone_list: Properly implement
difference() patch set, this fixes breakage of streamed_mutation_test
and row_cache_test.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170123195643.9876-1-duarte@scylladb.com>
* seastar 6d80c6a...397685c (4):
> Merge "add label to the io_queue" from Amnon
> rpc: Modify the shutdown code to wait and handle exceptions
> tls.cc: Fix shutdown_input/output to conform with expected socket behaviour
> core: Add counter for polls
The difference method wasn't properly implemented. The version in this
patch correctly computes the difference and returns a range tombstone
list contains those range tombstones in "this" but absent from the
other, specified range tombstone list.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
* seastar 38aaa4a...6d80c6a (2):
> DPDK: Change the metrics registration with label support
> metric: Fix the error: could not convert {...} from <brace-enclosed initializer list> to struct metric_definition_impl
Add a counter field to RELEASE, just before the date, and fix it at zero.
This allows custom package builds to override it in a way that sorts before
the official packages.
Example:
Official release: 1.6.0-0.20160120.<githash>
Custom release 1: 1.6.0-1.avi.20160121.<githash>
Custom release 2: 1.6.0-2.avi.20160122.<githash>
The counter (0/1/2) ensures that the build number dominates over the date
when sorting.
Message-Id: <20170122102814.19649-1-avi@scylladb.com>
From Avi:
In many cases, batch statements are used to mutate a single partition, or
a number of partitions that is smaller than the number of statements within
the batch. We can detect this case and reduce the numbers of mutations
applied, and in some cases, convert a logged batch into an unlogged batch.
Ref #1689.
continuous_data_consumer::fast_forward_to() returns a future which was
later ignored by data_consume_context::fast_forward_to().
With the current implementation, the future in question is always ready
and that's why the problem didn't manifest itself in the form of crashes
or invalid results.
Message-Id: <20170120105746.7300-1-pdziepak@scylladb.com>
Batch statements are often used to insert multiple rows into the same
partition. Recognize this case and merge mutations to the same partition.
If the result is a single mutation, there is an additional win (already
present in the code), where a logged batch can be converted into an unlogged
batch.
Ref #1689.
Add a boolean to short circuit the read path on empty range
hoping for some speedup.
tested in read write with cs using:
cl=QUORUM duration=1m -mode native cql3 -rate threads=700 -node localhost
Will do some additional benchmark.
Fixes#1056
Signed-off-by: Benoît Canet <benoit@scylladb.com>
Message-Id: <20170118194451.16836-1-benoit@scylladb.com>
That's because a single shard is used to calculate generation for new
sstables in upload directory, and that will result in that single shard
sharing all the resources with other shards.
For refresh without upload dir, it currently works fine because we
reshuffle column family dir instead.
flush_upload_dir() is now a free function, takes a distributed database
object, and uses calculate_shard_from_sstable_generation() to decide
which shard will move sstable using its own generation namespace.
Fixes#2008.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <b0cccf7bbb61416ff8718bac92fdca90cc5fb9c9.1484253232.git.raphaelsc@scylladb.com>
This patch ensures that when adding a shared sstable, we select only
one cpu to update that column family's stats. This is important so we
don't overestimated the on-disk size of sstables when resharding
This fixes only a temporary miscount of the current load, since shared
sstables are eventually re-written, but a fixes a permanent miscount
of the total load.
Refs #1592
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170119144823.31041-1-duarte@scylladb.com>
Currently eviction is performed until occupancy of the whole region
drops below the 85% threshold. This may take a while if region had
high occupancy and is large. We could improve the situation by only
evicting until occupancy of the sparsest segment drops below the
threshold, as is done by this change.
I tested this using a c-s read workload in which the condition
triggers in the cache region, with 1G per shard:
lsa-timing - Reclamation cycle took 12.934 us.
lsa-timing - Reclamation cycle took 47.771 us.
lsa-timing - Reclamation cycle took 125.946 us.
lsa-timing - Reclamation cycle took 144356 us.
lsa-timing - Reclamation cycle took 655.765 us.
lsa-timing - Reclamation cycle took 693.418 us.
lsa-timing - Reclamation cycle took 509.869 us.
lsa-timing - Reclamation cycle took 1139.15 us.
The 144ms pause is when large eviction is necessary.
The change improves worst case latency. Reclamation time statistics
over 30 second period after cache fills up, in microseconds:
Before:
avg = 1524.283148
stdev = 11021.021118
min = 12.934000
max = 144356.000000
sum = 257603.852000
samples = 169
After:
avg = 1317.362414
stdev = 1913.542802
min = 263.935000
max = 19244.600000
sum = 175209.201000
samples = 133
Refs #1634.
Message-Id: <1484730859-11969-1-git-send-email-tgrabiec@scylladb.com>
* seastar 240b0bf...ff098c8 (15):
> metrics::impl::shard(): check if reactor is initialized before using it
> reactor: introduce engine_is_ready()
> fix metric name
> Merge "Add label support to the metric layer" from Amnon
> core: Avoid memory leak when submission to syscall_work_queue fails
> core: Avoid memory leak when submission to smp_message_queue fails
> core: append_challenged_posix_file_impl: Make exception-safe
> Merge "Log backtrace in report_failed_future" from Tomasz
> install-dependencies.sh: add systemtap-sdt-dev to Ubuntu/Debian dependencies
> core: add fsqual.cc/.hh to core
> dpdk: Fix compile error with rte_pci.h
> fstream_test: fix spurious failures due to BOOST_REQUIRE_EQUAL thread-unsafety
> reactor: unregister metrics of queue on shard 0
> build: track system header changes too
> Prometheus: do not rely on collectd for the hostname
close() operation is like a destructor, it cannot fail. It just
reports errors, but close itself succeeds. So we should proceed with
the closing even if it fails.
Message-Id: <1484245886-7269-1-git-send-email-tgrabiec@scylladb.com>
As Tomek pointed out, previous code, regardless of version mismatch, of generating
comparator description string was not correct (as in: in sync with origin).
This modifies it to look at
1.) Actual clustring size
2.) Compound-ness
3.) Dense-ness
to determine whether we should generate a compound desc, and whether it
should contain a trailing utf8-desc type.
v2: Simplify non-dense base column addition and ensure it handles
thrift non-utf8 (as per comments from tomek)
Message-Id: <1484670171-18362-1-git-send-email-calle@scylladb.com>
As the metrics migration progressed, some include to scollectd.hh left
behind.
Because of the nature of the scollecd implementation those include
brings alot of code with them to the header files and eventually to many
source file.
This patch remove those include and add a missing include to
storage_proxy.cc.
The reason the compiler didn't complain is an indication to the
problematic nature of those include in the first place.
Before this patch, change in metrics.hh would cause 169 files to
compile, after this change 17.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1484667536-2185-1-git-send-email-amnon@scylladb.com>
Fixes#2019
According to the Java driver and cassandra, all versions < 3
include the PK in the comparator descriptor string.
This broke for us when bumping the cassandra version 2.1 -> 2.2
Message-Id: <1484657580-14411-1-git-send-email-calle@scylladb.com>
This patch ensures that the host only announces and registers the
MATERIALIZED_VIEWS feature if it was started with the experimental
flag.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170116123412.21365-1-duarte@scylladb.com>
After we call unlink_leftmost_without_rebalance(), we must unlink all
elements before mutatation is destroyed. We did this properly from
~reader, but it would not be called if reader construction failed,
which it may.
Message-Id: <1484572581-6537-1-git-send-email-tgrabiec@scylladb.com>
Since ce083308a1
"random_mutation_generator: Generate RTs by default" random mutation
generator produces range tombstones. However, so far the tests were run
with all features disabled (because of incomplete initialization of all
services) which meant that RANGE_TOMBSTONE feature was not enabled and
the code couldn't handle range tombstones that weren't just prefixes.
This patch solves the problem by forcing all features to be enabled when
tests are run.
Message-Id: <20170116103324.22956-1-pdziepak@scylladb.com>
It still has problems:
- while resharding a very large leveled compaction strategy table, a huge
amount of tiny sstables are generated, overwhelming the file descriptor
limits
- there is a large impact on read latency while resharding is going on
(cherry picked from commit cf27d44412)
(forward-ported from branch-1.6)
Since scylla-housekeeping running as scylla user, it doesn't have a permission
to create a file on /etc/scylla.d.
So introduce /var/lib/scylla-housekeeping which owns by scylla user, place uuid
file on the directory.
Fixes#2009
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1484235946-12463-1-git-send-email-syuu@scylladb.com>
query_mutations_locally() takes one_or_two_partition_ranges by
reference and requires, indirectly, that it is kept alive until
operation resolves. However, we were passing expiring value to it, the
result of unwrap().
Fixes dtest failure in consistent_bootstrap_test.py:TestBootstrapConsistency.consistent_reads_after_bootstrap_test
Another potential problem was that we were dereferencing "s" in the same
expression which move-constructs an argument out of it.
Message-Id: <1484222759-4967-1-git-send-email-tgrabiec@scylladb.com>
On DPDK 16.11 dpdk_nic_bind.py is renamed to dpdk-devbind.py, so we are
getting "file not found" both on packaging and scripts, fixed that.
Also fixed inconsistent packaging.
Since Seastar copied dpdk_nic_bind.py to its scripts/ directory, there're two
different versions of the script, .rpm/.deb packaging different one:
dist/redhat: seastar/dpdk/tools/dpdk_nic_bind.py
dist/ubuntu: seastar/scripts/dpdk_nic_bind.py
That's won't work because we sharing setup scripts between two
distributions, so I changed dist/ubuntu package to use DPDK one.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1484191955-28006-1-git-send-email-syuu@scylladb.com>
Explicitly generate tables' IDs of tables from the system_traces KS using
generate_legacy_id() in order to ensure all Nodes create these tables with
the same IDs.
This is going to prevent hitting issue #420.
Fixes#1976
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1484153725-31030-1-git-send-email-vladz@scylladb.com>
This patch changes the random_mutation_generator so it generates range
tombstones by default. This fixes a bug where reversibly applying
range tombstones wasn't being tested.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20170110164822.28747-1-duarte@scylladb.com>
boost::intrusive::value_traits_pointers was introduced in boost 1.56, while
we also support boost 1.55. Replace with an equivalent expression.
(with additions by Asias)
Message-Id: <20170110084700.19994-1-avi@scylladb.com>
Commit f0c28e1 ("db/schema_tables: Add schema_functions and
schema_aggregates tables") forgot to add the newly added tables to the
db::schema_tables::ALL list, which is used for authorization checks, for
example.
Fixes the following auth_test.py dtest failures:
('Unable to connect to any servers', {'127.0.0.1': Unauthorized('Error from server: code=2100 [Unauthorized] message="User cathy has no SELECT permission on <table system.schema_functions> or any of its parents"',)})
Message-Id: <1484045277-4997-1-git-send-email-penberg@scylladb.com>
"Intended to reduce memory usage when resharding by sharing sstable
components among shards. File descriptors are also shared from now
on, meaning that a much smaller number of file descriptors will be
used during resharding.
Fixes #1951."
branch 'excessive_memory_usage_v4' of github.com:raphaelsc/scylla
* 'excessive_memory_usage_v4' of github.com:raphaelsc/scylla:
db: avoid excessive memory usage during resharding
checked_file_impl: add support to dup
sstables: group sstable components that can be shared among shards
sstables: rename sstable member
After resharding, sstables may be owned by all shards, which
means that file descriptors and memory usage for metadata will
increase by a factor equal to number of shards. That can easily
lead to OOM.
SSTable components are immutable, so they can be stored in one
shard and shared with others that need it. We use the following
formula to decide which shard will open the sstable and share
it with the others: (generation % smp::count), which is the
inverse of how we calculate generation for new sstables.
So if no resharding is performed, everything is shard-local.
With this approach, resource usage due to loaded sstables will
be evenly distributed among shards.
For this approach to work, we now only populate keyspaces from
shard 0. It's now the sole responsible for iterating through
column family dirs. In addition, most of population functions
are now free and take distributed database object as parameter.
Fixes#1951.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
"This patch series adds support for CQL 3.3.1. The changes to CQL are listed
here:
https://github.com/apache/cassandra/blob/cassandra-2.2/doc/cql3/CQL.textile#changes
The following CQL features are already supported by Scylla:
- TRUNCATE TABLE alias
- Double-dollar string literals
- Aggregate functions: MIN, MAX, SUM, and AVG
This series adds the following CQL features:
- New data types: tinyint, smallint, date, and time
- CQL binary protocol v4 (required by the new data types)
- Advertise Cassandra 2.2.8 version from Scylla so that drivers correctly
detect the presence of CQL 3.3.1
The following CQL features are not supported by Scylla:
- Role-based access control (issue #1941)
- JSON data type
- User-defined functions (UDFs)
- User-defined aggregates (UDAs)
The following CQL binary protocol v4 changes are not implemented by this
series:
- Read_failure and Write_failure error codes are not implemented.
They error codes not used by the smart drivers but as they are
propagated to application code, we eventually need to wire them up
to our storage proxy implementation.
- Function_failure error code is only used by user-defined functions
and the fromJson function, which are not implemented by Scylla.
Fixes #1284."
* 'penberg/cql-3.3.1/v5' of github.com:cloudius-systems/seastar-dev:
version: Bump Cassandra version to 2.2.8
db/schema_tables: Add schema_functions and schema_aggregates tables
tests/type_tests: TIME type test cases
tests/cql_query_test: TIME type test cases
cql3: TIME data type support
tests/type_tests: DATE type test cases
tests/cql_query_test: DATE type test cases
cql3: DATE type support
date.h: 64-bit year and days representation
licenses: Add utils/date.h license
utils/date.h: Import date and time library sources
tests/type_tests: TINYINT and SMALLINT type test cases
tests/cql_query_test: TINYINT and SMALLINT type test cases
cql3: TINYINT and SMALLINT data type support
types: Fix integer_type_impl::parse_int() for bytes
mutation_result_merger::get() assumes that the merged result may be a
short read if at least one of the partial results is a short read (in
other words, if none of the partial results is a short read, then the
merged result is also not a short read). However this is not true;
because we update the memory accounter incrementally, we may stop
scanning early. All the partial results are full; but we did not scan
the entire range.
Fix by changing the short_read variable initialization from `no`
(which assumes we'll encounter a short read indication when processing
one of the batches) to `this->short_read()`, which also takes into
account the memory accounter.
Fixes#2001.
Message-Id: <20170108111315.17877-1-avi@scylladb.com>
We need 64-bit year and days representation to support the boundary
values of the CQL data type, which is implemented using Joda Time
library's DateTime type.
Noone overrides file_writer::write() so there is no reason to inhibit
optimisations and cause compiler to emit indirect calls.
Message-Id: <20170104163618.26251-1-pdziepak@scylladb.com>
Ubuntu Packaging Guide says if there's no upstream package (means it's not
ported from Debian), revision should be "0ubuntu1", not "ubuntu1" which is we
currently using.
On Debian, Debian Policy Manual says it's conventional to restart revision from 1 when upstream version increased, so we should specify it to "1".
To do it in single script, we will generate the revision on building time.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1483498658-27491-1-git-send-email-syuu@scylladb.com>
This patch gets housekeeping to create a uuid file if a path to a uuid
file is upplied but the file is missing.
Because it import the uuid lib, uuid parameters where renamed.
Fixes#1987
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1483866553-13855-2-git-send-email-amnon@scylladb.com>
We intend to share immutable sstable components among shards to
reduce excessive memory usage when resharding shared sstables.
This change is about grouping those components into a structure,
and using foreign ptr to make sure that the structure will be
deleted by whichever shard created it.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Rename _components to _recognized_components because _components
will be used to name a field with shareable components.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
* seastar 1c8e389...240b0bf (15):
> file/dup: don't decrease refcnt twice when file is explicitly closed
> reactor: Add missing CentOS 7.2 dependency systemtap-sdt-devel
> reactor: Cleaning the smp queue metrics when shuting down
> metrics: metrics keep the value map while unregistering
> change the reactor load metrics to utilization
> Merge "ASan fiber switches" from Paweł
> tls: Add missing credentials_builder::set_client_auth method
> collectd: create metrics with the right format
> io_queue: remove owner number from metric name
> reactor: change the load metric name to load
> Merge "reactor: stop using signals for task_quota timer"
> metrics: Allow initializing the metric_group in its constructor
> Update DPDK to 16.11
> Revert "rpc: Avoid using zero-copy interface of output_stream"
> core::metrics_groups: add a clear() method
Transform the supervisor_notify() and related functions into
the "supervisor" class and place this class implementation in
a separate .cc file.
This is going to fix the compilation breakage of tests introduced
by a
commit 8014adc2a1
init: serialize the creation of system_traces KS objects
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1483663955-20096-1-git-send-email-vladz@scylladb.com>
"Reduce the size of mutation_partition by implementing intrusive set using
bi::rbtree_algorithms directly and using tree nodes optimized for size.
This will reduce the size of mutation_partition by:
24 bytes + <number of cql rows> * 8 bytes
This should have a positive impact on performance because mutation_partitions
are stored both in memtable and cache.
Fixes #742."
* 'haaawk/742' of github.com:cloudius-systems/seastar-dev:
intrusive_set: rename size() to calculate_size()
Make intrusive_set_external_comparator::_value_traits static
Implement intrusive set using rbtree_algorithms
mutation_partition: make apply_reversibly_intrusive_set nongeneric
mutation_partition: take schema in find_row and clustered_row
mutation_partition: Extract intrusive set logic to a class.
mutation_partition: Replace value_comp with key_comp calls
Before this patch system table writes were not writing to commit log
because database::add_column_family() disables writes to commit log
for the table which is added if _commitlog is not set at that
time. Fix by initializing commit log before system tables are created.
Fixes#1986.
Fixes recent regression in
batch_test.py:TestBatch.replay_after_schema_change_test after
scylla-jmx was updated to not flush system tables on nodetool flush.
Could cause system keyspace writes to be delayed for more than before
under heavy write workload. Refs #1926.
Message-Id: <1483618117-4535-1-git-send-email-tgrabiec@scylladb.com>
During a range scan, we try to avoid sorting according to partition range
when we can do so. This is when we scan fewer than smp::count shards --
each shard's range is strictly ordered with respect to the others.
However, we use the wrong key for the sort -- we use the shard number. But
if we started at shard s > 0 and wrapped around to shard 0, then shard 0's
range will be after the range belonging to shard s, but will sort before it.
Fix by storing the iteration order as the sort key. We use that when we
know that shards do not overlap (shards < smp::count) and the index within
the source partition range vector when they do.
Fixes#1998.
Message-Id: <20170105114253.17492-1-avi@scylladb.com>
Serialize the creation of a system_traces KS objects when
they do not exist - the initial cluster boot.
Avoid creating them in parallel by different cluster Nodes
in order to avoid issue #420.
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1483552503-12873-3-git-send-email-vladz@scylladb.com>
This hopefully will make it more apparent that
the time complexity of this method is O(N) not O(1).
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
_value_traits can be shared among all instances
and there's no need to store it in every single one.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
1.6 truncates paged queries early to avoid overrunning server memory
with too-large query results, but in the case of partition range queries,
this terminates too early due to an uninitialized variable holding the
maximum result size. This results in slow performance due to additional
round trips.
Fix by initializing the maximum result size from the result_memory_tracker
running on the coordinating shard.
Fixes#1995.
Message-Id: <20170105103915.10633-1-avi@scylladb.com>
This new implementation takes less memory because it
does not store comparator.
It also uses tree nodes optimized for size. This means
that instead of storing an enum field |color| they embed
this information inside pointer to parent.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
apply_reversibly_intrusive_set is used only in one place
and always with rows_type. There's no need for it to be generic.
This will allow changing intrusive set implementation.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
The integer_type_impl::parse_int() function uses boost::lexical_cast()
under the hood, which parses 8-bit numbers as characters. Fix the
function to lexical cast to 64-bit integer and convert the result to
integer_type_impl template type.
Previously, if the Prometheus port (by default, 0.0.0.0:9180) could not
be opened, the following message appeared in the log about 10 seconds into
the run, and Scylla crashed.
ERROR 2017-01-01 19:31:04,066 [shard 0] seastar - Exiting on unhandled exception: std::system_error (error system:98, Address already in use)
The puzzled user would have no idea *which* address was already in use, why,
or why Scylla stopped.
In this patch, before the above message we get the much more informative
message:
ERROR 2017-01-01 19:58:19,080 [shard 0] init - Could not start Prometheus API server on 0.0.0.0:9180: std::system_error (error system:98, Address already in use)
We continue to print the original message - and exit - in this case,
under the assumption that it's better not to run the database while
improperly configured.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20170102121304.2060-1-nyh@scylladb.com>
The test case is interested in the upper boundary of 32-bit integer
because we already test the lower boundary in assertions below. The old
test passed, of course, but it wasn't very interesting.
Message-Id: <1483522773-6008-1-git-send-email-penberg@scylladb.com>
Fix the following build breakage:
FAILED: build/release/gen/cql3/CqlParser.o
g++ -MMD -MT build/release/gen/cql3/CqlParser.o -MF build/release/gen/cql3/CqlParser.o.d -std=gnu++1y -g -Wall -Werror -fvisibility=hidden -pthread -I/home/penberg/scylla/seastar -I/home/penberg/scylla/seastar/fmt -I/home/penberg/scylla/seastar/build/release/gen -march=nehalem -Ifmt -DBOOST_TEST_DYN_LINK -Wno-overloaded-virtual -DFMT_HEADER_ONLY -DHAVE_HWLOC -DHAVE_NUMA -DHAVE_LZ4_COMPRESS_DEFAULT -O2 -DBOOST_TEST_DYN_LINK -Wno-maybe-uninitialized -DHAVE_LIBSYSTEMD=1 -I. -I build/release/gen -I seastar -I seastar/build/release/gen -c -o build/release/gen/cql3/CqlParser.o build/release/gen/cql3/CqlParser.cpp
In file included from ./query-request.hh:31:0,
from ./locator/token_metadata.hh:51,
from ./locator/abstract_replication_strategy.hh:29,
from ./database.hh:26,
from ./service/storage_proxy.hh:44,
from ./db/schema_tables.hh:43,
from ./db/system_keyspace.hh:46,
from ./cql3/functions/function_name.hh:45,
from ./cql3/selection/selectable.hh:48,
from ./cql3/selection/writetime_or_ttl.hh:45,
from build/release/gen/cql3/CqlParser.hpp:63,
from build/release/gen/cql3/CqlParser.cpp:44:
./tracing/tracing.hh:357:5: error: ‘scollectd’ does not name a type
scollectd::registrations _registrations;
^~~~~~~~~
Message-Id: <1482939751-8756-1-git-send-email-penberg@scylladb.com>
Commit d41cd48a made the is_joined() method a future<bool> because
only cpu 0 knows its real value. This makes this function inconvenient
to use. So this patch reverts commit d41cd48a, and instead sets this
flag's value on all shards, so each shard can read its value locally
(and immediately).
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20161228160450.5831-1-nyh@scylladb.com>
* seastar f32e4c2...1c8e389 (2):
> Merge "migrate network related seastar collectd metrics to the new metrics registration API" from Vlad
> file: add dup() support
This patch adds an overload to the apply() function,
which takes a clustering_row by reference, to copy. This will be
needed by future patches, when merging base table updates with the
existing data.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1482881106-3202-1-git-send-email-duarte@scylladb.com>
This patch removes references to the old begin_range_tombstone and
end_range_tombstone mutation_fragments, which have been replaced by a
single range_tombstone fragment.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1482880820-2831-1-git-send-email-duarte@scylladb.com>
- Change an exception type thrown by a tracing::tracing::set_trace_probability()
to make it different from the one thrown by an std::stod() when it fails to
parse a given string.
- Catch the std::out_of_range exception thrown by a tracing::tracing::set_trace_probability() and
wrap the exception string into the httpd::bad_param_exception() object.
- Throw a httpd::bad_param_exception() with a
"Bad format in a probability value: <a user given probability string value>"
message if std::invalid_argument is caught.
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Message-Id: <1465300738-1557-1-git-send-email-vladz@cloudius-systems.com>
Refuse to boot if we don't have at least 1 GiB per shard, unless in developer
mode.
The primary violator here is docker, but since it starts in developer mode,
it won't get fixed. We need some extra logic for this case.
Message-Id: <20161221090222.28677-1-avi@scylladb.com>
"This patchset contains fixes for the changes introduced in "Query result
size limiting". It also improves handling of short data reads.
I order to minimise chances of digest mismatch during data queries replicas
that were asked just to return a digest also keep track of the size of the
data (in the IDL representation) so that they would stop at the same point
nodes doing full data queries would. Moreover, data queries are not
affected by per-shard memory limit and the coordinator sends individual
result size limits to replicas in order not to depend on hardcoded values.
It is still possible to get digest mismatches if the IDL changes (e.g. a
new field is added), but, hopefully, that won't be a serious problem."
* 'pdziepak/short-read-fixes/v4' of github.com:cloudius-systems/seastar-dev:
query: introduce result_memory_accounter::foreign_state
storage_proxy: fix short reads in parallel range queries
storage_proxy: pass maximum result size to replicas
mutation_partition: use result limiter for digest reads
query: make result_memory_limiter constants available for linker
result_memory_limiter: add accounter for digest reads
idl: allow writers to use any output stream
result_memory_limiter: split new_read() to new_{data, mutation}_read()
idl: is_short_read() was added in 1.6
mutation_partition: honour allowed_short_read for static rows
storage_proxy: fix _is_short_read computation
storage_proxy: disallow short reads if got no live rows
storage_proxy: don't stop after result with no live rows
This reverts commit aa392810ff, reversing
changes made to a24ff47c637e6a5fd158099b8a65f1191fc2d023; it uses
boost::intrusive::detail directly, which it must not, and doesn't compile on
all boost versions as a consequence.
* seastar 0b98024...f32e4c2 (11):
> Merge "Moving the reactor counters to the metric layer" from Amnon
> metrics: Metrics function should take variable as a refernce
> Revert "Merge ""Moving the reactor counters to the metric layer from Amnon"
> Merge ""Moving the reactor counters to the metric layer from Amnon
> Revert "fstream: Auto-close data_sink and data_source"
> rpc: Avoid resource unit leaks on failure
> fstream: Auto-close data_sink and data_source
> http: Move metrics registration to the metrics layer
> output_stream: add batching to zero copy interface
> Revert "slab: Move the metrics registration to the metrics layer"
> slab: Move the metrics registration to the metrics layer
"Reduce the size of mutation_partition by implementing intrusive set using
bi::rbtree_algorithms directly and using tree nodes optimized for size.
This will reduce the size of mutation_partition by:
24 bytes + <number of cql rows> * 8 bytes
This should have a positive impact on performance because mutation_partitions
are stored both in memtable and cache.
Fixes #742."
* 'haaawk/742' of github.com:cloudius-systems/seastar-dev:
intrusive_set: rename size() to calculate_size()
Make intrusive_set_external_comparator::_value_traits static
Implement intrusive set using rbtree_algorithms
mutation_partition: make apply_reversibly_intrusive_set nongeneric
mutation_partition: take schema in find_row and clustered_row
mutation_partition: Extract intrusive set logic to a class.
mutation_partition: Replace value_comp with key_comp calls
This hopefully will make it more apparent that
the time complexity of this method is O(N) not O(1).
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
_value_traits can be shared among all instances
and there's no need to store it in every single one.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
This new implementation takes less memory because it
does not store comparator.
It also uses tree nodes optimized for size. This means
that instead of storing an enum field |color| they embed
this information inside pointer to parent.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
apply_reversibly_intrusive_set is used only in one place
and always with rows_type. There's no need for it to be generic.
This will allow changing intrusive set implementation.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
file output streams take the responsibility of closing the file, they
will close the file as part of closing the stream.
During sstable writing we create sstable object and keep file
references there as well. Sstable object also has responsibility for
closing the files, and does so from sstable::~sstable().
Double close was supposed to be avoided by a construct like this:
writer.close().get();
_file = {};
However if close() failed, which can happen when write-ahead failed,
_file would not be cleared, and both the writer and sstable would
close the file. This will result in a crash in
append_challenged_posix_file_impl::close(), which is not prepared to
be closed twice.
Another problem is that if exception happened before we reached that
construct, we still should close the writer. Currently we don't, so
there's no double close on the file, but that's a bug which needs to
be fixed and once that's fixed double close on _file will be even more
likely.
The fix employed here is to not keep files inside sstable object when
writing. As soon as the writer is constructed, it's the only owner of
the file.
Fixes#1764.
Message-Id: <1482428648-22553-1-git-send-email-tgrabiec@scylladb.com>
Range queries used to be performed sequentially and the shard performing
part of the read was reading state of the merger's memory accounter
directly. Now, they may be performed in parallel so it is safer to just
pass relevant data by value to the intersted shards so that they are not
reading something that another shard is modyfing at the same time.
Since query is done in parallel there is a chance of overread. However,
the parallelism is high only in sparsely populated tables and that's
when the overread is less serious problem.
Since a1cafed370 "storage_proxy: handle
range scans of sparsely populated tables" nonsingular range queries may
be performed in parallel on multiple shards. The consequence of this
that result may be added to the merger out of order. This requires more
complex logic for handling short reads.
As soon as mutation_result_merger gets a short read it starts to discard
all subsequently received results that are known to contain partitions
with larger keys.
Then when the final result is being prepared the merger may need to
combine and sorts results which ordering is not known. If at least one
of these results is a short one all partitions with larger keys are
removed.
Due to request being performed in parallel it is possible that even
though there was a short read the merger has got enough live data to
satisfy specified limits. If this has happened the short read flag is
not set on the final result.
We may want to change the default individual result size limit in the
future. If it is provided by the coordinator and not hardcoded in the
replicas this can be done without causing data query digest mismatches
or wasteful mutation query results.
Even if we are performing a digest query we should do proper result
memory accounting so that the result ends exactly in the same place that
it would if it was a data query. This is to avoid digest mismatches
between replicas.
It was the observation that ring_position_range_sharder doesn't support
wrapping ranges that started the nonwrapping_range madness, but that
class still has some leftover wrapping ranges. Close the circle by
removing them.
Message-Id: <20161123153113.8944-1-avi@scylladb.com>
Digest reads differ from data reads in a way that they do not really
consume any memory. We still want them to stop in the same place that
data reads would, but the per-shard semaphore shouldn't be updated by
them.
Original IDL generated code was hardcoded to always use bytes_ostream.
This patch makes the output stream a template parameter so that any
valid output stream can be used.
Unfortunately, making IDL writers generic requires updates in the code
that uses them, this is fixed in C++17 which would be able to deduce the
parameter in most cases.
For data queries it is very important that all replicas get limited in
the same place (this includes replicas returning only digest). That's
why they shouldn't be affected by per-shard result memory limit.
Moreover, we should make sure that individual memory limits are the
same, making the coordinator provide it for replicas which allow to
safely change it in the future.
Mutation queries are not as sensitive but it is still beneficial to make
sure that all replicas use the same individual limit.
If after reconciliation the coordinator ends up with no live rows and
short reads are allowed a retry may not make any progress if replicas
end their reads in the same place. The solution is to disallow short
reads on retries which are caused by final result having no live rows.
mutation_result_merger merges results from different shards and stops as
soon as a shard returned a short read or memory usage on the merging
shard is too high. However, it should never stop unless at least one
live rows is in the merged result.
* 'tgrabiec/calculate-hash-once-compaction' of github.com:cloudius-systems/seastar-dev:
sstables: Calculate key hash only once during compaction
tests: sstables: Add more test cases to tombstone_purge_test
db: Expose column_family::add_sstable
tests: sstables: Ensure timestamps are increasing
tests: sstables: Simplify tombstone_purge_test
- moved to seastar thread
- extracted sstable creation and validation logic
- reduced code duplication
- switched to mutation_reader assertions
- used result of compact_sstable() to locate the new sstable
- rather than setting gc timestamp in the past, bump the clock
before compacting
Some code paths were obtaining db_clock timestamp to only convert it
to gc_clock later. Avoid this. In the future we could make gc_clock
cheaper cause it has low precision.
Message-Id: <1482401190-2035-1-git-send-email-tgrabiec@scylladb.com>
Shared sstables will now be resharded in the same order to guarantee
that all shards owning a sstable will agree on its deletion nearly
the same time, therefore, reducing disk space requirement.
That's done by picking which column family to reshard in UUID order,
and each individual column family will reshard its shared sstables
in generation order.
Fixes#1952.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <87ff649ed24590c55c00cbb32bffd8fa2743e36e.1482342754.git.raphaelsc@scylladb.com>
Currently server-side timestamps use a clock with millisecond
precision. Timestamps have microsecond resolution, with lower bits
used to serialize mutations originating from given client.
Timestamps for column drops always use just the millisecond base. A
column drop which is executed after an insert may thus be given lower
timestamp than the insert, even when the two are serialized on the
client side over same connection.
Use microsecond precision to reduce chances of that event.
This is supposed to fix sporadic failures of
schema_test.py:TestSchema.drop_column_queries_test dtest.
Message-Id: <1482343119-27698-1-git-send-email-tgrabiec@scylladb.com>
Refs #1943.
* 'tgrabiec/optimize-bloom-filter' of github.com:cloudius-systems/seastar-dev:
db: Compute key hash once in partition_presence_checker
bloom_filter: Allow checking presence using pre-hashed key
db: Use incremental selector in partition_presence_checker
RPC messaging service is initialized before the Tracing service, so
we should prevent creation of tracing spans before the service is
fully initialized.
We will use an already existing "_down" state and extend it in a way
that !_down equals "started", where "started" is TRUE when the local
service is fully initialized.
We will also split the Tracing service initialization into two parts:
1) Initialize the sharded object.
2) Start the tracing service:
- Create the I/O backend service.
- Enable tracing.
Fixes issue #1939
Signed-off-by: Vlad Zolotarov <vladz@scylladb.com>
Message-Id: <1481836429-28478-1-git-send-email-vladz@scylladb.com>
If a node gets more MUTATION request that it can handle via RPC it will
stop reading from this RPC connection, but this will prevent it from
getting MUTATION_DONE responses for requests it coordinates because
currently MUTATION and MUTATION_DONE messages shares same connection.
To solve this problem this patches moves MUTATION_DONE messages to
separate connection.
Fixes: #1843
Message-Id: <20161201155942.GC11581@scylladb.com>
This reverts commit b81a57e8eb.
With exponential range scanning, we should now be able to survive
msb ignore bits of 12, which allows better sharding on large clusters.
When murmur3_partitioner_ignore_msb_bits = 12 (which we'd like to be the
default), a scan range can be split into a large number of subranges, each
going to a separate shard. With the current implementation, subranges were
queried sequentially, resulting in very long latency when the table was empty
or nearly empty.
Switch to an exponential retry mechanism, where the number of subranges
queried doubles each time, dropping the latency from O(number of subranges)
to O(log(number of subranges)).
If, during an iteration of a retry, we read at most one range
from each shard, then partial results are merged by concatentation. This
optimizes for the dense(r) case, where few partial results are required.
If, during an iteration of a retry, we need more than one range per
shard, then we collapse all of a shard's ranges into just one range,
and merge partial results by sorting decorated keys. This reduces
the number of sstable read creations we need to make, and optimizes for
the sparse table case, where we need many partial results, most of which
are empty.
We don't merge subranges that come from different partition ranges,
because those need to be sorted in request order, not decorated key order.
[tgrabiec: trivial conflicts]
Message-Id: <20161220170532.25173-1-avi@scylladb.com>
This patchset implements the multiple CQL3 statements relating to
materialized views, as well as ensuring other statements now take
materialized views into account. It also adds the necessary internal
data structures to hold materialized view metadata.
This patch adds a set of tests for materialized view schema
handling, complementing the dtests for the same feature.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch introduces the do_with_cql_env_thread() function, which
behaves like do_with_cql_env() except that it executes the
user-specified function in the context of a Seastar thread.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds an utility function that creates a raw select
statement from a set of columns and a where clause. It is intended to
be used to create the prepared select statement used by the view
class.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the alter_view_statement, which enables users to
change the properties of a materialized view.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the MATERIALIZED_VIEWS_FEATURE to the set of cluster
features and requires its presence to allow creating a view. This
ensures view schemas can be safely propagated across nodes.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch propagates the for_view argument, used by
statement_restrictions to ensure IS NOT NULL can be used when creating
a materialized view.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch checks for additional permissions when modifying a table
with views, since that update will require reading from the table and
writing into its views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch forbids dropping a column family if there are still views
associated with it, and also forbids dropping a view through the drop
table statement.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures that changes to a base table's schema
are reflected in that table's materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
To minimize code duplication, have query_processor use
do_with_parser() instead of manually creating the CqlParser.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds an utility function to rename a column occurring a
textual where clause. It is intended to change a view's where clause
when users alter the underlying base table.
To do this, we rely on functions that transform a textual where clause
into a set of relations, which allows to reliably rename the column.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds an utility function to rename an identifier
occurring in a cql3 relation. This function will be used when renaming
an identifier in a view's where clause.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch builds the mutations to announce a new view. Aside from
including the view schema, we include the base table mutations so
that a node is resilient against receiving create view mutations
before the base table create mutations.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch allows a view schema to be frozen. To unfreeze such a
schema, we add an is_view attribute to the schema idl.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch replaces the add_table_to_schema_mutation() function with
add_table_or_view_to_schema_mutation().
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts update_column_family from schema_tables into
database so it can be used when adding materialized views, in future
patches.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the drop_column_family() function to remove
a view schema from the list of views of its base table.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds code for parsing the views schema table upon init and
also ensures that when adding a view column family, that we add it to
its base table list of views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch moves some duplicate code into the
add_column_family_and_create_directory() function. It also saves some
superfluous keyspace lookups and readies the code to be used by
materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes read_table_mutations() so that it can now
read schemas from other tables besides the column families
schema table.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures we don't provide access to materialized views over
thrift. This includes preventing updates but also omitting them when
describing a keyspace.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds utility functions to keyspace_metadata to select only
the tables or only the views out of all the schemas.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the view class, which will contains functions related
to populating a view, either from the base table's write path or from
the view building mechanism which copies over already existing data in
the base table.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
The view_ptr class contains a schema_ptr known to represent a
materialized view. It is intended to be used by functions that require
such a schema, and thus obviate the need for the function to check for
schema::is_view().
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a view_info optional field to the schema. It's
presence indicates the schema represents a materialized view.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch fixes a regression introduced in 0518895, where we counted
one extra row per partition when it contained live, non static rows.
We also simplify the visitor logic further, since now we don't need to
count rows one by one. Also remove a bunch of unused fields.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1482234083-2447-1-git-send-email-duarte@scylladb.com>
Will allow us to calculate the hash once and use it on many filters
instead of calculating the hash for each filter separately.
Another change made is to avoid precomputing all indexes during filter
operations, and have for_each_index() template instead which invokes a
functor.
This reduces the number of sstables we need to check to only those
whose token range overlaps with the key. Reduces cache update
time. Especially effective with leveled compaction strategy.
Refs #1943.
Incremental selector works with an immutable sstable set, so cache
updates need to be serialized. Otherwise we could mispopulate due to
stale presence information.
Presence checker interface was changed to accept decorated key in
order to gain easy access to the token, which is required by
the incremental selector.
"nonwrapping_range<ring_position> and nonwrapping_range<token> are used
in many places. Let's make an alias for them to make it less verbose.
Also there is a query::partition_range in query-request.hh which is the alias of
nonwrapping_range<ring_position>. query::partition_range is used in
places not related to query at all. Let's unify the usage project wide."
* tag 'asias/repair_dht_token_range/v2' of github.com:cloudius-systems/seastar-dev:
Convert to use dht::partition_range_vector and dht::token_range_vector
dht: Introduce dht::partition_range_vector and dht::token_range_vector
Get rid of query::partition_range
Convert to use dht::partition_range
Convert to use dht::token_range
dht: Rename token_range to token_range_endpoints
dht: Introduce dht::token_range an dht::partition_range
std::vector<dht::partition_range> and std::vector<dht::token_range> are
used in a lot of places, introduce dht::partition_range_vector and
dht::token_range_vector as the alias.
nonwrapping_range<ring_position> and nonwrapping_range<token> are used
in many places. Let's make an alias for them to make it less verbose.
Also there is a query::partition_range in query-request.hh which is the alias of
nonwrapping_range<ring_position>. query::partition_range is used in
places not related to query at all. Let's unify the usage project wide.
"In 7c873f0d (repair: Reduce unnecessary streaming traffic), we optimize
in cases when 1) all the remote nodes has the same checksum and 2) local node
has zero checksum.
In this series, we make the optimization more generec and cover more cases."
* tag 'asias/repair/node_reducer/v3' of github.com:cloudius-systems/seastar-dev:
repair: Reduce unnecessary streaming traffic even more
repair: Add hash specialization for partition_checksum
"This patchset ensures the partition limit is enforced at
the storage_proxy level. Uppers layers like the pager may
already be depending on this behavior."
* 'enforce-row-limit/v3' of https://github.com/duarten/scylla:
query_pagers: Don't trim returned rows
select_statement: Don't always trim result set
query_result_merger: Limit rows
mutation_query: to_data_query_result enforces row limit
"This patchset ensures the partition limit is enforced at
the storage_proxy level. To achieve this, we add the partition
count to query::result, and allow the result_merger to trim
excess partitions."
* 'enforce-partition-limit/v3' of https://github.com/duarten/scylla:
storage_proxy: Decrease limits when retrying command
storage_proxy: Don't fetch superfluous partitions
query::result: Add partition count
column_family: Use counters in query::result::builder
query_result_builder: Use the underlying counters
mutation_partition: Count partitions in query_compacted
mutation_partition: Remove tabs in query_compacted
query::result::builder: Add partition count
query_result_merger: Limit partitions
A case could be made that we should have counters for them no matter
what, since it can help us reason about the distribution of memory among
the groups. But with the hierarchy being broken in 1.5 it becomes even
more important. Now by looking solely at dirty, we will have no idea
about how much memory we are using in those groups.
After this patch, the dirty_memory_manager will register its metrics
for the 3 groups that we have, and the legacy names will be used to show
totals.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <0d04ca4c7e8472097f16a5dc950b77c73766049e.1481831644.git.glauber@scylladb.com>
"Original, naive db::make_streaming_reader() implementation created a set
of memtable and sstable readers for every partition range. This caused
bad interaction with the code limiting sstable readers concurrency and
was suboptimal.
This series introduces multi range mutation reader that takes mutation
source and a sorted, disjoint vector of ranges. It creates only a single
set of memtable and sstable readers and fast forwards it to the next
range once the current one is completed."
* 'pdziepak/multi-range-reader/v1' of github.com:cloudius-systems/seastar-dev:
db: use multi range reader for streaming readers
dht: describe split_range[s]_to_shards() guarantees
repair: remove outdated fixme
test/mutation_reader_test: add multi_range_reader test
tests/mutation_reader: extract key creation code
mutation_reader: add multi_range_reader
A naive approach was to create a set of readers for each range and pass
them all to combining reader. This however performed badly if the number
of ranges was high.
The solution is to use multi range reader which uses only a single set
of readers and fast forwards from range to range when necessary. This
adds another requirement that the ranges passed to
make_streaming_reader() are sorted and disjoint.
We are going to require these functions to return sorted and disjoint
ranges. They already do so (provided that the input ranges are sorted
and disjoint), but if the guarantee is not explicitly stated it may
disappear some day.
So far, the only way to combine outputs of multiple readers was to use
combining reader. It is very general and, in particular, supports case
when the readers emit mutations from overlapping ranges.
However, we have cases (e.g. streaming) when we need to read from
several disjoint ranges. Combining reader is a suboptimal solution as it
requires to creating a reader for each range and ignores the fact that
they do not overlap.
This patch introduces multi_range_mutation_reader which takes a
mutation_source and a sorted set of disjoint ranges. Internally, it uses
mutation_reader::fast_forward_to() to move to the next range once the
current one is completed.
Since storage_proxy::query() now respects the read_command limits, we
can remove the trimming logic from query_pagers.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Trimming the result set is only needed when the query contains an "IN"
relation, an ORDER BY clause, and defines a limit, which is the case
where we query different ranges concurrently. We don't use the
result_merger to trim since we first need to reorder the rows.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch makes the row limit enforced by the storage_proxy layer.
It adds a row limit to the query_result_merger, useful when merging
results for concurrent queries.
More importantly, it provides guarantees that upper layers may be
relying on implicitly (e.g., the paging code).
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes mutation_query::to_data_query_result() so that it
enforces the row limit alongside the partition limit and the
per-partition limit.
In the following patch, we'll enforce the row limit in an upper layer,
but this lets us optimize the case where only when replica replies.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes a read_command's limits when retrying it, so that
we don't ask for more rows than necessary.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures we keep track of how many partitions we've queried
so we don't ask for more than the number we need.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a partition count to query::result, filled by the
query::result::builder. The partition count is present whenever the
result carries data, being absent only for the case where the result
contains only a digest.
We also ensure that counts are present for an empty query::result.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes column_family::query() to use the counters in the
builder to determine how many partitions and rows to ask for and also
to implement the stop condition. This saves a continuation to do the
bookkeeping, and allows us to remove data_query_result.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the query_result_builder to use the counters
provided by the query::result::builder. It also ensures they are kept
current.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes mutation_partition::query_compacted() to count the
number of partitions written to the underlying writer.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a partition count to the query::result::builder. It is
intended to be incremented by users, and later used to build a
query::result.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a partition limit to the query_result_merger, useful
when merging results for concurrent queries. This change also makes
the partition limit enforced by the storage_proxy layer, no changes
being needed by the upper layers, namely the Thrift interface.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
We saw the message twice for the same feature check. This is a bit
confusing.
INFO 2016-12-15 11:26:23,993 [shard 0] gossip - Checking if need_features {RANGE_TOMBSTONES} in features {}
INFO 2016-12-15 11:26:23,993 [shard 0] gossip - Checking if need_features {RANGE_TOMBSTONES} in features {}
INFO 2016-12-15 11:26:23,993 [shard 0] gossip - Checking if need_features {LARGE_PARTITIONS} in features {}
INFO 2016-12-15 11:26:23,993 [shard 0] gossip - Checking if need_features {LARGE_PARTITIONS} in features {}
This is because
ss._range_tombstones_feature = gms::feature(RANGE_TOMBSTONES_FEATURE);
ss._large_partitions_feature = gms::feature(LARGE_PARTITIONS_FEATURE);
The first message is printed when gms::feature(RANGE_TOMBSTONES_FEATURE)
is constructed. The second message is printed when the
ss._range_tombstones_feature is copy-constructed.
Add the helper function to enable the a feature and log the feature is
enabled.
When a feature is enabled, we see
INFO 2016-12-15 11:29:32,443 [shard 0] gossip - Feature LARGE_PARTITIONS is enabled
INFO 2016-12-15 11:29:32,443 [shard 0] gossip - Feature RANGE_TOMBSTONES is enabled
in the log.
"This series makes Scylla limit size of query results it produces in case they
grow unreasonably large. This is possible because CQL paging queries do not
guarantee that the returned page is going to have page_size rows and pages
smaller than tha *do not* indicate end of stream. Non-paged queries and Thrift
requests do not have such flexibility and they also get all the requested data
(though their memory usage is still accounted for and may limit paged queries).
There is a maximum result size (1 MB) and all results builders will stop after
reaching it. Moreover, there is a per-shard limitation on the amount of memory
used by all results combined (10%). To avoid tiny results a query has to
reserve (wait if necessary) 4 kB before starting executing, after that it can
consume more memory without any additional waiting provided it is below
individual and shard-local limits.
Enabling the cluster to return less rows than requested also means some changes
for the coordinator. Firstly, if it receives such short result from a replica
retrying it with a larger limit obviously makes no sense whatsoever. Instead,
in such cases the coordinator removes the clustering rows it has incomplate
information about and sends short result back to the client. Moreover, even
if no replica returned short response reconciliation may have made it so. In
this case, the coordinator do not necessairly need to retry the query as well.
Unfortunately, with the current implementation short responses ruin data
queries since they will cause a digest mismatch.
Three new metrics were added:
* database_bytes_total_result_memory -- total memory used by query results
* database_total_operations_short_data_queries -- data queries that were
limited by size, particulary bad as it basically forces coordinator to
retry them as mutation queries
* database_total_operations_short_mutation_queries -- mutation queries limited
by size"
* 'pdziepak/short-paged-reads/v4' of github.com:cloudius-systems/seastar-dev:
storage_proxy: clean up after primary_key introduction
cql3: allow short reads with paged queries
storage_proxy: handle intentional short reads
storage_proxy: make sure coordinator has complete data
storage_proxy: honour partition limit
storage_proxy: use cmd limits to determine that replica reached end
db: add metrics for short reads and memory used for results
data_query: limit result size
mutation_query: limit result size
db: create result_memory_accounters when starting query
query_builder: add partition_slice getter
reconcilable_result: keep result_memory_tracker object
mutation_compactor: honour stop_iteration from consumers
db: add result_memory_limiter
query: add result size limiter
reconcilable_result: properly propagate short_read flag
query_pagers: handle short reads properly
query: allow short reads
serializer_impl: add serializer for bool_class<Tag>
primary_key was introduced as a replacement for
std::pair<dht::decorated_key, std::optional<clustering_key>>. In order
to simplify patch introducing its fields were named 'first' and
'second'. This patch changes the names to something less useless,
removes old row_address alias and removes is_missing_rows() in favour of
primary_key::less_compare_clustering comparator.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
If the result is going to be too large the replica may decide to make it
shorter and coordinator should handle this properly (i.e. do not retry).
Moreover, coordinator could avoid some retries by setting the short_read
flag itself.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
got_incomplete_information() ensures that the coordinator has received
all required data from all replicas.
(see 77dbe3c12f "storage_proxy: fix
reconciliation with limits" for the examples when that may not be the
case).
However, this function is called only if reconciled result has at least
as much rows as the user asked for. This was correct when we had only
total row limit: if the result was shorter than that either all replicas
sent all data they have or the coordinator will retry anyway. However,
since then we got partition limit and per partition row limit and a
request may be limited by one of these while being still below the total
row limit.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
At the moment the coordinator does not care much for the partition
limit. In particular it doesn't check whether after reconciliation the
result still contains enough partitions.
This patch makes it honour the partition limit and increase it in the
retried queries if necessary.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Coordinator may retry a query with larger limits. However, code
determining whether replica has no more data always used the original
limits. This may cause a livelock.
For example, consider cluster having the following partitions (deletions
cover live cells):
node1:
pk=0, v=0
pk=1, v=1
node2
delete pk=0
delete pk=1
pk=2, v=2
pk=3, v=3
Now, if there is a query SELECT * FROM cf LIMIT 2 the first node is
going to send partitions 0 and 1 while second node is going to send 2
and 3 + tombstones for 0 and 1. The coordinator will decide that it
needs to retry the request with larger row limit since node1 may have
some information about partitions 2 and 3 that are newer than what node2
has sent.
However, when the second response arrives node1 will still sent only two
rows since it has no more data. Because the coordinator uses original
row limit it will not notice that this node reached the end and we are
going to get another retry without making any progress.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
This pach ensures than when we start executing a query a minimum result
size is reserved from result_memory_limiter.
Moreover, range queries need a way of merging memory usage information
from different shards.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
This patch introduces an infrastrucutre for limiting result size.
There is a shard-local limit which makes sure that all results combined
do not use more than 10% of the shard memory.
There is also an invidual limit which restricts a result to 4 MB.
In order
In order to avoid sending tiny results there is minimum guaranteed size
(4 kB), which the query needs to reserve before it starts producing the
result.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
reconcilable_result can be merged with another or transformed into
query::result. Make sure that short_read information is never lost.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Currently, the paging implementation assumes that the server retunrs
either as many rows as it was asked for all reached the end. Soon,
that's not going to be true so instead of making any assumptions about
the number of the rows returned use the new "short read" flag to
determine whether there is going to be more data.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
When paging is used the cluster is allowed to return less rows than the
client asked for. However, if such possibility is used we need a way of
telling that to the coordinator and the paging implementation so that
they can differentiate between short reads caused by the replica running
out of data to sent and short reads caused by any other means.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
The test assumed that mutations added to the commitlog are visible to
reads as soon as a new segment is opened. That's not true because
buffers are written back in the background, and new segment may be
active while the previous one is still being written or not yet
synced.
Fix the test so that it expectes that the number of mutations read
this way is <= the number of mutations read, and that after all
segments are synced, the number of mutations read is equal.
Message-Id: <1481630481-19395-1-git-send-email-tgrabiec@scylladb.com>
"The current criteria for memtable flush is not being respected. The
problem is demonstrated to happen when the dirty memory group is over
limit, and so is the system table extra allowance. In that situation,
both the normal region and the system table region will be under
pressure and try to flush.
More specifically, because the normal region inherits from the system
region, if the normal region is under pressure (over the soft limit
threshold), the system region will certainly be as well, even though it
has an extra allowance. This is because after virtual dirty, we start
blocking when we reach half the region, but memory itself can grow up to
100 % of the region. So the total amount of memory used will be
certainly bigger than the system pressure threshold, which is now 50 %
plus the allowance.
To fix that, this patch reworks the flush logic so that the regions are
not dependent on each other.
Fixes#1918"
* 'flush-criteria-v6' of github.com:glommer/scylla:
config: get rid of memtable_total_space
database: rework dirty memory hierarchy
system keyspace: write batchlog mutation in user memory
database: remove flush_token
database: abstract pressure condition notification
database: encapsulate semaphore_units into a flush_permit
database: remove friendship declaration
database: simplify flush_one
database: make memtable_list aware in cases it can't flush
In 7c873f0d (repair: Reduce unnecessary streaming traffic), we optimize
in cases when 1) all the remote nodes has the same checksum and 2) local node
has zero checksum.
In this patch, we make the optimization more generec and cover more cases.
1) With RF = 3, 3 nodes cluster, rm data on node3 then run repair on node2
Before:
INFO 2016-12-09 16:24:31,961 [shard 0] repair - Found differing range (-4091524285777924069, -4086237930244473115]
on nodes {127.0.0.3, 127.0.0.1}, in = {127.0.0.3, 127.0.0.1}, out = {127.0.0.3, 127.0.0.1}
INFO 2016-12-09 16:24:31,963 [shard 0] repair - Found differing range (-609511120964672970, -605253169726090861]
on nodes {127.0.0.1, 127.0.0.3}, in = {127.0.0.1, 127.0.0.3}, out = {127.0.0.1, 127.0.0.3}
INFO 2016-12-09 16:24:31,964 [shard 0] repair - Found differing range (-7655412157560911259, -7652234653747163387]
on nodes {127.0.0.3, 127.0.0.1}, in = {127.0.0.3, 127.0.0.1}, out = {127.0.0.3, 127.0.0.1}
INFO 2016-12-09 16:24:31,965 [shard 0] repair - Found differing range (-4133815130045531703, -4128528774512080749]
on nodes {127.0.0.3, 127.0.0.1}, in = {127.0.0.3, 127.0.0.1}, out = {127.0.0.3, 127.0.0.1}
INFO 2016-12-09 16:24:31,967 [shard 0] repair - Found differing range (-605253169726090861, -600995218487508751]
on nodes {127.0.0.1, 127.0.0.3}, in = {127.0.0.1, 127.0.0.3}, out = {127.0.0.1, 127.0.0.3}
INFO 2016-12-09 16:24:31,968 [shard 0] repair - Found differing range (438510347741343837, 441475345714861354]
on nodes {127.0.0.1, 127.0.0.3}, in = {127.0.0.1, 127.0.0.3}, out = {127.0.0.1, 127.0.0.3}
After:
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-660606535827658284, -656348584589076175]
on nodes {127.0.0.1, 127.0.0.3}, in = {}, out = {127.0.0.3}
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-4234255885181099833, -4228969529647648879]
on nodes {127.0.0.3, 127.0.0.1}, in = {}, out = {127.0.0.3}
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-4228969529647648879, -4223683174114197925]
on nodes {127.0.0.3, 127.0.0.1}, in = {}, out = {127.0.0.3}
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-4223683174114197925, -4218396818580746971]
on nodes {127.0.0.3, 127.0.0.1}, in = {}, out = {127.0.0.3}
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-7728494745277112315, -7725317241463364443]
on nodes {127.0.0.3, 127.0.0.1}, in = {}, out = {127.0.0.3}
INFO 2016-12-09 16:30:29,204 [shard 0] repair - Found differing range (-720217853167807818, -715959901929225709]
on nodes {127.0.0.1, 127.0.0.3}, in = {}, out = {127.0.0.3}
Before, we need to fetch data from both node 1 and node 3 and send data back to node 1 and node 3, i.e., 2 IN, 2 OUT
After, we only need to fetch data from node 3, i.e. 0 IN, 1 OUT
We saved 3X traffic, with higher RF, we can save even more.
2) With RF = 3, 3 nodes cluster, rm data on node3 then run repair on node3
Before:
INFO 2016-12-09 16:20:11,448 [shard 0] repair - Found differing range (-8533861887892628919, -8052600134279395253]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
INFO 2016-12-09 16:20:11,465 [shard 0] repair - Found differing range (7190719703944308372, 7692358524564683543]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
INFO 2016-12-09 16:20:11,486 [shard 0] repair - Found differing range (-3305328316052774469, -2671876682129336880]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
INFO 2016-12-09 16:20:11,494 [shard 0] repair - Found differing range (-2190610927722759275, -1305178847032904465]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:20:11,518 [shard 0] repair - Found differing range (-4747032371925842389, -4070378863644120252]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:20:11,519 [shard 0] repair - Found differing range (-1137497074548854552, -592479316010344531]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
After:
INFO 2016-12-09 16:29:22,433 [shard 0] repair - Found differing range (67885601051654285, 447405341661896387]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:29:22,454 [shard 0] repair - Found differing range (-2190610927722759275, -1305178847032904465]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:29:22,473 [shard 0] repair - Found differing range (2523396860109747637, 3083778975065200884]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:29:22,474 [shard 0] repair - Found differing range (-3305328316052774469, -2671876682129336880]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
INFO 2016-12-09 16:29:22,487 [shard 0] repair - Found differing range (-4747032371925842389, -4070378863644120252]
on nodes {127.0.0.2, 127.0.0.1}, in = {127.0.0.2}, out = {}
INFO 2016-12-09 16:29:22,493 [shard 0] repair - Found differing range (-1137497074548854552, -592479316010344531]
on nodes {127.0.0.1, 127.0.0.2}, in = {127.0.0.1}, out = {}
This shows the new more generic methods covers the optimization we had before as well.
Issue #1918 describes a problem, in which we are generating smaller
memtables than we could, and therefore not respecting the flush
criteria.
That happens because group sizes (and limits) for pressure purposes, and
the the soft threshold is currently at 40 %. This causes system group's
soft threshold to be way below regular's virtual dirty limit and close
to regular group's soft threshold. The system group was very likely to
become under soft pressure when regular was because writes to regular
group are not yet throttled when they cross both soft thresholds.
This is a direct consequence of the linear hierarchy between the regions
and to guarantee that it won't happen we would have acqire the semaphore
of all ancestor regions when flushing from a child region. While that
works, it can lead to problems on its own, like priority inversion if
the regions have different priorities - like streaming and regular, and
groups lower in the hierarchy, like user, blocking explicit flushes
from their ancestors
To fix that, this patch reorganizes the dirty memory region groups so
that groups are now completely independent. As a disadvantage, when
streaming happen we will draw some memory from the cache, but we will
live with it for the time being.
Fixes#1918
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Batchlog is a potentially memory-intensive table whose workload is
driven by user needs, not system's. Move it to the user dirty memory
manager.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We had a flush_token structure in addition to the flush_permit because
we needed to keep a pointer to the dirty_memory_manager and apply
changes to the region group upon the region destruction. Since Tomek's
latest series, this is no longer needed and now this structure doesn't
have a place in the world anymore. Simplify the code by removing it.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Done in a separate patch to reduce clutter in the main patch.
Soon we'll be testing for one more condition.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We will soon need to hold more than a semaphore_units<> object per
flush, potentially.
Preparation patch for that.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
flush_one has to make sure that we're using the correct
dirty_memory_manager object, because we could be flushing from a region
group different than the one the flush request originated.
It's simpler to just assume flush_one will be dealing with the right
object, and use a different object instead of "this" when calling it.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Some of our CFs can't be flushed. Those are the ones who are not marked
as having durable writes. We treat them just the same from the point of
view of the flush logic, but they provide a function that doesn't do
anything and just returns right away.
We already had troubles with that in the past, and that also poses a
problem for an upcoming patch reworking the flush memtable pick
criteria.
It's easier, simpler, and cleaner, to just make the memtable_list aware
it can't flush. Achieving that is also not very complicated: we just
need a special constructor that doesn't take a seal function and then we
make sure that it is initialized to an empty std::function
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Currently, we weren't completing a query as early as possible if it
reached the partition limit, we instead had to wait until reaching the
end of the specified partition ranges. This patches fixes that by
including a check to the partition limit in the termination condition.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20161213114559.26438-1-duarte@scylladb.com>
column_mapping is not safe to access across shards, because data_type
is not safe to access. One of the manifestation of this is that
abstract_type::is_value_compatible_with() always fails if the two
types belong to different shards.
During replay, column_mapping lives on the replaying shard, and is
used by converting_mutation_partition_applier against the schema on
the target shard. Since types in the mapping will be considered
incompatible with types in the schema, all cells will be dropped.
Fix by using column_mapping in a safe way, by copying it to the target
shard if necessary. Each shard maintains its own cache of column
mappings.
Fixes#1924.
Message-Id: <1481310463-13868-1-git-send-email-tgrabiec@scylladb.com>
* seastar 0773e98...6fbd792 (2):
> tls: Only run our "verify" function in client session
> Merge "Clean the metric definition" from Amnon
Includes patch from Amnon adjusting the metrics registration due to seastar
API changes.
* seastar 0a74317...0773e98 (6):
> tls: Add support for client cetrificate verification & priority strings
> semaphore: add consume_units
> semaphore: add available_units()
> thread: check need_preempt for threads in a scheduling group as well
> tutorial: fix semaphore example, and text
> stop_iteration: add && and || operators
"This series:
- We can make reader with ranges
- Fix possible use after free of 'si'
- Streaming ranges now are sorted and merged
- Fix shard_begin shard_end end loop in both streaming and repair"
A range now alternates between different shards: the first part of the
range goes to shard X, the next to shard X+1, but after a while we go
back to shard X. So we can't do a simple loop between shard_begin and
shard_end.
Fix by using the newly introduced dht::split_range_to_shards
Use the cf.make_streaming_reader with ranges to simplify the code a bit.
Now that we have the new interface to make readers with ranges, we can
simplify the code a lot.
1) Less readers are needed
before: number of ranges of readers
after: smp::count readers at most
2) No foreign_ptr is needed
There is no need to forward to a shard to make the foreign_ptr for
send_info in the first phase and forward to that shard to execute the
send_info in the second phase.
3) No do_with is needed in send_mutations since si now is a
lw_shared_ptr
4) Fix possible user after free of 'si' in do_send_mutations
We need to take a reference of 'si' when sending the mutation with
send_stream_mutation rpc call, otherwise:
msg1 got exception
si->mutations_done.broken()
si is freed
msg2 got exception
si is used again
The issue is introduced in dc50ce0ce5 (streaming: Make the mutation
readers when streaming starts) which is master only, branch 1.5 is not
affected.
Allow to make a streaming reader with a vector of ranges in addition to
a single range. This will be used soon in following streaming patch.
We can make the reader more efficient later.
This patch fixes a typo in i_partitioner::tri_compare() where we were
using std::max instead of std::min, thus avoiding accessing random
memory and getting random results.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20161211165043.17816-1-duarte@scylladb.com>
This patch adds uuid file support for ubuntu system. It also split the
behaviour between restart and daily checks. The first run in r mode and
the second in d mode.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Allows scylla-housekeeping getting the uuid from a file instead of the
command line.
If the file is missing no uuid will be used.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
"Function to calculate maximum purgeable timestamp is made 10 times faster when
compacting sstables overlap with 10% of all sstables.
That's possible with an incremental selector that will incrementally select
sstables based on key being compacted.
Currently, we iterate through all non-compacting sstables and consult their
bloom filter to determine max purgeable timestamp, and that will be very
expensive for compactions that are frequently deciding whether or not to purge
tombstones."
* 'filter_overhead_fix_v4' of github.com:raphaelsc/scylla:
compaction: reduce bloom filter overhead with incremental selector
tests: add test for sstable set's incremental selector
sstable_set: introduce incremental selector
compatible_ring_position: add function to return token
The procedure to calculate max purgeable timestamp is optimized
by only visiting sstables that overlap with key being currently
compacted. That's done using incremental sstable selector.
Function to calculate maximum purgeable timestamp is made 10 times
faster when compacting sstables overlap with 10% of all sstables.
Fixes#1322.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Incrementally select sstables from sstable set using token
in ascending order.
For leveled strategy, it returns all sstables that belong
to current interval. For other strategies, it just return
all sstables from the set.
Useful for compaction which needs all sstables that overlap
with key being currently compacted to calculate maximum
purgeable timestamp.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
The semaphore future may be unavailable for many reasons. Specifically,
if the task quota is depleted right between sem.wait() and the .then()
clause in get_units() the resulting future won't be available.
That is particularly visible if we decrease the task quota, since those
events will be more frequent: we can in those cases clearly see this
counter going up, even though there aren't more requests pending than
usual.
This patch improves the situation by replacing that check. We now verify
whether or not there are waiters in the semaphore.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <113c0d6b43cd6653ce972541baf6920e5765546b.1481222621.git.glauber@scylladb.com>
When row cache is disabled, update_cache() will do nothing to the
memtable. Active readers may keep the memtable alive for unbounded
amount of time, preventing it from going away. This doesn't play well
with virtual dirty accounting. Soon before calling update_cache(), the
memory which was subtracted during flush is added back to the amount
of virtual dirty memory. If there was write pressure all along, we
will be at the dirty memory limit. When we give back subtracted memory
this will put virtual dirty way above the limit. This will stall all
writes until another memtable flush drags virtual dirty down or
readers finally release the memtable. We want to prevent upward
jumps of virtual dirty.
First part of the fix is to ensure that as long as the memtable's
region is in the dirty group, we will not revert flushed memory. This
must happen synchronously from region's memory being removed from the
group in order to prevent upward virtual dirty jumps. To make this
easier, tracking of flushed memory was moved to the memtable object.
Another part of the fix is to gradually clear the memtable when cache
is disabled in a similar fashion as when it's moved to cache. This
ensures that the actual memory held by memtable's region is released
sooner than it dies.
Refs #1879
"Due to my misreading of Cassandra code, I thought it would ignore new
components in the Statistics component; however, it doesn't, and the change
(introduced in bdd11648ac ("sstables: add
intra-node sharding metadata") breaks sstable2json and likely any
Cassandra code that touches sstables.
To fix, move the sharding data into a new component ("Scylla.db"), which
Cassandra does ignore. The new component is designed to be extensible so
we don't experience the same issue later on."
* tag 'asias/repair/subranges/refactor_fix/v1' of github.com:cloudius-systems/seastar-dev:
repair: Limit the number of sub ranges
repair: Use estimated_keys_for_range in repair_cf_range
repair: Extract the target_partitions into repair_info class
repair: Put request_transfer_ranges into repair_info class
repair: Introduce check_failed_ranges helper
repair: Introduce do_streaming helper
repair: Make the neighbors const reference
repair: Introduce repair_info
repair: Attach the repair id in the stream plan name
The problem is that replay will unlink any segments which were on disk
at the time the replay starts. However, some of those segments may
have been created by current node since the boot. If a segment is part
of reserve for example, it will be unlinked by replay, but we will
still use that segment to log mutations. Those mutations will not be
visible to replay after a crash though.
The fix is to record preexisting segents before any new segments will
have a chance to be created and use that as the replay list.
Introduced in abe7358767.
dtest failure:
commitlog_test.py:TestCommitLog.test_commitlog_replay_on_startup
Message-Id: <1481117436-6243-1-git-send-email-tgrabiec@scylladb.com>
- Add "coordinator" and "replica" categories
- Use a new seastar/metrics_registration framework
* 'rearrange-storage-proxy-stats-v4' of github.com:cloudius-systems/seastar-dev:
service::storage_proxy: rework the collectd counters registration
service/storage_proxy: regroup collectd statistics
The Cassandra derived sstable tools (and likely Cassandra itself) object to
a new sub-component in the Statistics component; create a new Scylla
component instead to host this data.
Allow declaring discriminated unions (with an enum type as the
discriminant and any sstable serializable type as a value) and sets
of these unions, with the disciminant as the key. Parsers and writers
are auto-generated.
Currently housekeeping timer won't be reset when we restart scylla-server.
We expect the service to be run at each start, it will be consistent with
upstart script in Ubuntu 14.04
When we restart scylla-server, housekeepting timer will also be restarted,
so let's replace "OnBootSec" with "OnActiveSec".
Fixes: #1601
Signed-off-by: Amos Kong <amos@scylladb.com>
Message-Id: <a22943cc11a3de23db266c52fd476c08014098c4.1480607401.git.amos@scylladb.com>
To reduce duplicated code and simplified scripts introduce scylla_lib.sh
for shellscripts which provides functions to classify distributions,
and load all sysconfig files.
This also fixes script bugs to misdetect Debian and RHEL.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1480667672-9453-2-git-send-email-syuu@scylladb.com>
Since Ubuntu 15.10/16.04 still uses Upstart to manage GUI session (not as init), when we directly launch Scylla on Ubuntu's GUI Terminal(not using systemctl or initctl), raise(SIGSTOP) mistakenly calls (Because GUI session has "UPSTART_JOB" environment variable, won't happen when running Scylla as systemd service).
To avoid this, we need to verify UPSTART_JOB == "scylla-server".
If it's part of GUI session UPSTART_JOB has to be "unity7", we need to avoid raise(SIGSTOP) in that case.
Fixes#1199
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1480620421-28967-1-git-send-email-syuu@scylladb.com>
So that memory is released gradually (impacting latency less) and
sooner than when memtable is destroyed. Active readers may keep the
memtable alive for unbounded amount of time.
Refs #1879
When memtable is flushing, it subtracts _flushed_memory from groups's
size to gradually allow more writes. Ideally _flushed_memory would be
equal to region's size when flush ends, so the group's size would
reach zero. When the memtable and its region are gone the group size
should remain the same as after the flush. This is ensured by adding
back _flushed_memory to group's size right before the region is
removed from the group.
Calling clear() before region is removed from the group breaks the
accounting because it will shrink the region, but will not affect the
amount of memory subtracted due to _flushed_memory. So group's size
would decrease more than we want (twice the region's size). The fix is
to change clear() so that it reverts _flushed_memory by the amount by
which the region size is reduced. This will keep the groups's size
constant as long as _flushed_memory > 0.
The implementation assumes that memtable's region group is owned by
dirty_memory_manager, and tries to obtain a reference to it like this:
boost::intrusive::get_parent_from_member(_region.group(), &dirty_memory_manager::_region_group));
This is undefined behavior when the region's group does not come from
dirty manager. It's safer to be explicit about this dependency by
taking a reference to dirty_memory_manager in the constructor.
If exception is triggered early in boot when doing an I/O operation,
scylla will fail because io checker calls storage service to stop
transport services, and not all of them were initialized yet.
Scylla was failing as follow:
scylla: ./seastar/core/sharded.hh:439: Service& seastar::sharded<Service>::local()
[with Service = gms::gossiper]: Assertion `local_is_initialized()' failed.
Aborting on shard 0.
Backtrace:
0x000000000048a2ca
0x000000000048a3d3
0x00007fc279e739ff
0x00007fc279ad6a27
0x00007fc279ad8629
0x00007fc279acf226
0x00007fc279acf2d1
0x0000000000c145f8
0x000000000110d1bc
0x000000000041bacd
0x00000000005520f1
0x00007fc279aeaf1f
Aborted (core dumped)
Refs #883.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Signed-off-by: Asias He <asias@scylladb.com>
Message-Id: <963f7b0f5a7a8a1405728b414a7d7a6dccd70581.1479172124.git.asias@scylladb.com>
A range is diveded into N sub ranges so that each sub range contains 100
partitions. So N depends on the number of partitions in that range. N
can grow unbounded and the memory usage of vector to hold these sub
ranges can go unbouded.
Limit the max number of sub ranges a range can divided into.
The downside is that the limited sub range will make we include more
partitions in the checksum.
Fixes#1917
Use the new seastar's metrics_registration framework:
- Change the registration syntax.
- Add a long description for each counter.
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Instead of putting all statistics under the same "storage_proxy" category
separate them into 2 groups according to where the corresponding counters
are updated:
- "storage_proxy_replica"
- "storage_proxy_coordinator"
Fixes#1763
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
When requests hit the commitlog, each of them will be assigned a replay
position, which we expect to be ordered. If reorders happen, the request
will be discarded and re-applied. Although this is supposed to be rare,
it does increase our latencies, specially when big requests are
involved. Processing big requests is expensive and if we have to do it
twice that adds to the cost.
The commitlog is supposed to issue replay positions in order, and it
coudl be that the code that adds them to the memtables will reorder
them. However, there is one instance in which the commitlog will not
keep its side of the bargain.
That happens when the reserve is exhausted, and we are allocating a
segment directly at the same time the reserve is being replenished. The
following sequence of events with its deferring points will ilustrate
it:
on_timer:
return this->allocate_segment(false). // defer here // then([this](sseg_ptr s) {
At this point, the segment id is already allocated.
new_segment():
if (_reserve_segments.empty()) {
[ ... ]
return allocate_segment(true).then ...
At this point, we have a new segment that has an id that is higher than
the previous id allocated.
Then we resume the execution from the deferring point in on_timer():
i = _reserve_segments.emplace(i, std::move(s));
The next time we need to allocate a segment, we'll pick it from the
reserve. But the segment in the reserve has an id that is lower than the
id that we have already used.
Reorders are bad, but this one is particularly bad: because the reorder
happens with the segment id side of the replay position, that means that
every request that falls into that segment will have to be reinserted.
This bug can be a bit tricky to reproduce. To make it more common, we
can artificially add a sleep() fiber after the allocate_segment(false)
in on_timer(). If we do that, we'll see a sea of reinsertions going on
in the logs (if dblog is set to debug).
Applying this patch (keeping the sleep) will make them all disappear.
We do this by rewriting the reserve logic, so that the segments always
come from the reserve. If we draw from a single pool all the time, there
is no chance of reordering happening. To make that more amenable, we'll
have the reserve filler always running in the background and take it out
of the timer code.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <49eb7edfcafaef7f1fdceb270639a9a8b50cfce7.1480531446.git.glauber@scylladb.com>
The information (last compacted keys) is lost after node is restarted
or schema is updated, which causes strategy to be rebuilt.
We need it for strategy to guarantee uniform distribution of token
range across sstables, or we could end up with 1 sstable of level L
overlapping with lots of sstables of level L+1, and that results in
a compaction of undesired length.
That information can be generated from scratch by getting last key
of newest sstable in each level > 0.
Fixes#1906.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <35ebd15977d5a8418239febb160c796cdc0e98fa.1480533805.git.raphaelsc@scylladb.com>
Streaming memtable have a delayed mode where many flushes are coalesced
together into one, with the actual flush happening later and propagated
to all the previous waiters.
However, the timer that triggers the actual flush was not using the
newly introduced flush infrastructure. This was a minor problem because
those flushes wouldn't try to take the semaphore, and so we could have
many flushes going on at the same time.
What was a potential performance issue became a correctness issue when
we moved the reversal of the dirty memory accounting out of
revert_potentially_cleaned_up_memory() into remove_from_flush_manager().
Since the latter is only called through the flush infrastructure, it
simply wasn't called. So the deferral of the reversal exposed this bug.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <0d5755375bc27524b8cfb9970c76d492b14d9eea.1480522742.git.glauber@scylladb.com>
This fix splits build_ami.sh --repo to three different options:
--repo-for-install is for Scylla package installation, only valid
during AMI construction.
--repo-for-update will be stored at /etc/yum.repos.d/scylla.repo, to
receive update package on AMI.
--repo is both, for installation and update.
Fixes#1872
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1480438858-6007-1-git-send-email-syuu@scylladb.com>
"The goal of this series is to prevent unbounded memory use
in cases when requests are timing out. Write requests which timed
out may still occupy memory for a while because of local mutation
application. This memory is not accounted for and can build up.
First part of the fix changes local mutation application so that it times out
at about the same time as the request handler. Then the life
time of the request handler is extended to cover any background activity
of that request which hasn't timed out yet. This has two main effects:
(1) by timing out local writes we prevent build up of background activity
for timed out requests
(2) we ensure that memory used by background activity is not left
behind unaccounted for. This will prevent CQL server from admitting
more requests than memory usage limit allows.
Fixes #1756."
* tag 'tgrabiec/prevent-oom-on-timeouts-v5' of github.com:cloudius-systems/seastar-dev:
storage_proxy: Do not flood logs with timeout errors
database: Add counter for timed out writes
storage_proxy: Delay timeout response until background work ceases
storage_proxy: Propagate timeout to local writes
storage_proxy: Use shared ownership for abstract_write_response_handler
storage_proxy: Add counter for all alive write handlers
db: Allow writes to be timed out
db: Introduce counters for failed reads and writes
commitlog: Allow allocations to be timed out
utils/logalloc: Add ability to timeout run_when_memory_available() task
utils/flush_queue: Add ability to wait with a timeout
Timeout errors are flooding the log after local mutate can time
out. We don't log remote mutate timeouts, so for consistency we won't
log local ones as well.
There is a database counter for timed out writes which can be
consulted in order to check if they're occuring.
Perhaps this would be better solved by a generic log message
throttling/coalescing mechanism, but that's not ready yet.
Write requests which timed out may still occupy memory for a while due
to local write. It should time out soon as well but there is a time
window in which it has not yet. If we don't delay timeout response,
the request would be seen as not consuming any memory too early. This
in turn would cause the CQL server to allow more requests than we
want. In some cases causing OOM or exceeding memory limits and causing
excessive cache eviciton.
Fixes#1756.
Currently the counter uses _response_handlers.size(), but after later
patches we may have an active (timed out) write with no response
handler, so count live instances instead.
Problem will cause size tiered to return small jobs when there are
more than max_threshold sstables of similar size. For example, if
max_threshold is 32, and there are 36 sstables of similar size,
strategy will only return 4 sstables to be compacted. That's because
we incorrectly create a new bucket when it meets the max threshold.
What we should do is to allow buckets to grow beyond max threshold
and trim them when selecting the most suitable one for compaction.
Important to mention that estimation for size tiered will now
work better when there are more than max_threshold sstables of
similar size.
Fixes#1901.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <080bad70d6cb86eaf52ac1bdd6765ac47aab5b03.1478316140.git.raphaelsc@scylladb.com>
This reverts commit 0e9b75d406.
commitlog_test fails with this:
Running 14 test cases...
ERROR 2016-11-28 20:48:00,565 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:00,578 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:10,591 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:20,601 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
tests/commitlog_test.cc(203): fatal error in "test_commitlog_discard_completed_segments": critical check dn <= nn failed
ERROR 2016-11-28 20:48:20,645 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:20,837 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
WARN 2016-11-28 20:48:20,838 [shard 0] commitlog - Exception in segment reservation: std::system_error (error system:2, No such file or directory)
ERROR 2016-11-28 20:48:20,952 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:31,064 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:31,083 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:31,098 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:31,111 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
ERROR 2016-11-28 20:48:31,113 [shard 0] commitlog - Segment reserve is full! Ignoring and trying to continue, but shouldn't happen
WARN 2016-11-28 20:48:31,116 [shard 0] commitlog - Could not allocate 16388 k bytes output buffer (16388 k required)
*** 1 failure detected in test suite "tests/commitlog_test.cc"
WARN 2016-11-28 20:48:31,117 [shard 0] commitlog - Exception in segment reservation: std::system_error (error system:2, No such file or directory)
New sstables are loaded and added in parallel, meaning that scylla can
potentially return stale data if a new sstable containing a tombstone
wasn't loaded yet. Compaction should also not run until all new sstables
are added for similar reasons.
Fix is about separating blocking and non-blocking steps to allow
atomic add of multiple new sstables.
Fixes#1368.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <14283b8a4a69127071d1fabef320a93c91817ec2.1480356073.git.raphaelsc@scylladb.com>
When requests hit the commitlog, each of them will be assigned a replay
position, which we expect to be ordered. If reorders happen, the request
will be discarded and re-applied. Although this is supposed to be rare,
it does increase our latencies, specially when big requests are
involved. Processing big requests is expensive and if we have to do it
twice that adds to the cost.
The commitlog is supposed to issue replay positions in order, and it
coudl be that the code that adds them to the memtables will reorder
them. However, there is one instance in which the commitlog will not
keep its side of the bargain.
That happens when the reserve is exhausted, and we are allocating a
segment directly at the same time the reserve is being replenished. The
following sequence of events with its deferring points will ilustrate
it:
on_timer:
return this->allocate_segment(false). // defer here // then([this](sseg_ptr s) {
At this point, the segment id is already allocated.
new_segment():
if (_reserve_segments.empty()) {
[ ... ]
return allocate_segment(true).then ...
At this point, we have a new segment that has an id that is higher than
the previous id allocated.
Then we resume the execution from the deferring point in on_timer():
i = _reserve_segments.emplace(i, std::move(s));
The next time we need to allocate a segment, we'll pick it from the
reserve. But the segment in the reserve has an id that is lower than the
id that we have already used.
Reorders are bad, but this one is particularly bad: because the reorder
happens with the segment id side of the replay position, that means that
every request that falls into that segment will have to be reinserted.
This bug can be a bit tricky to reproduce. To make it more common, we
can artificially add a sleep() fiber after the allocate_segment(false)
in on_timer(). If we do that, we'll see a sea of reinsertions going on
in the logs (if dblog is set to debug).
Applying this patch (keeping the sleep) will make them all disappear.
We do this by rewriting the reserve logic, so that the segments always
come from the reserve. If we draw from a single pool all the time, there
is no chance of reordering happening. To make that more amenable, we'll
have the reserve filler always running in the background and take it out
of the timer code.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <2606b97df39997bcf3af84a23adf17e094ffb0b8.1480107174.git.glauber@scylladb.com>
"We currently write the size_estimates system table for every schema
on a periodic basis, currently set to 5 minutes, which can interfere
with an ongoing workload.
This patchset virtualizes it such that queries are intercepted and we
calculate the results on the fly, only for the ranges the caller is interested in.
Fixes#1616"
* 'virtual-estimates/v4' of github.com:duarten/scylla:
size_estimates_virtual_reader: Add unit test
db: Delete size_estimates_recorder
size_estimates: Add virtual reader
column_family: Add support for virtual readers
storage_service: get_local_tokens() returns a future
nonwrapping_range: Add slice() function
range: Find a sequence's lower and upper bounds
system_keyspace: Build mutations for size estimates
size_estimates: Store the token range as bytes
range_estimates: Add schema
murmur3_partitioner: Convert maximum_token to sstring
When we finish writing a memtable, we revert the dirty memory charges
immediately. When we do that, dirty memory will grow back to what it
was, and soon (we hope) will go down again when we release the requests
for real.
During that time, we may not accept new requests. Sealing can take a
long time, specially in the face of Linux issues like the ones we have
seen in the past. It also will take proportionally more time if the
SSTables end up being small, which is a possibility in some scenarios.
This patch changes the dirty_memory_manager so that the charges won't be
reverted right after we finish the flush. Rather, we will hold on to it,
and revert it right before we update the cache. We don't need to do it
for all classes of memtable writes, because after we finish flushing,
flush_one() will destroy the hashed element anyway.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <2d5a8f6ca57d5036f4850ac163557bca59b8063d.1480004384.git.glauber@scylladb.com>
GCC 5.3.1 was unable to convert bound to optional<bound>.
sstables/sstables.cc:2494:123: error: no matching function for call to
‘nonwrapping_range<dht::ring_position>::nonwrapping_range(dht::ring_position,
dht::ring_position)’
(dtr.right.exclusive ? dht::ring_position::starting_at :
dht::ring_position::ending_at)(std::move(t2)));
In file included from ./dht/i_partitioner.hh:52:0,
from ./query-request.hh:28,
from ./clustering_key_filter.hh:27,
from sstables/sstables.hh:35,
from sstables/sstables.cc:38:
./range.hh:441:14: note: candidate: nonwrapping_range<T>::nonwrapping_range(
const wrapping_range<U>&) [with T = dht::ring_position]
explicit nonwrapping_range(const wrapping_range<T>& r)
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <95bbf984cd73a61739c8da99cf6cd5e94f1d1457.1479954360.git.raphaelsc@scylladb.com>
Since not all distributions have a version of LZ4 with
LZ4_compress_default(), we use it conditionally.
This is specially important beginning with version 1.7.3 of LZ4,
which deprecates the LZ4_compress() function in favour of
LZ4_compress_default() and thus prevents Scylla from compiling
due to the deprecated warning.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20161124092339.23017-1-duarte@scylladb.com>
In Thrift, SliceRange defines a count that limits the number of cells
to return from that row (in CQL3 terms, it limits the number of rows
in that partition). While this limit is honored in the engine, the
Thrift layer also applies the same limit, which, while redundant in
most cases, is used to support the get_paged_slice verb.
Currently, the limit is not being reset per Thrift row (CQL3
partition), so in practice, instead of limiting the cells in a row,
we're limiting the rows we return as well. This patch fixes that by
ensuring the limit applies only within a row/partition.
Fixes#1882
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <20161123220001.15496-1-duarte@scylladb.com>
With the default value of 12, a node's range is partitioned into
4096 * smp::count sub-ranges which are queried sequentually for a range
scan. If the number of rows in the table is smaller than the required
result size, we will query all of them. This can take so long that we
time out.
A better fix is to query multiple sub-ranges in parallel and merge them,
but for that we need to resurrect the non-sequential merger.
"Clusters with a large number of nodes, or a low number of vnodes, and a
high number of shards, or a combination, suffer from an aliasing problem:
both vnodes and intra-node sharding consider the most significant bits
to select the owning node and owning shard respectively. Since the same
bits are used for both, a low number of vnodes leads to some shards
being overcommitted relative to others.
This series fixes the problem by sharding on bits 0:47 of the token
(murmur3 partitioner only), leaving the most significant 12 bits for
vnodes. Simulation shows that this value provides reasonable sharding
for 100-node, 30-shard clusters.
In order to prevent re-sharding sstables on each boot, token ranges for
the range are stored in a new sub-component of the sstable Statistics
component. With the default 12 ignored bits we have 4096 token ranges
for non-Level-compacted SSTables, which takes some space but is still
reasonable.
Fixes #1277."
Sharding on the most significant token bits aliases with the vnode mechanism,
which also uses the most significant bits; this requires a huge number of
vnodes to achieve good sharding.
This patch teaches the murmur3 partitioner to ignore the most significant
N bits when calculating a token's hard, so we use token bits which still have
some entropy. In effect, with changes the token range layout from
shard 0
shard 1
...
shard S-1
to
shard 0
shard 1
...
shard S-1
shard 0
shard 1
...
shard S-1
...
shard 0
shard 1
...
shard S-1
Where the number of repetitions of the block is 2^(ignored msb bits).
For compatibility, the default is zero ignored bits, matching the pre-patch
state, until we wire things up.
Instead of calculating the owning shard from the sstable's partition
key range, delegate to the new sstable method for getting owning shard
infomation. This insulates us from changes in the sharding algorithm.
When we load an sstable, we don't know beforehand which shards it belongs
to; we don't want to open it until we do. Add a method that allows us
to read just the sharding data, without opening anything else.
Add a metadata component that describes token ranges that are spanned by
this sstable. With the current sharding algorithm, where each shard owns
a single token range, the first/last partition key is sufficient to
describing sharding information, but for multi-range algorithms, this
is not sufficient.
We have a semaphore controlling the amount of background work generated
by the memtable flush process. However, because we are not moving it
inside the memtable post-flush continuation, the units are being
released when we star the flush and not when we finish it.
That's not the intended behavior and that can cause flushes to
accumulate.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <b7dc1866ed3473b9b1862c433d59c5ebd8575dbc.1479839600.git.glauber@scylladb.com>
Instead of calculating the offset for each statistic component manually,
use a loop to iterate over all components, accumulating the offset as we
go along.
Introduce a new function that reuses the file_writer code to compute
the serialized size of an sstable object, by serializing it into memory
and discarding the result.
write() doesn't need to change its input; so change it to const.
The only snag is that describe_type() isn't and can't be made const-correct,
so cheat when it is called and const_cast the input.
This helps in writing a generic serialized_size() that is const correct,
in the next patch.
Recently we have changed our shutdown strategy to wait for the
_request_controller semaphore to make sure no other allocations are
in-flight. That was done to fix an actual issue.
The problem is that this wasn't done early enough. We acquire the
semaphore after we have already marked ourselves as _shutdown and
released the timer.
That means that if there is an allocation in flight that needs to use a
new segment, it will never finish - and we'll therefore neve acquire
the semaphore.
Fix it by acquiring it first. At this point the allocations will all be
done and gone, and then we can shutdown everything else.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <5c2a2f20e3832b6ea37d6541897519a9307294ed.1479765782.git.glauber@scylladb.com>
storage_proxy has an optimization where it tries to query multiple token
ranges concurrently to satisfy very large requests (an optimization which is
likely meaningless when paging is enabled, as it always should be). However,
the rows-per-range code severely underestimates the number of rows per range,
resulting in a large number of "read-ahead" internal queries being performed,
the results of most of which are discarded.
Fix by disabling this code. We should likely remove it completely, but let's
start with a band-aid that can be backported.
Fixes#1863.
Message-Id: <20161120165741.2488-1-avi@scylladb.com>
We current pass a region group to the memtable, but after so many recent
changes, that is a bit too low level. This patch changes that so we pass
a memtable list instead.
Doing that also has a couple of advantages. Mainly, during flush we must
get to a memtable to a memtable_list. Currently we do that by going to
the memtable to a column family through the schema, and from there to
the memtable_list.
That, however, involves calling virtual functions in a derived class,
because a single column family could have both streaming and normal
memtables. If we pass a memtable_list to the memtable, we can keep
pointer, and when needed get the memtable_list directly.
Not only that gets rid of the inheritance for aesthetic reasons, but
that inheritance is not even correct anymore. Since the introduction of
the big streaming memtables, we now have a plethora of lists per column
family and this transversal is totally wrong. We haven't noticed before
because we were flushing the memtables based on their individual sizes,
but it has been wrong all along for edge cases in which we would have to
resort to size-based flush. This could be the case, for instance, with
various plan_ids in flight at the same time.
At this point, there is no more reason to keep the derived classes for
the dirty_memory_manager. I'm only keeping them around to reduce
clutter, although they are useful for the specialized constructors and
to communicate to the reader exactly what they are. But those can be
removed in a follow up patch if we want.
The old memtable constructor signature is kept around for the benefit of
two tests in memtable_tests which have their own flush logic. In the
future we could do something like we do for the SSTable tests, and have
a proxy class that is friends with the memtable class. That too, is left
for the future.
Fixes#1870
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <811ec9e8e123dc5fc26eadbda82b0bae906657a9.1479743266.git.glauber@scylladb.com>
Now that access to the size_estimates system is virtualized, we no
longer need the recorder.
Fixes#1616
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch add a virtual mutation_reader so that queries
to the size_estimates system table are handled by the engine
without needing to perform any IO.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Virtual readers allow queries to selected tables, usually system
tables, to be answered by the engine. This is useful for tables which
aren't written by users and whose contents can be calculated on
demand.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the get_local_tokens() function in storage_service
to return a future instead of requiring running under a seastar::thread.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch add the slice() function to nonwrapping range, which uses
its bounds to slice an input sequence.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts a pair of functions from mutation_partition to
calculate the lower and upper bounds of a sequence from a
nonwrapping_range.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds a function to system_keyspace responsible for creating
a mutation to a partition of the size_estimates system table from a
set of range_estimates.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch changes the range_estimates struct so that the tokens are
represented as utf8 encoded bytes. This will make future patches
require less conversions.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch ensures we can convert the maximum_token to an sstring.
For Cassandra, the minimum and maximum tokens have the same
representation. So, we use the string representation of the
maximum_token for the maximum_token.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"This series is just a cleanup which intention is to deal with all
confusion related to the way T::memory_usage() functions work.
* T::memory_usage() which returned external memory usage are renamed
to T::external_memory_usage()
* T::memory_usage() is introduced where needed to avoid repeating
sizeof(T) + T::external_memory_usage()"
Paweł Dziepak (6):
rename memory_usage() to external_memory_usage() where applicable
streamed_mutation: add memory_usage() to mutation fragment types
keys: add memory_usage()
partition_snapshot_accounter: use range_tombstone::memory_usage()
mutation_rebuilder: use memory_usage()
frozen_mutation: use memory_usage()
In the last iterations of this patchset, we have moved explicit flushes
to acquire the semaphore directly and the coalescing inside the
memtable_list. As a result, we are no longer keeping any kind of action
for them inside the condition variable. Checking for them has no longer
a purpose.
This is a cleanup patch that remove does checks.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <732676ccfe4ac93eb57aa799ec94b841499a01a6.1479500646.git.glauber@scylladb.com>
If a Column Family is non-durable, then its flushes will never create a
memtable flush reader. Our current flush logic depends on that being
created and destroyed to release the semaphore permits on the flush.
We will remove the permits ourselves it there is an exception, but not
under normal circumnstances. Given this issue, however, it would be more
adequate to always try to remove the permits after we flush. If the
permits were already removed by the flush reader, then this test will
just see that the permit is not in the map and return. But if it is
still there, then it is removed.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <049334c3b4bef620af2c7c045e6c84347dcf9013.1479498026.git.glauber@scylladb.com>
This patch introduces memory_usage() to static_row, clustering_row and
range_tombstone so that we can avoid repeating sizeof(T) +
x.external_memory_usage().
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Renaming the function to external_memory_usage() makes it clear that
sizeof(T) is not included, something that was a source of confusion in
the past.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
"Fixes #1856
Commitlog replay reads are being issued without a priority. That means
they will lose to compaction every time."
* 'issue-1856-v2' of github.com:glommer/scylla:
commitlog: use read ahead for replay requests
commitlog: use commitlog priority for replay
commitlog: close replay file
We supported install CentOS7 .rpm on RHEL7, but we haven't supported
building on RHEL7, since there is little difference between CentOS,
and that causes build error.
This patch fixes the error, now we can produce .rpm for RHEL7 wihout
using CentOS.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1479431134-8032-1-git-send-email-syuu@scylladb.com>
This patch addresses post-merge follow up comments by Tomek.
Basically, what we do is:
- we don't need to signal() from remove_from_flush_manager(), because
the explicit flushes no longer wait on the condition variable. So we
don't.
- We now wait on the stop() flushes (regardless of their return status)
so we can make sure that the _flush_queue will indeed be done with.
- we acquire the semaphore before shutting down the dirty_memory_manager
to make sure that there are no pending flushes
- the flush manager that holds the semaphore has to match in the exception
handler
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <a23ab5098934546c660a08de64cd9294bb3a2008.1479400239.git.glauber@scylladb.com>
Aside from putting the requests in the commitlog class, read ahead
will help us going through the file faster.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Right now replay is being issued with the standard seastar priority.
The rationale for that at the time is that it is an early event that
doesn't really share the disk with anybody.
That is largely untrue now that we start compactions on boot.
Compactions may fight for bandwidth with the commitlog, and with such
low priority the commitlog is guaranteed to lose.
Fixes#1856
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Replay file is opened, so it should be closed. We're not seeing any
problems arising from this, but they may happen. Enabling read ahead in
this stream makes them happen immediately. Fix it.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
The leakage results in deleted sstables being opened until shutdown, and disk
space isn't released. That's because column_family::rebuild_sstable_list()
will not remove reference to deleted sstables if an exception was triggered in
sstables::delete_atomically(). A sstable only has its files closed when its
object is destructed.
The exception happens when a major compaction is issued in parallel to a
regular one, and one of them will be unable to delete a sstable already deleted
by the other. That results in remove_by_toc_name() triggering boost::filesystem
::filesystem_error because TOC and temporary TOC don't exist.
We wouldn't have seen this problem if major compaction were going through
compaction manager, but remove_by_toc_name() and rebuild_sstable_list() should
be made resilient.
Fixes#1840.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <d43b2e78f9658e2c3c5bbb7f813756f18874bf92.1479390842.git.raphaelsc@scylladb.com>
"This patchset allows Scylla to determine the size of a memtable instead
of relying in the user-provided memtable_cleanup_threshold. It does that
by allowing the region_group to specify a soft limit which will trigger
the allocation as early as it is reached.
Given that, we'll keep the memtables in memory for as long as it takes
to reach that limit, regardless of the individual size of any single one
of them. That limit is set to 1/4 of dirty memory. That's the same as
last submission, except this time I have run some experiments to gauge
behavior of that versus 1/2 of dirty memory, which was a preferred
theoretical value.
After that is done, the flush logic is reworked to guarantee that
flushes are not initiated if we already have one memtable under flush.
That allow us to better take advantage of coalescing opportunities with
new requests and prevents the pending memtable explosion that is
ultimately responsible for Issue 1817.
I have run mainly two workloads with this. The first one a local RF=1
workload with large partitions, sized 128kB and 100 threads. The results
are:
Before:
op rate : 632 [WRITE:632]
partition rate : 632 [WRITE:632]
row rate : 632 [WRITE:632]
latency mean : 157.8 [WRITE:157.8]
latency median : 115.5 [WRITE:115.5]
latency 95th percentile : 486.7 [WRITE:486.7]
latency 99th percentile : 534.8 [WRITE:534.8]
latency 99.9th percentile : 599.0 [WRITE:599.0]
latency max : 722.6 [WRITE:722.6]
Total partitions : 189667 [WRITE:189667]
Total errors : 0 [WRITE:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:05:00
END
After:
op rate : 951 [WRITE:951]
partition rate : 951 [WRITE:951]
row rate : 951 [WRITE:951]
latency mean : 104.8 [WRITE:104.8]
latency median : 102.5 [WRITE:102.5]
latency 95th percentile : 155.8 [WRITE:155.8]
latency 99th percentile : 177.8 [WRITE:177.8]
latency 99.9th percentile : 686.4 [WRITE:686.4]
latency max : 1081.4 [WRITE:1081.4]
Total partitions : 285324 [WRITE:285324]
Total errors : 0 [WRITE:0]
total gc count : 0
total gc mb : 0
total gc time (s) : 0
avg gc time(ms) : NaN
stdev gc time(ms) : 0
Total operation time : 00:05:00
END
The other workload was the workload described in #1817. And the result
is that we now have a load that is very stable around 100k ops/s and
hardly any timeouts, instead of the 1.4 baseline of wild variations
around 100k ops/s and lots of timeouts, or the deep reduction of
1.5-rc1."
* 'issue-1817-v4' of github.com:glommer/scylla:
database: rework memtable flush logic
get rid of max_memtable_size
pass a region to dirty_memory_manager accounting API
memtable: add a method to expose the region_group
logalloc: allow region group reclaimer to specify a soft limit
database: remove outdated comment
database: uphold virtual dirty for system tables.
If we're talking to just one replica, the digest is not going to be used,
so better not to calculate it at all. The optimization helps with
LOCAL_ONE queries where the result is large, but does not contain large
blobs (many small rows).
This patch adds a digest_algorithm parameter to the READ_DATA verb that
can take on two values: none and MD5 (default), and sets it to none when
we're reading from one replica.
In the future we may add other values for more hardware-friendly digest
algorithms.
Message-Id: <1479380600-19206-1-git-send-email-avi@scylladb.com>
Currenlty we make the mutation readers for streaming at different
time point, i.e.,
do_for_each(_ranges.begin(), _ranges.end(), [] (auto range) {
make a mutation reader for this range
read mutations from the reader and send
})
If there are write workload in the background, we will stream extra
data, since the later the reader is made the more data we need to send.
Fix it by making all the readers before starting to stream.
Fixes#1815
Message-Id: <1479341474-1364-2-git-send-email-asias@scylladb.com>
If sstable Summary is not present Scylla does not refuses to boot but
instead creates summary information on the fly. There is a bug in this
code though. Summary files is a map between keys and offsets into Index
file, but the code creates map between keys and Data file offsets
instead. Fix it by keeping offset of an index entry in index_entry
structure and use it during Summary file creation.
Reviewed-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20161116165421.GA22296@scylladb.com>
The way we currently flush memtables, we seal the current one but wait
on a semaphore for the actual flush to proceed.
This is pointless, because if the flush is not proceeding we'll use up
memory for the new entries anyway, be them in a newly opened memtable or
not. As a matter of fact, by opening a new memtable we are foregoing
coalescing opportunities.
After recent changes to the flush paths, we are now in a position to do
differently. We move the semaphore earlier, and if we can't acquire it
we keep appending to the current memtable.
For explicit flushes, we'll queue and prioritize them over memory-based
flushes. This has the nice property of potentially coalescing various
flushes for the same CF into one.
Coalescing flushes for the same CF is particularly helpful for
commitlog-initiated flushes that can't complete within the flush period.
What we see currently, is that under heavy load the commitlog will keep
sealing memtables adding to the existing load.
Another interesting property of this approach is that we can keep the
disk utilization higher, by allowing a new flush to start before the
memtable is fully sealed. By design, every time a memtable is finished
flushing it will call revert_potentially_cleaned_up_memory() to revert
the virtual memory charges. That is the perfect moment for us to act.
It indicates that all the data flushing part is done.
The way we'll do it is by keeping the semaphore_units alive for this
memtable. When the flush ends, we destroy that object. This will
effectively trigger the next flush if there is a next flush that can be
initiated.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
After recent changes to the memtable code, there is no reason for us to
uphold a maximum memtable size. Now that we only flush one memtable at a
time anyway, and also have soft limit notifications from the
region_group_reclaimer, we can just set the soft limit to the target
size and let all of that be handled by the dirty_memory_manager.
It does have the added property that we'll be flushing when we globally
reach the soft limit threshold. In conditions in which we have multiple
CF writes fighting for memory, that guarantees that we will start
flushing much earlier than the hard limit.
The threshold is set to 1/4 of dirty memory. While in theory we would
prefer the memtables to go as big as 1/2 of dirty memory, in my
experiments I have found 1/4 to be a better fit, at least for the
moment.
The reason for such behavior is that in situations where we have slow
disks, setting the soft limit to 1/2 of dirty will put us in a situation
in which we may not have finished writing down the memtable when we hit
the limit, and then throttle. When set the threshold to 1/4 of dirty, we
don't throttle at all.
This behavior could potentially be fixed by not doing the full
memtable-based throttling after we do the commitlog throttling, but that
is not something realistic for the moment.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We would like to know from which region is a particular flush coming
from, and account accordingly. The reasoning behind that, is that soon
we'll be driving the flushes internally from the dirty_memory_manager
without explcitly triggering them.
We need to start a flush before the current one finishes, otherwise
we'll have a period without significant disk activity when the current
SSTable is being sealed, the caches are being updated, etc.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
That is technically not needed because a memtable inherits from group. So
whenever we have a memtable, we can use it's group() method to obtain a
group for it, and then from there go to the region_group.
However, region() is a const method in the memtable, so we have to play
trick with the const_cast, or remove the constness from the region. An
alternative to that, which I prefer, is to expose a method for the
region_group directly from the memtable object that does the right thing
and bypasses all that.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
The region_group_reclaimer will let us know every time we are over the
limit we have specified for memory usage.
However, For some applications, we would be interested in knowing about
memory build up earlier, so we can start doing something about it before
we reach that condition.
This patch introduce soft limit notifications for the
region_group_reclaimer. After this patch is applied, start_reclaim() is
called earlier, and stop_reclaim() later, after the soft condition is
abated.
There are methods that allow one to easily test if the pressure
condition is a soft limit condition or a hard, threshold condition and
act accordingly. Whether to act on both conditions or just one of them
is up to the application.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Currently the virtual dirty mechanism is not properly set for system
tables. We haven't divided the system table allowance by two, which
means it won't start thottling earlier as it was supposed to.
In practice, this has little effect because system table requests are
very well behaved, their sizes well known, and they tend to be
force-flushed. But we should be consistent.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
"Cache entries for wide partitions are usually smaller than other
entries and the cost of recreating them is higher so it makes sense
to keep them longer than ordinary entries."
This will allow us to see how big is an amount
of evictions of cached info about wide partitions.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Query pager needs to handle results that contain partitions with
possibly multiple clustering rows quite differently than results with
just one row per partition (for example a page may end in a middle of
partition). However, the logic dealing with partitions with clustering
rows doesn't work correctly for SELECT DISTINCT queries, which are
much more similar to the ones for schemas without clustering key.
The solution is to set _has_clustering_keys to false in case of SELECT
DISTINCT queries regardless of the schema which will make pager
correctly expect each partition to return at most one rows.
Fixes#1822.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1478612486-13421-1-git-send-email-pdziepak@scylladb.com>
Now that the histogram has its own unit expressed in its template
parameter, there is no reason to convert it to nano just so we may need
to convert it back if the histogram needs another unit.
This patch will keep everything as a duration until last moment, and
then we'll convert when needed.
This was suggested by Amnon.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <218efa83e1c4ddc6806c51913d4e5f82dc6d231e.1479139020.git.glauber@scylladb.com>
After the conversion to nonwrapping ranges, construct_range_to_endpoint_map()
may be called with semi-infinite token ranges, but it does not expect this,
calling nonwrapping_range::end()->value() unconditionally.
Fix by checking whether this is a semi-infinite range on the right, and
replace ->value() by maximum_token() instead.
Fixes `nodetool describering` (once more).
Message-Id: <1478983010-29630-1-git-send-email-avi@scylladb.com>
Exception handling was broken because after io checker, storage_io_error
exception is wrapped around system error exceptions. Also the message
when handling exception wasn't precise enough for all cases. For example,
lack of permission to write to existing data directory.
Fixes#883.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <b2dc75010a06f16ab1b676ce905ae12e930a700a.1478542388.git.raphaelsc@scylladb.com>
"JMX metrics were found to be either not showing, or showing absurd
values. Turns out there were multiple things wrong with them. The
patches were sent separately but conflict with one another. This series
is a collection of the patches needed to fix the issues we saw.
Fixes#1832, #1836, #1837"
We are tracking latencies in microseconds, but almost everywhere else
they are reported in microseconds. Instead of just converting, this
patch tries to be a bit more future proof and embed the unit into the
type - and we then default to microseconds.
I have verified that the JMX measures now report sane values for both
the storage proxy and the column family. nodetool cfhistograms still
works fine. That one is reported in nanoseconds, but through the
estimated_histogram, not ihistogram.
Fixes#1836
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We have recently fixed a bug due to which the constructor parameters for
moving average were inverted, leading to the numbers being just plain
wrong. However, the calculation of alpha was already inverted, meaning
it was right by accident and now that's wrong.
With the wrong alpha, the values we see are still correct, but they move
very quickly. The intention of this code is obviously to smooth things
out.
This was found out by Nadav. I have tested and confirmed that the smoothing
factor now works as expected.
Fixes #1837
Signed-off-by: Glauber Costa <glauber@scylladb.com>
moving_averages constructor is defined like this:
moving_average(latency_counter::duration interval, latency_counter::duration tick_interval)
But when it is time to initialize them, we do this:
... {tick_interval(), std::chrono::minutes(1)} ...
As it can be seen, the interval and tick interval are inverted. This
leads to the metrics being assigned bogus values.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <d83f09eed20ea2ea007d120544a003b2e0099732.1478798595.git.glauber@scylladb.com>
Snapshot destructor may free some objects managed by the LSA. That's why
partition_snapshot_reader destructor explicitly destroys the snapshot it
uses. However, it was possible that exception thrown by _read_section
prevented that from happenning making snapshot destoryed implicitly
without current allocator set to LSA.
Refs #1831.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1478778570-2795-1-git-send-email-pdziepak@scylladb.com>
After 7c28ed, the schemas defined in the test became compressed by
default. This patch changes the test so that it is explicit about
which schemas shouldn't define a compressor.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1478646530-5558-1-git-send-email-duarte@scylladb.com>
"Mostly small changes/additions to the API calls to match Cv3
requirements/semantics, i.e. updated scylla-jmx can implement required
nodetool etc calls in a working fashion."
The CQL 3.1 documentation specifies that for disabling compression,
users should use an empty string:
ALTER TABLE mytable WITH COMPRESSION = {'sstable_compression': ''};
However, Cassandra also accepts the absence of the sstable_compression
option to disable compression. The patch 7c28ed prevented this behavior
in Scylla, which this patch aims to fix.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Message-Id: <1478639499-4183-1-git-send-email-duarte@scylladb.com>
Current speculation target selection logic has several bugs in multi-dc
setup. It may select a non local target for CL=LOCAL and it may select
more than one target to speculate, one of which is non local.
Examples:
1. Two dataceneters: DC1 RF 2, DC2 RF 2 and read with LOCAL_QUORUM.
In this scenario db::filter_for_query() will return both replicas from
local DC and speculation target selection logic will peek one one which
will be in different DC.
2. Two dataceneters: DC1 RF 2, DC2 RF 2 and read with LOCAL_ONE + RRD.DC_LOCAL
In this scenario db::filter_for_query() will return all nodes in local DC and
there already be enough nodes to speculate, but current logic will add
one node from non local dc as a speculation target.
The patch below fixed both of those scenarios.
Message-Id: <20161103154637.GS7766@scylladb.com>
Since continuity flag introduction row cache contains a single dummy
entry. cache_tracker knows nothing about it so that it doesn't appear in
any of the metrics. However, cache destructor calls
cache_tracker::on_erase() for every entry in the cache including the
dummy one. This is incorrect since the tracker wasn't informed when the
dummy entry was created.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1478608776-10363-1-git-send-email-pdziepak@scylladb.com>
partition_range passed to row_cache::make_reader
has to be kept alive as long as the resulting reader
is used.
Otherwise weird things start to happen.
This used to work just because of a pure luck.
When I started changing the row_cache implementation
I run into very weird behaviors for this tests.
Signed-off-by: Piotr Jastrzebski <piotr@scylladb.com>
Message-Id: <2c9e337dbbcf35f4e1394cad043eda10b8c2bd4a.1478602876.git.piotr@scylladb.com>
Commit 8fca1887c2 ("storage_service: fix range wrapping in
describe_ring") fixed incorrect range wrapping code for describe_ring,
but fails when the number of endpoints for a token is greater than one,
because the endpoints are stored in an unordered vector.
Fix by comparing the endpoints in a way that ignores their order.
Message-Id: <1478460826-15923-1-git-send-email-avi@scylladb.com>
Fixes#1775
stream lacks a check "is_open", which is a bummer. We have to both
prevent exception propagation and add a flag of our own to make sure
exceptions in producer code reaches consumer, and does not simply
get lost in the reactor.
Message-Id: <1478508817-18854-1-git-send-email-calle@scylladb.com>
This patchset adds missing properties to the create_view_statement,
such as whether the view is compact or the order of its clustering
columns.
Fixes#1766
"The atomic sstable deletion provides exception safety at the cost of
quadratic behavior in the number of sstables awaiting deletion. This
causes high cpu utilization during startup.
Change the code to avoid quadratic complexity, and add some unit tests.
See #1812."
In order to ensure exception safety, the atomic sstable deletion code
creates a copy of the list of sstables pending deletion, modifies that
copy, and then replaces the original data with the copy. This guarantees
that any exception does not change the data, since the assignment does
not require allocation.
However, it does result in quadratic behavior. During startup, all
sstables are loaded on each shard, and each shard deletes sstables that
are do not have any partitions served by that shard; this results in
almost all sstables being deleted from all shards, with all that work
going to shard 0; the list grows to O(nr sstables), and there are
O((nr sstables) * (nr shards)) operations to perform.
Fix by replacing the copy-modify-assign method with an in-place update,
but one that is designed to only commit changes after all allocations
have been made; in addition, instead of using a list, use a hash table,
removing another source of quadratic behavior.
Fixes#1812 (the quadratic beahvior part).
"Currently, partition range queries are processed in parallel on all
shards. This is inefficient because we are likely to drop the results
from all but one shard, assuming a well-populated column family. We
are multiplying our work by a factor of smp::count.
While this is worthwhile in its own right, it is really an excuse to
sneak in the range/shard generator (patch 5), which is preliminary for
a new sharding algorithm, dividing tokens among shards based on the
middle-significant bits rather than the most-siginificant bits (which
alias with vnodes)
Fixes #1573."
Instead of asking a shard for cmd->partition_limit and cmd->row_limit,
just ask it for the number of partitions and rows still needed to
satisfy the query. This removes the need to trim the shard's result.
Since every shard might cause the row_limit quota to be satisfied, every
shard might be the last one we need. Hence it is better to process shards
sequentially, stopping if the quota is reached or the range is exhausted.
The original code tried to yield to reduce latency, but this is now
unnecessary, as we're doing a lot less work per iteration (if it becomes
necessary, we should do it on the replica shard, not the coordinating shard).
Building on the single-range sharder, add a sharder for vectors of
partition ranges. This helps with wrapped ranges, which are translated
into a vector containing two shards.
Divides a ring_position range into a sequence of shard/range pairs. This
allows sequential iteration over shards in ring order.
The current multi-partition query executes on all shards in parallel, but
this is very wasteful, as most of the data will be thrown away if it is not
included in the page. With the generator, we can switch to sequential
execution.
When performing a range query, we want to iterate over shards, running the
query on each shard in order until the query range is exhausted or we have
the right number of rows.
To be able to do this, introduce token_for_next_shard(), which allows us
to determine the boundary between shards.
It is a sort-of inverse to shard_of(), in that
shard_of(token_for_next_range(t)) == shard_of(t) + 1
- Add a inserts, updates, deletes members to cql_stats.
- Store cql_stats& in a modification_statement and increment the corresponding counter according to the value of a "type" field.
- Store cql_stats& in a batch_statement and increment the statistics for each BATCH member.
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
- Add a "reads" counter to a cql3::cql_stats struct.
- Store a reference for a query_processor::_cql_stats in the select_statement object.
- Increment a "reads" counter where needed.
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Right now we are calculating latencies only when we are about to add an
item to the memtable.
That's incorrect and misleading, for two reasons. First, it leaves the
commitlog latencies out. But second, it is done after the memtable wall
effect is applied, which means we are not counting throttle time neither
in the memtables or in the commitlog.
To do that, we'll start the latency_counter object as soon as possible
and move it all the way to apply_in_memory(). That should span the
entire write operation.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <4e424780d290fd5938046060df2b17e2b470b717.1478111467.git.glauber@scylladb.com>
There are two variants of apply_in_memory() being called in do_apply():
with and without the commitlog. The main differences are that when the
commitlog is involved, we need to wait for its future to complete before
moving to apply_in_memory. That can easily be factored out by providing
an always-ready future if we don't have the commitlog enabled, and
waiting on that.
The second, is that the commitlog version can cause apply_in_memory to
generate an exception if there is replay position reordering. However,
there is no harm in appending the exception handler to both versions. In
one of them it's an impossible exception, but that's fine.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <8cee0cad9b1930a057a24e095f0a655069ae8be2.1478111467.git.glauber@scylladb.com>
There are places in which we need to use the column family object many
times, with deferring points in between. Because the column family may
have been destroyed in the deferring point, we need to go and find it
again.
If we use lw_shared_ptr, however, we'll be able to at least guarantee
that the object will be alive. Some users will still need to check, if
they want to guarantee that the column family wasn't removed. But others
that only need to make sure we don't access an invalid object will be
able to avoid the cost of re-finding it just fine.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <722bf49e158da77ff509372c2034e5707706e5bf.1478111467.git.glauber@scylladb.com>
Since size estimates are stored as wrapped ranges, we call compat::wrap()
to convert from the now-standard unwrapped ranges back to wrapped ranges.
However, compat::wrap() relies on the ranges being in sorted order,
but our input is not. This leads to a crash as we find an unexpected
empty token in the middle of the vector.
Sort it so compat::wrap() works as expected.
Fixes#1804.
Message-Id: <1478161908-25051-1-git-send-email-avi@scylladb.com>
Wrapping ranges are a pain, so we are moving wrap handling to the edges.
Since cql can't generate wrapping ranges, this means thrift and the ring
maintenance code; also range->ring transformations need to merge the first
and last ranges.
Message-Id: <1478105905-31613-1-git-send-email-avi@scylladb.com>
We were ignored --smp option taken from io.conf since iotune didn't supported
it, but now it supported we can pass it.
(We need to pass it because we need to measure io performance on same condition
with scylla)
Fixes#1768
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1478082591-27205-1-git-send-email-syuu@scylladb.com>
Under the hood, the selectable::add_and_get_index() function
deliberately filters out duplicate columns. This causes
simple_selector::get_output_row() to return a row with all duplicate
columns filtered out, which triggers and assertion because of row
mismatch with metadata (which contains the duplicate columns).
The fix is rather simple: just make selection::from_selectors() use
selection_with_processing if the number of selectors and column
definitions doesn't match -- like Apache Cassandra does.
Fixes#1367
Message-Id: <1477989740-6485-1-git-send-email-penberg@scylladb.com>
This patch adds an `update-version` script for updating the Scylla
version number in `SCYLLA-VERSION-GEN` file and committing the change to
git.
Example use:
$ ./scripts/update-version 1.4.0
which results into the following git commit:
commit 4599c16d9292d8d9299b40a3e44ef7ee80e3c3cf
Author: Pekka Enberg <penberg@scylladb.com>
Date: Fri Oct 28 10:24:52 2016 +0300
release: prepare for 1.4.0
diff --git a/SCYLLA-VERSION-GEN b/SCYLLA-VERSION-GEN
index 753c982..eba2da4 100755
--- a/SCYLLA-VERSION-GEN
+++ b/SCYLLA-VERSION-GEN
@@ -1,6 +1,6 @@
#!/bin/sh
-VERSION=666.development
+VERSION=1.4.0
if test -f version
then
Message-Id: <1477639560-10896-1-git-send-email-penberg@scylladb.com>
Fixes#1709.
* 'refresh-resilient-v3' of github.com:raphaelsc/scylla:
db: make refresh resilient to permission denied error
db: make it possible to use custom error handler with io checker
sstables: remove duplicated declaration of remove_by_toc_name
We use `data_resource` class in the CQL parser, which let's users refer
to a table resource without specifying a keyspace. This asserts out in
get_level() for no good reason as we already know the intented level
based on the constructor. Therefore, change `data_resource` to track the
level like upstream Cassandra does and use that.
Fixes#1790
Message-Id: <1477599169-2945-1-git-send-email-penberg@scylladb.com>
We store all auth perm strings in upper case, but the user might very
well pass this in upper case.
We could use a standard key comparator / hash here, but since the
strings tend to be small, the new sstring will likely be allocated in
the stack here and this approach yields significantly less code.
Fixes#1791.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <51df92451e6e0a6325a005c19c95eaa55270da61.1477594199.git.glauber@scylladb.com>
User may forget to set permission of new sstables in upload dir
before refreshing them, and that will result in shutdown.
io_checker is now able to work with a custom handler, so all we
have to do is to whitelist EACCES.
Fixes#1709.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
By default, io checker will cause Scylla to shutdown if it finds
specific system errors. Right now, io checker isn't flexible
enough to allow a specialized handler. For example, we don't want
to Scylla to shutdown if there's an permission problem when
uploading new files from upload dir. This desired flexibility is
made possible here by allowing a handler parameter to io check
functions and also changing existing code to take advantage of it.
That's a step towards fixing #1709.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
EC2 paravirtual instances uses pv-grub, which refers /boot/grub/menu.lst (grub0.9x config file) instead of grub2 config file.
So add boot parameters on /boot/grub/menu.lst when the file exists, and the instance is on EC2.
Fixes#1598
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1472056875-17512-1-git-send-email-syuu@scylladb.com>
wget is often used from scripts recording to logs; as it emits a log
line every second, the logs are huge and unreadable. Make it quieter.
Message-Id: <1477558534-32718-1-git-send-email-avi@scylladb.com>
"5ff699e09fcbd62611e78b9de601f6c8636ab2f0 ("row_cache: rework cache to
use fast forwarding reader") brought some significant changes to the
row cache implementation. Unfortunately, "significant changes" often
translates to "more bugs" and this time was no different.
This series contains fixes for the problems introduced in that rework
and makes failing dtest
bootstrap_test.py:TestBootstrap.local_quorum_bootstrap_test
pass again."
* 'pdziepak/cache-fixes/v1' of github.com:cloudius-systems/seastar-dev:
row_cache: avoid dereferencing invalid iterator
row_cache: set _first_element flag correctly
row_cache: fix clearing continuity flag at eviction
Conditions in row_cache::do_find_or_create_entry() make it possible that
std::prev(it) is going to be dereferenced even if it is a begin
iterator.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
If the continuity flag was set for the first element _first_element flag
would not be cleared. This shouldn't cause any correctness problems but
properly setting the flag allows to avoid some unnecessary key
comparisons.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
In original implementation the continuity flag indicated that cache has
full information about the range the between current partition and the
one following it, hence when evicting an entry the one preceeding it
had to have its continuity flag cleared.
This was changed, however, and now the continuiy flag tells whether the
cache is continuous between the current element and the one before it.
This means that eviction code needs to clear the flag for the entry
directly following the evicted one.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
* seastar 69acec1...98b5a2d (9):
> rpc: Silence warning about ignored failed future
> future: prioritise continuations that can run immediately
> iotune: relax aio restrictions
> build: support for static linking with boost
> rpc: Fix crash during connection teardown
> rpc: Move _connected flag to protocol::connection
> rpc test: fail test if exception is thrown during test execution
> rpc: do not assume underling semaphore type
> rpc: fix default resource limit
In boost 1.60, the executable's command-line arguments are expected to
be separated from the boost command-line arguments by '--'. Detect
this requirement and comply with it.
Message-Id: <1477212424-3831-1-git-send-email-avi@scylladb.com>
We calculate two sizes during the allocation: "size", which is the
in-segment size of this mutation, and "s", which is that plus the
overhead. cycle() must be called with the latter, not the former, as
doing otherwise may lead to buffer overflows.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <ccf346d8d0ebb44a1ba9fd069653bab0d7be0a61.1477063157.git.glauber@scylladb.com>
"This patchset reworks the commitlog logic to better handle conditions in
which we are getting requests faster than the disk can handle. It does
this by building a wall around the commitlog and only allowing
allocations to proceed when we are under the desired memory threshold.
The main advantage of that is that we can now easily set the commitlog
to work at disk speed, more or less allowing an "one byte in for each
byte out" approach instead of depending on the current cycle to finish.
As a result, max latencies are greatly reduced.
Testing Results
===============
To test this, I have ran a workload that times out frequently. That
workload use 10 threads to write 100 partitions (to isolate from the
effects of the memtable introduced latencies) in a loop and each
partition is 2MB in size.
After 10 minutes running this load, we are left with the following
percentiles:
latency mean : 51.9 [WRITE:51.9]
latency median : 9.8 [WRITE:9.8]
latency 95th percentile : 125.6 [WRITE:125.6]
latency 99th percentile : 1184.0 [WRITE:1184.0]
latency 99.9th percentile : 1991.2 [WRITE:1991.2]
latency max : 2338.2 [WRITE:2338.2]
After this patch:
latency mean : 54.9 [WRITE:54.9]
latency median : 43.5 [WRITE:43.5]
latency 95th percentile : 126.9 [WRITE:126.9]
latency 99th percentile : 253.9 [WRITE:253.9]
latency 99.9th percentile : 364.6 [WRITE:364.6]
latency max : 471.4 [WRITE:471.4]
I have run this with larger sizes as well, and it generally performs
much better than the baseline version. For sizes up to 5MB, I have seen
no timeouts in my setup. After that, I see some timeouts. Buffer
splitting is expected to make this better.
Aside from performance testing, this was also tested with batch and
periodic mode for various requests sizes."
Current tracker for pending allocations is a queue_size GAUGE. Add a
total_operations version so we have more insight on what's going on.
It will be called requests_blocked_memory for consistency with other
subsystems that track similar things.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
scylla-blocktune is a script that parses scylla.yaml and tunes the data file
and commitlog directories it references.
Tuning includes:
- set the I/O scheduler to noop
- disable merging
- tune dependent devices (like RAID members)
Message-Id: <1476357027-15014-2-git-send-email-avi@scylladb.com>
The current chunk size of 256 gives a 50% probability of a 128k read or
write getting split into two accesses. This reduces efficiency and
increases latency.
Change the chunk size to 1MB, with a 12% probability of cross-member
access.
Message-Id: <1476269082-2473-1-git-send-email-avi@scylladb.com>
Since Debian 8(jessie) does not provides g++-5, we frequently got compile error
because we are using older compiler.
To fix the problem, backport g++-5 from Debian 9(stretch).
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1476694318-10640-3-git-send-email-syuu@scylladb.com>
On Debian, lsb_release -r returns the version number something like '8.6'.
However, on this script we want to check major version only.
Therefore we can use VERSION_ID from /etc/os-release which only contains
major version number.
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1476694318-10640-2-git-send-email-syuu@scylladb.com>
Commit 7dcd70124a "tests/sstables: add
test for fast forwarding reader" added a test for skipping parts of
sstable. Unfortunately, it did not include the sstables it was trying to
read.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
The current incarnation of commitlog establishes a maximum amount of
writes that can be in-flight, and blocks new requests after that limit
is reached.
That is obviously something we must do, but the current approach to it
is problematic for two main reasons:
1) It forces the requests that trigger a write to wait on the current
write to finish. That is excessive; ideally we would wait for one
particular write to finish, not necessarily the current one. That
is made worse by the fact that when a write is followed by a flush
(happens when we move to a new segment), then we must wait for
*all* writes in that segment to finish.
1) it casts concurrency in terms of writes instead of memory, which
makes the aforementioned problem a lot worse: if we have very big
buffers in flight and we must wait for them to finish, that can
take a long time, often in the order of seconds, causing timeouts.
The approach taken by this patch is to replace the _write_semaphore
with a request_controller. This data structure will account the amount
of memory used by the buffers and set a limit on it. New allocations
will be held until we go below that limit, and will be released
as soon as this happens.
This guarantees that the latencies introduced by this mechanism are
spread out a lot better among requests and will keep higher percentile
latencies in check.
To test this, I have ran a workload that times out frequently. That
workload use 10 threads to write 100 partitions (to isolate from the
effects of the memtable introduced latencies) in a loop and each
partition is 2MB in size.
After 10 minutes running this load, we are left with the following
percentiles:
latency mean : 51.9 [WRITE:51.9]
latency median : 9.8 [WRITE:9.8]
latency 95th percentile : 125.6 [WRITE:125.6]
latency 99th percentile : 1184.0 [WRITE:1184.0]
latency 99.9th percentile : 1991.2 [WRITE:1991.2]
latency max : 2338.2 [WRITE:2338.2]
After this patch:
latency mean : 54.9 [WRITE:54.9]
latency median : 43.5 [WRITE:43.5]
latency 95th percentile : 126.9 [WRITE:126.9]
latency 99th percentile : 253.9 [WRITE:253.9]
latency 99.9th percentile : 364.6 [WRITE:364.6]
latency max : 471.4 [WRITE:471.4]
Signed-off-by: Glauber Costa <glauber@scylladb.com>
In a subsequent patch, I'll use this code in a different place. To
prepare for that, we move it out as a method. It also fits a lot better
inside the segment manager, so move it there.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Goal is to calculate a size that is lesser or equal than the
segment-dependent size.
This was originally written by Tomasz, and featured in his submission
"commitlog: Handle overload more gracefully"
Extracted here so it sits clearly in a different patch.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
It is mostly an optimization, and while it makes sense in this context,
it won't soon as we'll stop waiting for the current cycle specifically
to finish.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We'll do that so we can, in following patches, use static members from
the segment. Those are not defined at this point.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We track the amount of pending allocations but we don't really export
it. It will be crucial when we stop tracking pending writes.
This patch exports it through a method instead of the totals structure,
so we can easily change it. Current code probing pending_allocations
(the api code) is also converted to use the public method instead of the
totals struct.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
"This patchset enables mutation readers to be fast forwarded to a different
partition range. The main reason for introducing such feature are range
queries served from cache. If the cache is partially populated in the
requested range the reader will end up with multiple subranges that have
to be read from the sstables. Originally, each of these subranges would
require a new reader to be created, but with fast forwarding we can have
just one sstable reader. This is better since there is a chance that buffers
kept by the reader may be still useful after fast forwarding it.
In this series there are also patches that clean up cache readers in order
to make integration with fast forwarding easier. Namely, continuity flag is
changed to store information about range before the entry which significantly
simplifies the logic.
Fixes #1299."
* 'pdziepak/fast-forward-mutation-readers/v5' of github.com:cloudius-systems/seastar-dev: (24 commits)
sstables: keep separate stream history for single and range reads
sstables: drop sstable::{lower, upper}_bound()
row_cache: rework cache to use fast forwarding reader
row_cache: put cache entry flags in a struct
row_cache: add do_find_or_create_entry() to reduce code duplication
mutation_reader: forward fast_forward_to() calls
tests/row_cache: add fast_forward_to() to throttled reader
tests/row_cache: count mutations read from _underlying
memtable: add support for fast_forward_to()
drop key readers
tests/mutation_reader: test fast forwarding combined reader
database: enable fast forwarding of range_sstable_reader
combined_mutation_reader: implement fast_forward_to()
mutation_reader: make combinded_reader public
tests/sstables: add test for fast forwarding reader
tests: add more helpers to mutation reader assertions
sstables: enable fast forwarding for range readers
mutation_reader: introduce fast_forward_to()
sstables: implement mutation_reader::impl::fast_forward_to()
sstables: introduce index_reader
...
Single partition and partition range reads are expected to behave
considerably different so it is worth to have them use separate file
stream history. This also makes reads use different history for each
sstable which is also a good thing.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
This uncomfortably large patch overhauls cache range reader so that it
can take advantage of fast forwarding mutation readers.
A significant change in the cache itself is that the continuity flag now
is used to determine whether cache is contiguous between the previous
entry and the current one. This allows for a significant simplification
of the cache code and easier integration with reader fast forwarding.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Flags are easier to manage if they are in a single structure.
Especially, default initialization and move contstructors are simpler
and less error prone.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Originally, cache tests checked how many times a mutation reader was
created from the underlying mutation source to determine whether
continuity flag is working correctly.
This is not going to work with fast forwarding mutation readers so the
test is switched to count number of mutations (+ end of stream markers)
returned from underlying mutaiton readers which is much less fragile.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Fast forwarding of memtable readers is needed only for unit tests which
often use memtables as underlying data source for cache and the cache
readers.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
When fast forwarding a reader that combines sstable reader we must also
remember that the set of sstables for the new range may be different
than for the previous one. The reader introduced in this patch makes
sure that we read from correct sstables.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
We want to be able to fast forward sstable readers. However, just
implementing fast_forward_to() for combined_reader is not enough as the
sstables we are reading from may need to change.
Following patches are going to introduce a combined sstable reader that
derives from combined_reader. To make that possible we first need to
make combined_reader public.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
This patch introduces the interface for fast forwarding mutation
readers. The main user of this feature is going to be cache which, while
serving range query, may need to read multiple small ranges from the
sstables to populate itself with the missing entries.
Fast forwarding is an alternative to recreating a reader with different
range. Its main advantage is fact that it avoids dropping data that has
already been read.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
This patch allows sstable readers to be fast forwarded without making it
necessary to recreate the reader (and dropping all buffers in the
process). It is built on top of index_reader and ability of
data_consume_context to be fast forwarded.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
index_reader is a helper that implements index lookups. Its goal is to
avoid dropping read buffers if they still may be needed (for example to
get end bound of the range or after fast forwarding the reader).
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
That overload was used only by unit test and violated guarantee that
partition range lives until mutation reader is done.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
is_compactible() will pass on very small regions. full_compaction() is
only used in tests to force objects to be moved due to compaction, so
we want all reclaimable regions to be compacted.
The move constructor of partition_version was not invoking move
constructor of anchorless_list_base_hook. As a result, when
partition_version objects were moved, e.g. during LSA compaction, they
were unlinked from their lists.
This can make readers return invalid data, because not all versions
will be reachable.
It also casues leaks of the versions which are not directly attached
to memtable entry. This will trigger assertion failure in LSA region
destructor. This assetion triggers with row cache disabled. With cache
enabled (default) all segments are merged into the cache region, which
currently is not destroyed on shutdown, so this problem would go
unnoticed. With cache disabled, memtable region is destroyed after
memtable is flushed and after all readers stop using that memtable.
Fixes#1753.
Message-Id: <1476778472-5711-1-git-send-email-tgrabiec@scylladb.com>
This patch uses cf_properties instead to add the missing attributes to
the create_view_statement class.
Fixes#1766
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the VIEWS element to the cause enum so we can
mark failures due to incomplete support of materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts the definition of the default compressor into the
compression_parameters class, so that the table and view creation
statements don't have to explicitly deal with it.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch extracts the cf_properties class, which contains common
attributes of tables and materialized views.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Since we are exiting Scylla process in engine().at_exit() using
::_exit(0), even verify_seastar_io_scheduler() throwing an exception,
scylla always exit with 0.
Systemd misunderstands scylla-server.service was shutdown successfully
because of this, so we need to pass correct exit code to ::_exit() here.
Fixes#1674
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1475065607-15486-1-git-send-email-syuu@scylladb.com>
* seastar 207bf3d...ccd8649 (3):
> Merge "Augment semaphore with non-blocking operations" from Glauber
> Merge "More dynamic fstream patches" from Paweł
> Merge "fstream: add dynamic adjustments based on stream history" from Paweł
A 1MB response will require 2000 allocations with the current 512-byte
chunk size. Increase it exponentially to reduce allocation count for
larger responses (still respecting the upper limit).
Message-Id: <1476369152-1245-1-git-send-email-avi@scylladb.com>
Memory accounting code was attaching partition_snapshot to
partition_entry in order to calculate the size of partition_version
object. However, it is only allowed if partition_entry doesn't have
any snapshot attached already. In this case it always has one, created
by the flushing reader.
Change the accounting code to reuse existing partition_snapshot reference.
Fixes#1746
Message-Id: <1476449160-9252-1-git-send-email-tgrabiec@scylladb.com>
LSA tries to allocate zones as large as possible (while still leaving
enough free space for the standard allocator). It uses the amount of
free memory in order to guess how much it can get, but that obviously
doesn't account for fragmentation and the allocation attempt may fail.
This patch changes the LSA code so that it doesn't throw in case zone
couldn't be created but just returns a null pointer which should be
more performant if the LSA memory cannot grow any more.
Fixes#1394.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1476435031-5601-1-git-send-email-pdziepak@scylladb.com>
The expected behaviour in the scylla_setup script is that a question
will be followed by the answer.
For example, after asking if the scylla should be run as a service the
relevant actions will be taken before the following question.
This patch address two such mis-orders:
1. the scylla-housekeeping depends on the scylla-server, but the
setup should first setup the scylla-server service and only then ask
(and install if needed) the scylla-housekeeping.
2. The node_exporter should be placed after the io_setup is done.
Fixes#1739
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
Message-Id: <1476370098-25617-1-git-send-email-amnon@scylladb.com>
Change abstract_replication_strategy::create_replication_strategy() to
throw exceptions::configuration_error if replication strategy class
lookup to make sure the error is converted to the correct CQL response.
Fixes#1755
Message-Id: <1476361262-28723-1-git-send-email-penberg@scylladb.com>
* seastar f937fb0...207bf3d (11):
> Merge "iotune: gracefully exit on predictable exceptions" (Fixes#1623)
> core/semaphore: Add semaphore_units::release()
> Merge "rometheus API with grafana uses labels" from Amnon
> core/thread: Fix stack alloc-dealloc mismatch
> core/thread: Make jmp_buf_link::yield_at use the same time point as thread_scheduling_group
> file: support for XFS on older kernels
> reactor: fix bug when handling EBADF in flush_pending_aio()
> prometheus CPU should start in 0
> Collectd: bytes ordering depends on the type
> tests: Check that backtrace() doesn't corrupt signal mask
> core/thread: Add stack guards to seastar thread stacks
If we have a range query involving a wrapping range (i.e., from thrift),
and mutations from both halves of the result are involved, then
we will return the results in the wrong order (and potentially the wrong
partitions) since we order by token, so the results from the second half
of the wrapping range end up before the first.
Fix by splitting the two queries, and merging the second half with lower
priority compared to the first half.
Note: this will be fixed in a better way once we have the sharding iterator,
as then we can query sequentially.
Fixes#1761.
Message-Id: <1476262693-30162-1-git-send-email-avi@scylladb.com>
"This series address two issues that interfere with running the node_exporter as a service in ubuntu 16.
1. The service file should be packed in the deb file
2. When setting the node_exporter as a service it doesn't need to run with scylla use"
* 'amnon/node_exporter_ubuntu_v2' of github.com:cloudius-systems/seastar-dev:
node-exporter service: No need to run as scylla user
debian package: Include the node_exporter service file
the node-exporter does not need to run as scylla user. It can run
without scylla or without the scylla user being configure.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
"This patch-set re-implements the describe_splits_ex() verb to more closely
follow Cassandra's implementation, on which some clients rely.
Ref #1139
Ref #693"
* 'describe-splits/v2' of github.com:duarten/scylla:
thrift: Implement describe_splits_ex based on Cassandra
storage_service: Implement get_splits() function
sstables: Add function to get key samples
sstables/key: Add to_partition_key function
size_estimates_recorder: Increase estimate accuracy
sstables: Get estimates for a particular range
sstables/key: Make key::kind public
The script mistakenly split value at "," when cpuset list is separated
by comma. Instead of matching possible patterns of the argument, let's
pass all characters until reach to space delimiter or end of line.
Fixes#1716
Signed-off-by: Takuya ASADA <syuu@scylladb.com>
Message-Id: <1476171037-32373-1-git-send-email-syuu@scylladb.com>
This patch re-implements the describe_splits_ex() verb to more closely
follow Cassandra's implementation, on which some clients rely.
Ref #1139
Ref #693
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch implements the get_splits() function in storage_service,
used to split a particular token range in slices of approximately the
specified size, using the sample keys and estimates of the CF's
sstables.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch implements the get_key_samples() function, on which a
future patch will base an implementation of the describe_splits()
thrift verb closer to Cassandra's.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the estimated_keys_for_range() function, which
estimates the number of keys present between the specified range.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
"The version is taken from the installation rather than the API, a mode command
line indicated that this is part of the setup and uuid is used for the
interaction with the checkversion server."
* 'amnon/check_version_on_startup_v3' of github.com:cloudius-systems/seastar-dev:
scylla_setup: Check and report the scylla version
scylla-housekeeping: check version during setup
There is already queue_length-requests_blocked_memory, but it's a
gauge so does not reflect what happened between the sampling points.
total_operations-requests_blocked_memory will allow to see if there
were any (and how many) requests which were blocked by dirty memory.
Message-Id: <1476098616-12682-1-git-send-email-tgrabiec@scylladb.com>
Presents current heap profile recording.
Works in text mode or dumps to collapsed stacks format from which
flame graph can be generated.
To generate a flamegraph:
(gdb) scylla heapprof --flame
Wrote heapprof.stacks
$ flamegraph.pl --colors mem < heapprof.stacks > heapprof.svg
flamegraph.pl comes from:
https://github.com/brendangregg/FlameGraph.git
Text mode example:
(gdb) scylla heapprof --min 100000000
All (274699676, #10213)
\-- void* memory::cpu_pages::allocate_large_and_trim<memory::cpu_pages::allocate_large_aligned(unsigned int, unsigned int)::{lambda(unsigned int, unsigned int)#1}>(unsigned int, memory::cpu_pages::allocate_large_aligned(unsigned int, unsigned int)::{lambda(unsigned int, unsigned int)#1}) + 169 (268435456, #1)
memory::allocate_large_aligned(unsigned long, unsigned long) + 87
memory::allocate_aligned(unsigned long, unsigned long) + 48
aligned_alloc + 9
logalloc::segment_zone::segment_zone() + 304
logalloc::segment_pool::allocate_segment() + 477
logalloc::segment_pool::segment_pool() + 304
__tls_init.part.801 + 72
logalloc::region_group::release_requests() + 1333
logalloc::region_group::add(logalloc::region_group*) + 514
The branches are formatted like this:
-- <symbol> (<size>, #<count>)
Where <size> is total size of live objects and <count> is total
number of live objects, for all objects allocated from paths going
through this node.
Nodes which share the same <size> and <count> are stacked like this:
-- <symbol_1> (<size>, #<count>)
<symbol_2>
<symbol_3>
Message-Id: <1475583334-19524-1-git-send-email-tgrabiec@scylladb.com>
Limiting the concurrency of memtable flushes to 4 was a temporary
workaround for the fact that we lacked good write behind support. Now
that write behind is properly merged we can reduce the concurrency to
what it should be, one.
This means that memtable flushes will now be serialized, and only when
one of them ends will the next one begin. Disk parallelism is obtained
through the write-behind mechanism.
Fixes#1373
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Message-Id: <528f9ef928b5101bed952df600eb8555c275497a.1475881100.git.glauber@scylladb.com>
There is a limit to concurrency of sstable readers on each shard. When
this limit is exhausted (currently 100 readers) readers queue. There
is a timeout after which queued readers are failed, equal to
read_request_timeout_in_ms (5s by default). The reason we have the
timeout here is primarily because the readers created for the purpose
of serving a CQL request no longer need to execute after waiting
longer than read_request_timeout_in_ms. The coordinator no longer
waits for the result so there is no point in proceeding with the read.
This timeout should not apply for readers created for streaming. The
streaming client currently times out after 10 minutes, so we could
wait at least that long. Timing out sooner makes streaming unreliable,
which under high load may prevent streaming from completing.
The change sets no timeout for streaming readers at replica level,
similarly as we do for system tables readers.
Fixes#1741.
Message-Id: <1475840678-25606-1-git-send-email-tgrabiec@scylladb.com>
Make split_after() more generic by allowing split_point to be anywhere,
not just within the input range. If the split_point is before, the entire
range is returned; and if it is after, stdx::nullopt is returned.
"before" and "after" are not well defined for wrap-around ranges, so
but we are phasing them out and soon there will not be
wrapping_range::split_after() users.
This is a prerequisite for converting partition_range and friends to
nonwrapping_range.
Message-Id: <1475765099-10657-1-git-send-email-avi@scylladb.com>
Commit log replay is a synchronous operation in bootstrap, so services
will only be started after it's completed. By starting compaction before,
less bandwidth will be available to both and consequently boot will be
slowed down. Fix is simply about moving compaction, which is an
asynchronous operation after commitlog replay is over.
Fixes#1620.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <d2a173a4ee4d474317b970c6b39530e61067fea9.1475527955.git.raphaelsc@scylladb.com>
This patch adds the parsing for the "CREATE MATERIALIZED VIEW" statement,
following Cassandra 3 syntax. For example:
CREATE MATERIALIZED VIEW building_by_city
AS SELECT * FROM buildings
WHERE city IS NOT NULL
PRIMARY KEY(city, name);
It also adds the "IS NOT NULL" operator needed for this purpose.
As in Cassandra, "IS NOT NULL" can only be used for materialized
view creation, and not in a normal SELECT. It can only be used with
the NULL operand (i.e., "IS NOT 3" will be a syntax error).
The current implementation of this statement just does some sanity
checking (such as to verify that "city" is a valid column name and that
the "building" base table exists), complains that materialized views are
not yet supported:
SyntaxException: <ErrorMessage code=2000 [Syntax error in CQL query] message="Failed parsing statement: [CREATE MATERIALIZED VIEW building_by_city AS
SELECT * FROM buildings
WHERE city IS NOT NULL
PRIMARY KEY(city, name);] reason: unsupported operation: Materialized views not yet supported">
As mentioned above, the "IS NOT NULL" restriction is not allowed in
ordinary selects not creating a materialized views:
SELECT * FROM buildings WHERE city IS NOT NULL;
InvalidRequest: code=2200 [Invalid query] message="restriction 'city IS NOT null' is only supported in materialized view creation"
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <1475742927-30695-1-git-send-email-nyh@scylladb.com>
The latest virtual dirty patches broke the SSTable tests. The reason for
this is that those tests will flush synthetic memtables that do not have
a region_group attached to it.
Normally in cases like this we would just give the flush_reader an empty
region group. However, the memtable class constructor takes a
region_group pointer and that can be null according to the interface.
So we must conditionally test it.
If there isn't a region_group involved, the virtual dirty accounting
should be disabled: after all, we won't even have the baseline memory
to begin with.
One of the approaches to fix this could be to just provide null
accounter classes to be used as a surrogate for the accounting classes
in this case. However, since this is mostly used for tests, a much
simpler way is to just revert back to the scanning reader in that case.
The scanning reader is similar enough to the flush_reader, except that
it can handle partial ranges, slices, and delegate accesses to an
sstable post-flush. We don't need any of that, but as argued above,
there is no need to remove it either.
Signed-off-by: Glauber Costa <glommer@scylladb.com>
Message-Id: <1475667271-60806-1-git-send-email-glommer@scylladb.com>
Remove inclusions from header files (primary offender is fb_utilities.hh)
and introduce new messaging_service_fwd.hh to reduce rebuilds when the
messaging service changes.
Message-Id: <1475584615-22836-1-git-send-email-avi@scylladb.com>
"Description:
============
Scylla currently suffers from a brick wall behavior of the request throttler.
Requests pile up until we reach the dirty memory limit, at which point we stop
serving them until we have freed enough memory to allow for more requests.
The problem is that freeing dirty memory means writing an SSTable to completion.
That can take a long time, even if we are blessed with great disks. Those long
waiting times can and will translate into timeouts. That is bad behavior.
What this patch does is introduce one form of virtual dirty memory accounting.
Instead of allowing 100 % of the dirty memory to be filled up until we stop
accepting requests, we will do that when we reach 50 % of memory. However,
instead of releasing requests only when an SSTable is fully written, we start
releasing them when some memory was written.
The practical effect of that, is that once we reach 50 % occupancy in our dirty
memory region, we will bring the system from CPU speed to disk speed, and will
start accepting requests only at the rate we are able to write memory back.
Results
=======
With this patchset running a load big enough to easily saturate the disk,
(commitlog disabled to highlight the effects of the memtable writer), I am able
to run scylla for many minutes, with timeouts occurring only when I run out of
disk space, whereas without this patch a swarm of timeouts would start merely 2
seconds after the load started - and would never get stable.
In V2, I have sent a set of graphs illustrating the performance of this solution.
This version does not have any significant differences in that front.
For details, please refer to
https://groups.google.com/d/msg/scylladb-dev/iCvD-3Z-QqY/EM8KUh_MAQAJ
Accuracy of the accounting:
---------------------------
It is important for us to be as accurate as possible when accounting freed
memory, since every byte we mark as freed may allow one or more requests to be
executed. I have measured the accuracy of this approach (ignoring padding,
object size for the mutation fragments) to be 99.83 % of used memory in the
test workload I have ran (large, 65k mutations). Memtables under this circumnstance
tend to have a very high occupancy ratio because throttle breeds idle, and idle
breeds compact-on-idle.
Known Issues:
-------------
A lot of time can be elapsed between destroying the flush_reader and actually
releasing memory. The release of memory only happens when the SSTable is fully
sealed, and we have to flush the files, as well as finish writing all SSTable
components at this point. This happened in practice with a buggy kernel that
would result in flushes taking a long time.
After that is fixed, this is just a theoretical problem and in practice it
shouldn't matter given the time we expect those operations to take."
* 'virtual-dirty-v6' of github.com:glommer/scylla:
database: allow virtual dirty memory management
streamed_mutation: make _buffer private
add accounting of memory read to partition_snapshot_reader
move partition_snapshot_reader code to header file
LSA: allow a group to query its own region group
memtables: split scanning reader in two
sstables: use special reader for writing a memtable
LSA: export information about object memory footprint
LSA: export information about size of the throttle queue
database: export virtual dirty bytes region group
* seastar 18f7bb8...f937fb0 (5):
> Merge "Fix signal mask corruption" from Tomasz
> core/memory: Avoid violating strict aliasing when accessing allocation sites
> core/memory: Avoid indirection when storing allocation sites
> core/memory: Add a way to disable abort on allocation failure in some scope
> core/sharded: Allow mapper to take the service by non-const reference
Scylla currently suffers from a brick wall behavior of the request throttler.
Requests pile up until we reach the dirty memory limit, at which point we stop
serving them until we have freed enough memory to allow for more requests.
The problem is that freeing dirty memory means writing an SSTable to completion.
That can take a long time, even if we are blessed with great disks. Those long
waiting times can and will translate into timeouts. That is bad behavior.
What this patch does is introduce one form of virtual dirty memory accounting.
Instead of allowing 100 % of the dirty memory to be filled up until we stop
accepting requests, we will do that when we reach 50 % of memory. However,
instead of releasing requests only when an SSTable is fully written, we start
releasing them when some memory was written.
The practical effect of that is that once we reach 50 % occupancy in our dirty
memory region, we will bring the system from CPU speed to disk speed, and will
start accepting requests only at the rate we are able to write memory back.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
It is currently protected, but now all users go through
push_mutation_fragment(). So we can safely move its visibility to guarantee
that it stays that way.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
By default, we don't do any accounting. By specializing this class and providing
an accounter class, we can account how much memory are we reading as we read
through the elements.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
This is so we can template it without worrying about declaring the
specializations in the .cc file.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
The code that is common will live in its own reader, the iterator_reader. All
friendly private access to memtable attributes and methods happen through the
iterator reader.
After this patch, we are now left with the scanning_reader - same as always,
but now implemented on top of the iterator_reader, and a flush_reader, which
will be used by SSTable flushes only.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Right now the special reader doesn't do much, but the idea is that we will
soon replace it will a reader that specializes in flush, and is in turn able
to provide read-side on-flush functionality like virtual dirty.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
We allocate objects of a certain size, but we use a bit more memory to hold
them. To get a clerer picture about how much memory will an object cost us, we
need help from the allocator. This patch exports an interface that allow users
to query into a specific allocator to get that information.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
In order to allow Scylla’s docker container to handle multiple network
interfaces, the start-scylla script was refactored:
- `$IP` is now called `$SCYLLA_LISTEN_ADDRESS`, so it is less likely to
be confused or interfere with other environment variables.
- `$SCYLLA_LISTEN_ADDRESS` now checks its value and also tries to
resolve a hostname, if no IP was set to it.
- `$SCYLLA_LISTEN_DEVICE` can now be set as environment variable and
contain any available NIC device name (e.g. `eth0`). The script
automatically retrieves the IP address from the device.
Usage:
1. With `$SCYLLA_LISTEN_ADDRESS` as IP:
`docker run -t -i --rm --name scylla -e SCYLLA_LISTEN_ADDRESS=192.168.1.100 scylladb/scylla`
2. With `$SCYLLA_LISTEN_ADDRESS` as hostname:
`docker run -t -i --rm --name scylla -e SCYLLA_LISTEN_ADDRESS=containername.network.lan scylladb/scylla`
3. With `$SCYLLA_LISTEN_DEVICE`:
`docker run -t -i --rm --name scylla -e SCYLLA_LISTEN_DEVICE=eth0 scylladb/scylla`
Message-Id: <20161003151230.67672-1-marius@twostairs.com>
Currently, the code responsible for calculating ranges for the next
request could produce a wrap-around partition range. For example, if the
original range was (unimportant, A] and the last partition key A then
the output range would be (A, A].
This patch adds checks to make sure that in such cases the range is
removed.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1475497244-2790-1-git-send-email-pdziepak@scylladb.com>
cassandra_exception::prepare_message() is called from derived classes'
constructors before the base cassnadra_exception object is constructed.
This is technically illegal but harmless. Fix by marking the function
static.
Found by clang.
"This patch set ensures we can correctly handle queries
where the minimum token is specified."
* 'min-token/v3' of github.com:duarten/scylla:
cql_query_test: Add test case for min/max token bounds
token_restriction: Deal with minimum tokens
partitioner: Parse token from bytes
This object, similarly to a global_schema_ptr, allows to dynamically
create the trace_state_ptr objects on different shards in a context
of the original tracing session.
This object would create a secondary tracing session object from the
original trace_state_ptr object when a trace_state_ptr object is needed
on a "remote" shard, similarly to what we do when we need it on a remote
Node.
Fixes#1678Fixes#1647
Signed-off-by: Vlad Zolotarov <vladz@cloudius-systems.com>
Message-Id: <1474387767-21910-1-git-send-email-vladz@cloudius-systems.com>
This patch adds a call to the scylla-housekeeping check version during
setup, so a warning will be printed if a newer version is available.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
This changes are for running scylla during setup.
It contains the following changes:
1. get the current version from the command line (as the syclla does not
run at this stage).
2. It support a mode parameter in the command line to indicate that we
running during the installation.
3. It accept an external uuid that will be used with all interaction
with the check_version server.
Signed-off-by: Amnon Heiman <amnon@scylladb.com>
When max sstable size is increased, higher levels are suffering from
starvation because we decide to compact a given level if the following
calculation results in a number greater than 1.001:
level_size(L) / max_size_for_level_l(L)
Fixes#1720.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Uniform token range distribution across sstables in a level > 1 was broken,
because we were only choosing sstable with lowest first key, when compacting
a level > 0. This resulted in performance problem because L1->L2 may have a
huge overlap over time, for example.
Last compacted key will now be stored for each level to ensure sort of
"round robin" selection of sstables for compactions at level >= 1.
That's also done by C*, and they were once affected by it as described in
https://issues.apache.org/jira/browse/CASSANDRA-6284.
Fixes#1719.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Paging code assumes that clustering row range [a, a] contains only one
row which may not be true. Another problem is that it tries to use
range<> interface for dealing with clustering key ranges which doesn't
work because of the lack of correct comparator.
Refs #1446.
Fixes#1684.
Signed-off-by: Paweł Dziepak <pdziepak@scylladb.com>
Message-Id: <1475236805-16223-1-git-send-email-pdziepak@scylladb.com>
CQL server is supposed to throttle requests so that they don't
overflow memory. The problem is that it currently accounts for
request's memory only around reading of its frame from the connection
and not actual request execution. As a result too many requests may be
allowed to execute and we may run out of memory.
Fixes#1708.
Message-Id: <1475149302-11517-1-git-send-email-tgrabiec@scylladb.com>
This patch fixes a bug where queries such as the following are not
handled properly:
"SELECT * FROM ks.cf WHERE token(id) >
9207857967443869328 AND token(id) <= -9223372036854775808"
Here -9223372036854775808 represents the minimum token, which we were
just translating into a token with kind::key, thus returning incorrect
results.
Ref #1139
Ref #693Fixes#1717
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
This patch adds the from_bytes() function to the i_partitioner class,
whose purpose is parse a particular token and explicitly handle the
case when the minimum token is specified.
Signed-off-by: Duarte Nunes <duarte@scylladb.com>
Also add information about for how long has the oldest been sitting in the
queue. This is part of the backpressure work to allow us to throttle incoming
requests if we won't have memory to process them. Shortages can happen in all
sorts of places, and it is useful when designing and testing the solutions to
know where they are, and how bad they are.
This counter is named for consistency after similar counters from transport/.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
Currently, we export the region group where memtables are placed as dirty bytes.
Upcoming patches will optimistically mark some bytes in this region as free, a
scheme we know as "virtual dirty".
We are still interested in knowing the real state of the dirty region, so we
will keep track of the bytes virtually freed and split the counters in two.
Signed-off-by: Glauber Costa <glauber@scylladb.com>
2016-09-27 12:09:08 -04:00
890 changed files with 89449 additions and 28760 deletions
Use the [ScyllaDB user mailing list](https://groups.google.com/forum/#!forum/scylladb-users) for general questions and help.
# Reporting an issue
Please use the [Issue Tracker](https://github.com/scylladb/scylla/issues/) to report issues. Fill in as much information as you can in the issue template, especially for performance problems.
# Contributing Code to Scylla
To contribute code to Scylla, you need to sign the [Contributor License Agreement](http://www.scylladb.com/opensource/cla/) and send your changes as [patches](https://github.com/scylladb/scylla/wiki/Formatting-and-sending-patches) to the [mailing list](https://groups.google.com/forum/#!forum/scylladb-dev). We don't accept pull requests on GitHub.
This document is intended to help developers and contributors to Scylla get started. The first part consists of general guidelines that make no assumptions about a development environment or tooling. The second part describes a particular environment and work-flow for exemplary purposes.
## Overview
This section covers some high-level information about the Scylla source code and work-flow.
### Getting the source code
Scylla uses [Git submodules](https://git-scm.com/book/en/v2/Git-Tools-Submodules) to manage its dependency on Seastar and other tools. Be sure that all submodules are correctly initialized when cloning the project:
```bash
$ git clone https://github.com/scylladb/scylla
$ cd scylla
$ git submodule update --init --recursive
```
### Dependencies
Scylla depends on the system package manager for its development dependencies.
Running `./install_dependencies.sh` (as root) installs the appropriate packages based on your Linux distribution.
### Build system
**Note**: Compiling Scylla requires, conservatively, 2 GB of memory per native thread, and up to 3 GB per native thread while linking.
Scylla is built with [Ninja](https://ninja-build.org/), a low-level rule-based system. A Python script, `configure.py`, generates a Ninja file (`build.ninja`) based on configuration options.
To build for the first time:
```bash
$ ./configure.py
$ ninja-build
```
Afterwards, it is sufficient to just execute Ninja.
The full suite of options for project configuration is available via
```bash
$ ./configure.py --help
```
The most important options are:
-`--mode={release,debug,all}`: Debug mode enables [AddressSanitizer](https://github.com/google/sanitizers/wiki/AddressSanitizer) and allows for debugging with tools like GDB. Debugging builds are generally slower and generate much larger object files than release builds.
-`--{enable,disable}-dpdk`: [DPDK](http://dpdk.org/) is a set of libraries and drivers for fast packet processing. During development, it's not necessary to enable support even if it is supported by your platform.
Source files and build targets are tracked manually in `configure.py`, so the script needs to be updated when new files or targets are added or removed.
To save time -- for instance, to avoid compiling all unit tests -- you can also specify specific targets to Ninja. For example,
Unit tests live in the `/tests` directory. Like with application source files, test sources and executables are specified manually in `configure.py` and need to be updated when changes are made.
A test target can be any executable. A non-zero return code indicates test failure.
Most tests in the Scylla repository are built using the [Boost.Test](http://www.boost.org/doc/libs/1_64_0/libs/test/doc/html/index.html) library. Utilities for writing tests with Seastar futures are also included.
Run all tests through the test execution wrapper with
```bash
$ ./test.py --mode={debug,release}
```
The `--name` argument can be specified to run a particular test.
Alternatively, you can execute the test executable directly. For example,
```bash
$ build/release/tests/row_cache_test -- -c1 -m1G
```
The `-c1 -m1G` arguments limit this Seastar-based test to a single system thread and 1 GB of memory.
### Preparing patches
All changes to Scylla are submitted as patches to the public mailing list. Once a patch is approved by one of the maintainers of the project, it is committed to the maintainers' copy of the repository at https://github.com/scylladb/scylla.
Detailed instructions for formatting patches for the mailing list and advice on preparing good patches are available at the [ScyllaDB website](http://docs.scylladb.com/contribute/).
### Running Scylla
Once Scylla has been compiled, executing the (`debug` or `release`) target will start a running instance in the foreground:
```bash
$ build/release/scylla
```
The `scylla` executable requires a configuration file, `scylla.yaml`. By default, this is read from `$SCYLLA_HOME/conf/scylla.yaml`. A good starting point for development is located in the repository at `/conf/scylla.yaml`.
For development, a directory at `$HOME/scylla` can be used for all Scylla-related files:
The `scylla.yaml` file in the repository by default writes all database data to `/var/lib/scylla`, which likely requires root access. Change the `data_file_directories` and `commitlog_directory` fields as appropriate.
Scylla has a number of requirements for the file-system and operating system to operate ideally and at peak performance. However, during development, these requirements can be relaxed with the `--developer-mode` flag.
Additionally, when running on under-powered platforms like portable laptops, the `--overprovisined` flag is useful.
Multiple release branches are maintained on the Git repository at https://github.com/scylladb/scylla. Release 1.5, for instance, is tracked on the `branch-1.5` branch.
Similarly, tags are used to pin-point precise release versions, including hot-fix versions like 1.5.4. These are named `scylla-1.5.4`, for example.
Most development happens on the `master` branch. Release branches are cut from `master` based on time and/or features. When a patch against `master` fixes a serious issue like a node crash or data loss, it is backported to a particular release branch with `git cherry-pick` by the project maintainers.
## Example: development on Fedora 25
This section describes one possible work-flow for developing Scylla on a Fedora 25 system. It is presented as an example to help you to develop a work-flow and tools that you are comfortable with.
### Preface
This guide will be written from the perspective of a fictitious developer, Taylor Smith.
### Git work-flow
Having two Git remotes is useful:
- A public clone of Seastar (`"public"`)
- A private clone of Seastar (`"private"`) for in-progress work or work that is not yet ready to share
The first step to contributing a change to Scylla is to create a local branch dedicated to it. For example, a feature that fixes a bug in the CQL statement for creating tables could be called `ts/cql_create_table_error/v1`. The branch name is prefaced by the developer's initials and has a suffix indicating that this is the first version. The version suffix is useful when branches are shared publicly and changes are requested on the mailing list. Having a branch for each version of the patch (or patch set) shared publicly makes it easier to reference and compare the history of a change.
Setting the upstream branch of your development branch to `master` is a useful way to track your changes. You can do this with
As a patch set is developed, you can periodically push the branch to the private remote to back-up work.
Once the patch set is ready to be reviewed, push the branch to the public remote and prepare an email to the `scylladb-dev` mailing list. Including a link to the branch on your public remote allows for reviewers to quickly test and explore your changes.
### Development environment and source code navigation
Scylla includes a [CMake](https://cmake.org/) file, `CMakeLists.txt`, for use only with development environments (not for building) so that they can properly analyze the source code.
[CLion](https://www.jetbrains.com/clion/) is a commercial IDE offers reasonably good source code navigation and advice for code hygiene, though its C++ parser sometimes makes errors and flags false issues.
Other good options that directly parse CMake files are [KDevelop](https://www.kdevelop.org/) and [QtCreator](https://wiki.qt.io/Qt_Creator).
To use the `CMakeLists.txt` file with these programs, define the `FOR_IDE` CMake variable or shell environmental variable.
[Eclipse](https://eclipse.org/cdt/) is another open-source option. It doesn't natively work with CMake projects, and its C++ parser has many similar issues as CLion.
### Distributed compilation: `distcc` and `ccache`
Scylla's compilations times can be long. Two tools help somewhat:
- [ccache](https://ccache.samba.org/) caches compiled object files on disk and re-uses them when possible
- [distcc](https://github.com/distcc/distcc) distributes compilation jobs to remote machines
A reasonably-powered laptop acts as the coordinator for compilation. A second, more powerful, machine acts as a passive compilation server.
Having a direct wired connection between the machines ensures that object files can be transmitted quickly and limits the overhead of remote compilation.
The coordinator has been assigned the static IP address `10.0.0.1` and the passive compilation machine has been assigned `10.0.0.2`.
On Fedora, installing the `ccache` package places symbolic links for `gcc` and `g++` in the `PATH`. This allows normal compilation to transparently invoke `ccache` for compilation and cache object files on the local file-system.
Next, set `CCACHE_PREFIX` so that `ccache` is responsible for invoking `distcc` as necessary:
```bash
exportCCACHE_PREFIX="distcc"
```
On each host, edit `/etc/sysconfig/distccd` to include the allowed coordinators and the total number of jobs that the machine should accept.
This example is for the laptop, which has 2 physical cores (4 logical cores with hyper-threading):
`10.0.0.2` has 8 physical cores (16 logical cores) and 64 GB of memory.
As a rule-of-thumb, the number of jobs that a machine should be specified to support should be equal to the number of its native threads.
Restart the `distccd` service on all machines.
On the coordinator machine, edit `$HOME/.distcc/hosts` with the available hosts for compilation. Order of the hosts indicates preference.
```
10.0.0.2/16 localhost/2
```
In this example, `10.0.0.2` will be sent up to 16 jobs and the local machine will be sent up to 2. Allowing for two extra threads on the host machine for coordination, we run compilation with `16 + 2 + 2 = 20` jobs in total: `ninja-build -j20`.
When a compilation is in progress, the status of jobs on all remote machines can be visualized in the terminal with `distccmon-text` or graphically as a GTK application with `distccmon-gnome`.
One thing to keep in mind is that linking object files happens on the coordinating machine, which can be a bottleneck. See the next section speeding up this process.
### Using the `gold` linker
Linking Scylla can be slow. The gold linker can replace GNU ld and often speeds the linking process. On Fedora, you can switch the system linker using
```bash
$ sudo alternatives --config ld
```
### Testing changes in Seastar with Scylla
Sometimes Scylla development is closely tied with a feature being developed in Seastar. It can be useful to compile Scylla with a particular check-out of Seastar.
One way to do this it to create a local remote for the Seastar submodule in the Scylla repository:
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
