without C++23 `std::ranges::repeat_view`, it'd be cumbersume to
implement a loop without dummy variable. this change helps to
silence following warning:
```
test/boost/memtable_test.cc:1135:26: error: unused variable 'value' [-Werror,-Wunused-variable]
for (int value : boost::irange<int>(0, num_flushes)) {
^
```
Signed-off-by: Kefu Chai <kefu.chai@scylladb.com>
Schema related files are moved there. This excludes schema files that
also interact with mutations, because the mutation module depends on
the schema. Those files will have to go into a separate module.
Closes#12858
We currently have two method families to generate partition keys:
* make_keys() in test/lib/simple_schema.hh
* token_generation_for_shard() in test/lib/sstable_utils.hh
Both work only for schemas with a single partition key column of `text` type and both generate keys of fixed size.
This is very restrictive and simplistic. Tests, which wanted anything more complicated than that had to rely on open-coded key generation.
Also, many tests started to rely on the simplistic nature of these keys, in particular two tests started failing because the new key generation method generated keys of varying size:
* sstable_compaction_test.sstable_run_based_compaction_test
* sstable_mutation_test.test_key_count_estimation
These two tests seems to depend on generated keys all being of the same size. This makes some sense in the case of the key count estimation test, but makes no sense at all to me in the case of the sstable run test.
Closes#12657
* github.com:scylladb/scylladb:
test/lib/sstable_utils: remove now unused token_generation_for_shard() and friends
test/lib/simple_schema: remove now unused make_keys() and friends
test: migrate to tests::generate_partition_key[s]()
test/lib/test_services: add table_for_tests::make_default_schema()
test/lib: add key_utils.hh
test/lib/random_schema.hh: value_generator: add min_size_in_bytes
This series switches memtable and cache to use a new representation for mutation data,
called `mutation_partition_v2`. In this representation, range tombstone information is stored
in the same tree as rows, attached to row entries. Each entry has a new tombstone field,
which represents range tombstone part which applies to the interval between this entry and
the previous one. See docs/dev/mvcc.md for more details about the format.
The transient mutation object still uses the old model in order to avoid work needed to adapt
old code to the new model. It may also be a good idea to live with two models, since the
transient mutation has different requirements and thus different trade-offs can be made.
Transient mutation doesn't need to support eviction and strong exception guarantees,
so its algorithms and in-memory representation can be simpler.
This allows us to incrementally evict range tombstone information. Before this series,
range tombstones were accumulated and evicted only when the whole partition entry was evicted. This
could lead to inefficient use of cache memory.
Another advantage of the new representation is that reads don't have to lookup
range tombstone information in a different tree while reading. This leads to simpler
and more efficient readers.
There are several disadvantages too. Firstly, rows_entry is now larger by 16 bytes.
Secondly, update algorithms are more complex because they need to deoverlap range tombstone
information. Also, to handle preemption and provide strong exception guarantees, update
algorithms may need to allocate sentinel entries, which adds complexity and reduces performance.
The memtable reader was changed to use the same cursor implementation
which cache uses, for improved code reuse and reducing risk of bugs
due to discrepancy of algorithms which deal with MVCC.
Remaining work:
- performance optimizations to apply_monotonically() to avoid regressions
- performance testing
- preemption support in apply_to_incomplete (cache update from memtable)
Fixes#2578Fixes#3288Fixes#10587Closes#12048
* github.com:scylladb/scylladb:
test: mvcc: Extend some scenarios with exhaustive consistency checks on eviction
test: mvcc: Extract mvcc_container::allocate_in_region()
row_cache, lru: Introduce evict_shallow()
test: mvcc: Avoid copies of mutation under failure injection
test: mvcc: Add missing logalloc::reclaim_lock to test_apply_is_atomic
mutation_partition_v2: Avoid full scan when applying mutation to non-evictable
Pass is_evictable to apply()
tests: mutation_partition_v2: Introduce test_external_memory_usage_v2 mirroring the test for v1
tests: mutation: Fix test_external_memory_usage() to not measure mutation object footprint
tests: mutation_partition_v2: Add test for exception safety of mutation merging
tests: Add tests for the mutation_partition_v2 model
mutation_partition_v2: Implement compact()
cache_tracker: Extract insert(mutation_partition_v2&)
mvcc, mutation_partition: Document guarantees in case merging succeeds
mutation_partition_v2: Accept arbitrary preemption source in apply_monotonically()
mutation_partition_v2: Simplify get_continuity()
row_cache: Distinguish dummy insertion site in trace log
db: Use mutation_partition_v2 in mvcc
range_tombstone_change_merger: Introduce peek()
readers: Extract range_tombstone_change_merger
mvcc: partition_snapshot_row_cursor: Handle non-evictable snapshots
mvcc: partition_snapshot_row_cursor: Support digest calculation
mutation_partition_v2: Store range tombstones together with rows
db: Introduce mutation_partition_v2
doc: Introduce docs/dev/mvcc.md
db: cache_tracker: Introduce insert() variant which positions before existing entry in the LRU
db: Print range_tombstone bounds as position_in_partition
test: memtable_test: Relax test_segment_migration_during_flush
test: cache_flat_mutation_reader: Avoid timestamp clash
test: cache_flat_mutation_reader_test: Use monotonic timestamps when inserting rows
test: mvcc: Fix sporadic failures due to compact_for_compaction()
test: lib: random_mutation_generator: Produce partition tombstone less often
test: lib: random_utils: Introduce with_probability()
test: lib: Improve error message in has_same_continuity()
test: mvcc: mvcc_container: Avoid UB in tracker() getter when there is no tracker
test: mvcc: Insert entries in the tracker
test: mvcc_test: Do not set dummy::no on non-clustering rows
mutation_partition: Print full position in error report in append_clustered_row()
db: mutation_cleaner: Extract make_region_space_guard()
position_in_partition: Optimize equality check
mvcc: Fix version merging state resetting
mutation_partition: apply_resume: Mark operator bool() as explicit
We have enabled the command line options without changing a
single line of code, we only had to replace old include
with scylla_test_case.hh.
Next step is to add x-log-compaction-groups options, which will
determine the number of compaction groups to be used by all
instantiations of replica::table.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
With compaction groups, automatic flushing may not pick the user
table. Fix it by using explicit flush.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Use the newly introduced key generation facilities, instead of the the
old inflexible alternatives and hand-rolled code.
Most of the migrations are mechanic, but there are two tests that
were tricky to migrate:
* sstable_compaction_test.sstable_run_based_compaction_test
* sstable_mutation_test.test_key_count_estimation
These two tests seems to depend on generated keys all being of the same
size. This makes some sense in the case of the key count estimation
test, but makes no sense at all to me in the case of the sstable run
test.
Partition version merging can now insert sentinels, which may
temporarily increase unspooled memory. It is no longer true that
unspooled monotonically decreases, which the test verified. Relax it,
and only verify that unspooled is smaller than real dirty.
active_memtable() was fine to a single group, but with multiple groups,
there will be one active memtable per group. Let's change the
interface to reflect that.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
The latter is pretty popular test/lib header that disseminates the
former one over whole lot of unit tests. The former, in turn, naturally
includes sstables.hh thus making tons of unrelated tests depend on
sstables class unused by them.
However, simple removal doesn't work, becase of local_shard_only bool
class definition in sstable_utils.hh used in simple_schema.hh. This
thing, in turn, is used in keys making helpers that don't belong to
sstable utils, so these are moved into simple_schema as well.
When done, this affects the mutation_source_test.hh, which needs the
local_shard_only bool class (and helps spreading the sstables.hh
throughout more unrelated tests) and a bunch of .cc test sources that
used sstable_utils.hh to indirectly include various headers of their
demand.
After patching, sstables.hh touches 2x times less tests. As a side
effect the sstables_manager.hh also becomes 2x times less dependent
on by tests.
Continuation of 9bdea110a6
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Closes#12240
The "virtual dirty" term is not very informative. "Virtual" means
"not real", but it doesn't say in which way it isn't real.
In this case, virtual dirty refers to real dirty memory, minus
the portion of memtables that has been written to disk (but not
yet sealed - in that case it would not be dirty in the first
place).
I chose to call "the portion of memtables that has been written
to disk" as "spooled memory". At least the unique term will cause
people to look it up and may be easier to remember. From that
we have "unspooled memory".
I plan to further change the accounting to account for spooled memory
rather than unspooled, as that is a more natural term, but that is left
for later.
The documentation, config item, and metrics are adjusted. The config
item is practically unused so it isn't worth keeping compatibility here.
The test simulates a situation where 2 threads issue flushes to 2
tables. Both issue small flushes, but one has injected reactor stalls.
This can lead to a situation where lots of small sstables accumulate on
disk, and, if compaction never has a chance to keep up, resources can be
exhausted.
(cherry picked from commit b5684aa96d)
(cherry picked from commit 25407a7e41)
Now that memtable flush error handling was moved entirely
to table::seal_active_memtable, we don't need to notify_soft_pressure
to keep retry going. The inifinite retry loop should
eventually either succeed or die (by isolating the node or aborting)
on its own.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
This reverts commit aa8f135f64, reversing
changes made to 9a88bc260c. The patch
causes hangs during flush.
Also reverts parts of 411231da75 that impacted the unit test.
Fixes#10897.
There is a bug introduced in e74c3c8 (4.6.0) which makes memtable
reader skip one a range tombstone for a certain pattern of deletions
and under certain sequence of events.
_rt_stream contains the result of deoverlapping range tombstones which
had the same position, which were sipped from all the versions. The
result of deoverlapping may produce a range tombstone which starts
later, at the same position as a more recent tombstone which has not
been sipped from the partition version yet. If we consume the old
range tombstone from _rt_stream and then refresh the iterators, the
refresh will skip over the newer tombstone.
The fix is to drop the logic which drains _rt_stream so that
_rt_stream is always merged with partition versions.
For the problem to trigger, there have to be multiple MVCC versions
(at least 2) which contain deletions of the following form:
[a, c] @ t0
[a, b) @ t1, [b, d] @ t2
c > b
The proper sequence for such versions is (assuming d > c):
[a, b) @ t1,
[b, d] @ t2
Due to the bug, the reader will produce:
[a, b) @ t1,
[b, c] @ t0
The reader also needs to be preempted right before processing [b, d] @
t2 and iterators need to get invalidated so that
lsa_partition_reader::do_refresh_state() is called and it skips over
[b, d] @ t2. Otherwise, the reader will emit [b, d] @ t2 later. If it
does emit the proper range tombstone, it's possible that it will violate
fragment order in the stream if _rt_stream accumulated remainders
(possible with 3 MVCC versions).
The problem goes away once MVCC versions merge.
Fixes#10913Fixes#10830Closes#10914
When memtable receives a tombstone it can happen under some workloads
that it covers data which is still in the memtable. Some workloads may
insert and delete data within a short time frame. We could reduce the
rate of memtable flushes if we eagerly drop tombstoned data.
One workload which benefits is the raft log. It stores a row for each
uncommitted raft entry. When entries are committed they are
deleted. So the live set is expected to be short under normal
conditions.
Fixes#652.
Closes#10807
* github.com:scylladb/scylla:
memtable: Add counters for tombstone compaction
memtable, cache: Eagerly compact data with tombstones
memtable: Subtract from flushed memory when cleaning
mvcc: Introduce apply_resume to hold state for partition version merging
test: mutation: Compare against compacted mutations
compacting_reader: Drop irrelevant tombstones
mutation_partition: Extract deletable_row::compact_and_expire()
mvcc: Apply mutations in memtable with preemption enabled
test: memtable: Make failed_flush_prevents_writes() immune to background merging
Before the change, the test artificiallu set the soft pressure
condition hoping that the background flusher will flush the
memtable. It won't happen if by the time the background flusher runs
the LSA region is updated and soft pressure (which is not really
there) is lifted. Once apply() becomes preemptibe, backgroun partition
version merging can lift the soft pressure, making the memtable flush
not occur and making the test fail.
Fix by triggering soft pressure on retries.
Fixes#10801
Refs #10793
(cherry picked from commit 0e78ad50ea)
Closes#10802
When memtable receives a tombstone it can happen under some workloads
that it covers data which is still in the memtable. Some workloads may
insert and delete data within a short time frame. We could reduce the
rate of memtable flushes if we eagerly drpo tombstoned data.
One workload which benefits is the raft log. It stores a row for each
uncommitted raft entry. When entries are committed they are
deleted. So the live set is expected to be short under normal
conditions.
Fixes#652.
This patch prevents virtual dirty from going negative during memtable
flush in case partition version merging erases data previously
accounted by the flush reader. There is an assert in
~flush_memory_accounter which guards for this.
This will start happening after tombstones are compacted with rows on
partition version merging.
This problem is prevented by the patch by having the cleaner notify
the memtable layer via callback about the amount of dirty memory released
during merging, so that the memtable layer can adjust its accounting.
Before the change, the test artificiallu set the soft pressure
condition hoping that the background flusher will flush the
memtable. It won't happen if by the time the background flusher runs
the LSA region is updated and soft pressure (which is not really
there) is lifted. Once apply() becomes preemptibe, backgroun partition
version merging can lift the soft pressure, making the memtable flush
not occur and making the test fail.
Fix by triggering soft pressure on retries.
The test simulates a situation where 2 threads issue flushes to 2
tables. Both issue small flushes, but one has injected reactor stalls.
This can lead to a situation where lots of small sstables accumulate on
disk, and, if compaction never has a chance to keep up, resources can be
exhausted.
This reverts commit e0670f0bb5, reversing
changes made to 605ee74c39. It causes failures
in debug mode in
database_test.test_database_with_data_in_sstables_is_a_mutation_source_plain,
though with low probability.
Fixes#10780Reopens#652.
active_memtable().empty() becomes true once seal_active_memtable
succeeds with _memtables->add_memtable(), not when it is able
to flush the (once active) memtable.
In contrast, min_memtable_timestamp() returns api::max_timestamp
only if there is no data in any memtable.
Fixes#10793
Backport notes:
- Introduced in f6d9d6175f (currently in
branch-5.0)
- backport requires also 0e78ad50ea
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Closes#10798
When memtable receives a tombstone it can happen under some workloads
that it covers data which is still in the memtable. Some workloads may
insert and delete data within a short time frame. We could reduce the
rate of memtable flushes if we eagerly drpo tombstoned data.
One workload which benefits is the raft log. It stores a row for each
uncommitted raft entry. When entries are committed they are
deleted. So the live set is expected to be short under normal
conditions.
Fixes#652.
This patch prevents virtual dirty from going negative during memtable
flush in case partition version merging erases data previously
accounted by the flush reader. There is an assert in
~flush_memory_accounter which guards for this.
This will start happening after tombstones are compacted with rows on
partition version merging.
This problem is prevented by the patch by having the cleaner notify
the memtable layer via callback about the amount of dirty memory released
during merging, so that the memtable layer can adjust its accounting.
Before the change, the test artificiallu set the soft pressure
condition hoping that the background flusher will flush the
memtable. It won't happen if by the time the background flusher runs
the LSA region is updated and soft pressure (which is not really
there) is lifted. Once apply() becomes preemptibe, backgroun partition
version merging can lift the soft pressure, making the memtable flush
not occur and making the test fail.
Fix by triggering soft pressure on retries.
The flat_mutation_reader files were conflated and contained multiple
readers, which were not strictly necessary. Splitting optimizes both
iterative compilation times, as touching rarely used readers doesn't
recompile large chunks of codebase. Total compilation times are also
improved, as the size of flat_mutation_reader.hh and
flat_mutation_reader_v2.hh have been reduced and those files are
included by many file in the codebase.
With changes
real 29m14.051s
user 168m39.071s
sys 5m13.443s
Without changes
real 30m36.203s
user 175m43.354s
sys 5m26.376s
Closes#10194
Memtables are a replica-side entity, and so are moved to the
replica module and namespace.
Memtables are also used outside the replica, in two places:
- in some virtual tables; this is also in some way inside the replica,
(virtual readers are installed at the replica level, not the
cooordinator), so I don't consider it a layering violation
- in many sstable unit tests, as a convenient way to create sstables
with known input. This is a layering violation.
We could make memtables their own module, but I think this is wrong.
Memtables are deeply tied into replica memory management, and trying
to make them a low-level primitive (at a lower level than sstables) will
be difficult. Not least because memtables use sstables. Instead, we
should have a memtable-like thing that doesn't support merging and
doesn't have all other funky memtable stuff, and instead replace
the uses of memtables in sstable tests with some kind of
make_flat_mutation_reader_from_unsorted_mutations() that does
the sorting that is the reason for the use of memtables in tests (and
live with the layering violation meanwhile).
Test: unit (dev)
Closes#10120
`announce` now takes a `group0_guard` by value. `group0_guard` can only
be obtained through `migration_manager::start_group0_operation` and
moved, it cannot be constructed outside `migration_manager`.
The guard will be a method of ensuring linearizability for group 0
operations.
The functions which prepare schema change mutations (such as
`prepare_new_column_family_announcement`) would use internally
generated timestamps for these mutations. When schema changes are
managed by group 0 we want to ensure that timestamps of mutations
applied through Raft are monotonic. We will generate these timestamps at
call sites and pass them into the `prepare_` functions. This commit
prepares the APIs.
Instead of lengthy blurbs, switch to single-line, machine-readable
standardized (https://spdx.dev) license identifiers. The Linux kernel
switched long ago, so there is strong precedent.
Three cases are handled: AGPL-only, Apache-only, and dual licensed.
For the latter case, I chose (AGPL-3.0-or-later and Apache-2.0),
reasoning that our changes are extensive enough to apply our license.
The changes we applied mechanically with a script, except to
licenses/README.md.
Closes#9937
dirty_memory_manager monitors memory and triggers memtable flushing if
there is too much pressure. If bad_alloc happens during the flush, it
may break the loop and flushes won't be triggered automatically, leading
to blocked writes as memory won't be automatically released.
The solution is to add exception handling to the loop, so that the inner
part always returns a non-exceptional future (meaning the loop will
break only on node shutdown).
try/catch is used around on_internal_error instead of
on_internal_error_noexcept, as the latter doesn't have a version that
accepts an exception pointer. To get the exception message from
std::exception_ptr a rethrow is needed anyway, so this was a simpler
approach.
Fixes: #4174
Message-Id: <20220114082452.89189-1-mikolaj.sieluzycki@scylladb.com>
Move replica-oriented classes to the replica namespace. The main
classes moved are ::database, ::keyspace, and ::table, but a few
ancillary classes are also moved. There are certainly classes that
should be moved but aren't (like distributed_loader) but we have
to start somewhere.
References are adjusted treewide. In many cases, it is obvious that
a call site should not access the replica (but the data_dictionary
instead), but that is left for separate work.
scylla-gdb.py is adjusted to look for both the new and old names.
The database, keyspace, and table classes represent the replica-only
part of the objects after which they are named. Reading from a table
doesn't give you the full data, just the replica's view, and it is not
consistent since reconciliation is applied on the coordinator.
As a first step in acknowledging this, move the related files to
a replica/ subdirectory.
The gc_grace_seconds is a very fragile and broken design inherited from
Cassandra. Deleted data can be resurrected if cluster wide repair is not
performed within gc_grace_seconds. This design pushes the job of making
the database consistency to the user. In practice, it is very hard to
guarantee repair is performed within gc_grace_seconds all the time. For
example, repair workload has the lowest priority in the system which can
be slowed down by the higher priority workload, so that there is no
guarantee when a repair can finish. A gc_grace_seconds value that is
used to work might not work after data volume grows in a cluster. Users
might want to avoid running repair during a specific period where
latency is the top priority for their business.
To solve this problem, an automatic mechanism to protect data
resurrection is proposed and implemented. The main idea is to remove the
tombstone only after the range that covers the tombstone is repaired.
In this patch, a new table option tombstone_gc is added. The option is
used to configure tombstone gc mode. For example:
1) GC a tombstone after gc_grace_seconds
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'timeout'} ;
This is the default mode. If no tombstone_gc option is specified by the
user. The old gc_grace_seconds based gc will be used.
2) Never GC a tombstone
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'disabled'};
3) GC a tombstone immediately
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'immediate'};
4) GC a tombstone after repair
cqlsh> ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'repair'};
In addition to the 'mode' option, another option 'propagation_delay_in_seconds'
is added. It defines the max time a write could possibly delay before it
eventually arrives at a node.
A new gossip feature TOMBSTONE_GC_OPTIONS is added. The new tombstone_gc
option can only be used after the whole cluster supports the new
feature. A mixed cluster works with no problem.
Tests: compaction_test.py, ninja test
Fixes#3560
[avi: resolve conflicts vs data_dictionary]
Mutations are not guaranteed to come in the order of their timestamps.
If there is an expired tombstone in the sstable and a repair inserts old
data into memtable, the compaction would not consider memtable data and
purge the tombstone leading to data resurrection. The solution is to
disallow purging tombstones newer than min memtable timestamp.
When memtable contains both mutations and tombstones that delete them,
the output flushed to sstables contains both mutations. Inserting a
compacting reader results in writing smaller sstables and saves
compaction work later.
Performance tests of this change have shown a regression in a common
case where there are no deletes. A heuristic is employed to skip
compaction unless there are tombstones in the memtable to minimise
the impact of that issue.
Add full compaction in test_memtable_with_many_versions_conforms_to_mutation_source
in background. Without it, some paths in the partition snapshot reader
weren't covered, as the tests always managed to read all range
tombstones and rows which cover a given clustering range from just a
single snapshot. Now, when full_compaction happens in process of reading
from a clustering range, we can force state refresh with non-nullopt
positions of last row and last range tombstone.
Note: this inability to test affected only the reversing reader.
