Just remove static mark and stop using global qctx.
Grab config from _db instead of argument while at it.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Before patching system_keyspace methods to use query processor from
its instance, the respective call is needed.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
It's called only on start and actively uses both qctx and local
cache. Next patches will fix the whole setup code to stop using
global qctx/cache.
For now setup invocation is left in its place, but it must really
happen in start() method. More patching is needed to make it work.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
For now it's a reference, but all users of the cache will be
eventually switched into using system_keyspace.
In cql-test-env cache starting happens earlier than it was
before, but that's OK, it just initializes empty instances.
In main cache starts at the same time as before patching.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
Start happens at exactly the same place. One thing to take care
of is that it happens on all shards.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
The db::system_keyspace was made a class some time ago, time to create
a standard sharded<> object out of it. It needs query processor and
database. None of those depensencies is started early enough, so the
object for now starts in two steps -- early instances creation and
late start.
The instances will carry qctx and local_cache on board and all the
services that need those two will depend on system-keyspace. Its start
happens at exactly the same place where system_keyspace::setup happens
thus any service that will use system_keyspace will be on the same
safe side as it is now.
In the further future the system_keyspace will be equpped with its
own query processor backed by local replica database instance, instead
of the whole storage proxy as it is now.
Signed-off-by: Pavel Emelyanov <xemul@scylladb.com>
is_supported_by checks whether a given restriction
can be supported by some index.
Currently when a subscripted value, e.g `m[1]` is encountered,
we ignore the fact that there is a subscript and ask
whether an index can support the `m` itself.
This looks like unintentional behaviour leftover
from the times when column_value had a sub field,
which could be easily forgotten about.
Scylla doesn't support indexes on collection elements at all,
so simply returning false there seems like a good idea.
Signed-off-by: Jan Ciolek <jan.ciolek@scylladb.com>
Closes#10227
Permits have to wait for re-admission after having been evicted. This
happens via `reader_permit::maybe_wait_readmission()`. The user of this
method -- the evictable reader -- uses it to re-wait admission when the
underlying reader was evicted. There is one tricky scenario however,
when the underlying reader is created for the first time. When the
evictable reader is part of a multishard query stack, the created reader
might in fact be a resumed, saved one. These readers are kept in an
inactive state until actually resumed. The evictable reader shares it
permit with the to-be-resumed reader so it can check whether it has been
evicted while saved and needs to wait readmission before being resumed.
In this flow it is critical that there is no preemption point between
this check and actually resuming the reader, because if there is, the
reader might end up actually recreated, without having waited for
readmission first.
To help avoid this situation, the existing `maybe_wait_readmission()` is
split into two methods:
* `bool reader_permit::needs_readmission()`
* `future<> reader_permit::wait_for_readmission()`
The evictable reader can now ensure there is no preemption point between
`needs_readmission()` and resuming the reader.
Fixes: #10187
Tests: unit(release)
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <20220315105851.170364-1-bdenes@scylladb.com>
"
Namely the query result writer and the reconcilable result builder, used
for building results for regular queries and mutation queries (used in
read repair) respectively.
With this, there are no users left for the v1 output of the compactor,
so we remove that, making the compactor v2 all-the-way (and simpler).
This means that for regular queries, a downgrade phase is eliminated
completely, as regular queries don't store range tombstone in their
result, so no need to convert them.
Tests: unit(dev, release, debug)
"
* 'result-builders-v2/v1' of https://github.com/denesb/scylla:
reconcilable_result_builder: remove v1 support
query_result_builder: remove v1 support
mutation_compactor: drop v1 related code-paths
mutation_compactor: drop v1 support altogether from the API
tree: migrate to the v2 consumer APIs
test/boost/mutation_test: remove v1 specific test code
querier: switch to v2 compactor output
reconcilable_result_builder: add v2 support
query_result_writer: add v2 support
query_result_builder: make consume(range_tombstone) noop
The `result_try` and `result_futurize_try` are supposed to handle both
failed results and exceptions in a way similar to a try..catch block.
In order to catch exceptions, the metaprogramming machinery invokes the
fallible code inside a stack of try..catch blocks, each one of them
handling one exception. This is done instead of creating a single
try..catch block, as to my knowledge it is not possible to create
a try..catch block with the number of "catch" clauses depending on a
variadic template parameter pack.
Unfortunately, a "try" with multiple "catches" is not functionally
equivalent to a "try block stack". Consider the following code:
try {
try {
return execute_try_block();
} catch (const derived_exception&) {
// 1
}
} catch (const base_exception&) {
// 2
}
If `execute_try_block` throws `derived_exception` and the (1) catch
handler rethrows this exception, it will also be handled in (2), which
is not the same behavior as if the try..catch stack was "flat".
This causes wrong behavior in `result_try` and `result_futurize_try`.
The following snippet has the same, wrong behavior as the previous one:
return utils::result_try([&] {
return execute_try_block();
}, utils::result_catch<derived_exception>([&] (const auto&& ex) {
// 1
}), utils::result_catch<base_exception>([&] (const auto&& ex) {
// 2
});
This commit fixes the problem by adding a boolean flag which is set just
before a catch handler is executed. If another catch handler is
accidentally matched due to exception rethrow, the catch handler is
skipped and exception is automatically rethrown.
Tests: unit(dev, debug)
Fixes: #10211Closes#10216
In commit afab1a97c6, we added
test_tools.py - tests for the various tools embedded in the Scylla
executable. These tests need to know where the Scylla executable is,
and also where its sstables are stored. For this, the commit added two
test parameters - "--scylla-path" and "--workdir" - with which the
"run" script communicated this knowledge to the test.
However, that implementation meant that these tests only work if the
test was run via the test/cql-pytest/run script - they won't work if
the user ran Scylla/pytest manually, or through some other script not
passing these options.
This patch drops the "--scylla-path" and "--workdir" parameters, and
instead the test figures out this information on its own:
1. To find the Scylla executable, we begin by looking (using the
local_process_id(cql) function from the previous patch) for a
local process which listens to our CQL connection, and then find
the executable's path using /proc.
2. To find the Scylla data directory (which is what we really need, not
workdir which is just a shortcut to set all directories!), we
retrieve this configuration from the system.config table through CQL.
I tested that test_tools.py now works not only through test/cql-pytest/run
but also if I run Scylla manually and then run "pytest test_tools.py"
without any extra parameters.
Fixes#10209
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20220314151125.2737815-2-nyh@scylladb.com>
Generally, cql-pytest tests do not, and *should not* rely on looking up
messages in the Scylla log file: Relying on such messages makes it
impossible to run the same test against Cassandra or even a remotely-
installed Scylla, and the tests tend to break when logging (which is not
considered part of our API) changes. Moreover, usually what our dtests
achieve by looking at the log - e.g., figuring out when some event has
happened - can be achieved through official CQL APIs, and this is what
normal users do anyway (users don't normally dig through the log to
figure out when their operation completed).
However, sometimes we do want to write a test to confirm that during a
certain operation, a certain log message gets written to Scylla's log.
A desire to do this was raised by @fruch and @soyacz, so in this patch
I provide a mechanism to do this, and a trivial example - which checks
that a "Creating ..." message appears on the log whenever a table is
created, and "Dropping ..." when the table is deleted.
As is explained in detail in patches in the comment, Scylla's log file
is found automatically, without relying on Scylla's runner (such as
the script test/cql-pytest/run) communicating to the test where the log
file is. If the log file can't be found - e.g., we're testing a remote
Scylla, or if this isn't Scylla, the tests are skipped.
I would like all logfile-testing tests to be in the same file,
test_logs.py. As I explained above, I think it is a mistake for general
tests to check the log file just because they can. I think that the only
tests that should use the log file are tests deliberately written to
check what gets logged - and those can be collected in the same file.
As part of this patch, we add the utility function local_process_id(cql)
to find (if we can) the local process which listens to the connection
"cql". This utility function will later be useful in more places - for
example test_tools.py needs to find Scylla's executable.
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20220314151125.2737815-1-nyh@scylladb.com>
To make major compaction more resilient to low-
disk space conditions, 342bfbd65a
sorted the tables based on their live disk space used.
However, each shard still makes progress in its own pace.
This change serializes major compaction between tables
so we still compact in parallel on all shards, but one
(distributed) table at a time.
As a follow-up, we can consider serializing even at the single shard
level when disk space is critically low, so we can't even risk
parallel compaction across all shards.
Refs scylladb/scylla-dtest#2653
Test: unit(dev)
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Message-Id: <20220313153814.2203660-1-bhalevy@scylladb.com>
With compact_sstables() now living in compaction_manager::task,
release_exhausted no longer has to live inside compaction_descriptor,
which is a good direction because implementation detail is being
removed from the interface.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20220311023410.250149-2-raphaelsc@scylladb.com>
Table submits compaction request into manager, which in turn calls
back table to run the compaction when the time has come, i.e.:
table -> compaction manager -> table -> execute compaction
But manager should not rely on table to run compaction, as compaction
execution procedure sits one layer below the manager and should be
accessed directly by it, i.e:
table -> compaction manager -> execute compaction
This makes code easier to understand and update_compaction_history()
can now be noop for unit tests using table_state.
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20220311023410.250149-1-raphaelsc@scylladb.com>
"
This series hardens raft_group_registry::stop_servers
and uses it to drain_on_shutdown, called before
the database is stopped in cql_test_env.
(Not needed for main).
raft_group_registry deferred_stop is introduced right after
the service is started to make sure it's properly stopped
even if there's an exception at any point while starting.
Test: unit(dev)
"
* tag 'raft_group_registry-drain_on_shutdown-v1' of https://github.com/bhalevy/scylla:
cql_test_env: raft_group_registry::drain_on_shutdown before stopping the database
raft_group_registry: harden stop_servers
raft_group_registry: delete unused _shutdown_gate
When row_cache::make_reader() and memtable::make_flat_reader() see that the query result is empty, they return empty_flat_reader, which is a trivial implementation of flat_mutation_reader.
Even though empty_flat_reader doesn't do anything meaningful, it still needs to be created, handled in merging_reader and destroyed. Turns out this is costly.
This patch series replaces hot path uses of empty_flat_reader with an empty optional.
Performance effects:
`perf_simple_query --smp 1`
TPS: 138k -> 168k
allocs/op: 80.2 -> 71.1
insns/op: 49.9k -> 45.1k
`perf_simple_query --smp 1 --enable-cache=1 --flush`
TPS: 125k -> 150k
allocs/op: 79.2 -> 71.1
insns/op: 51.7k -> 47.2k
For a cassandra-stress benchmark (localhost, 100% cache reads) this translates to a TPS increase from ~42k to ~48k per hyperthread.
Note that this optimization is effective for single-partition reads where the queried partition is only in cache/sstables or only in memtables. Other queries (e.g. where the partition is in both cache in memtables and needs to be merged) are unaffected.
Closes#10204
* github.com:scylladb/scylla:
replica: Prefer row_cache::make_reader_opt() to row_cache::make_reader()
row_cache: Add row_cache::make_reader_opt()
replica: Prefer memtable::make_flat_reader_opt() to memtable::make_flat_reader()
memtable: Add memtable::make_flat_reader_opt()
[avi: adjust #include for readers/ split]
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
When there is nothing to read, make_flat_reader() returns an empty (no-op)
reader. But it turns out that constructing, combining and destroying that
empty reader is quite costly.
As an optimization, add an alternative version which returns an empty optional
instead.
database
We're currently stopping raft_gr before
shutting the database down, but we fail to do that if
anything goes wrong before that, e.g. if
distributed_loader::init_non_system_keyspaces fails.
This change splits drain_on_shutdown out of stop()
to stop the raft groups before the database is stopped
and does the rest in a deferred_stop placed right
after the rafr_gr registry is strated.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
stop_servers should never fail since it's called on
the shutdown path.
Use a local gate in stop_servers() to wait on all
background raft group server aborts.
Also, handle theoretical exceptions from server::abort()
to guarantee success.
Signed-off-by: Benny Halevy <bhalevy@scylladb.com>
Recent PR #10092 (propagating read timeouts on coordinator without
throwing) accidentally removed a line which cancelled
`abstract_read_resolver`'s `_timeout` timer after a read failure.
Because of that, it might happen that after a read failure the timer is
triggered and the `_done_promise` is set twice which triggers an assert
in seastar.
This commit brings back the line which cancels the timeout timer.
Fixes: #10193Closes#10206
This is a translation of Cassandra's CQL unit test source file
validation/operations/BatchTest.java into our our cql-pytest framework.
This test file includes 13 tests for various types of BATCH operations.
All tests pass on Scylla - no known or new bugs were reproduced.
Two of the tests involve very slow testing of TTLs, so after verifying
they work I marked them "skip" for now (we can always turn them on later,
perhaps after reducing the length or number of the sleeps).
Signed-off-by: Nadav Har'El <nyh@scylladb.com>
Message-Id: <20220313121634.2611423-1-nyh@scylladb.com>
Commit 1c99ed6ced added tracing logs
about the index chosen for the query, but aggregate queries have
a separate code path, which wasn't taken into account.
After this patch, tracing for aggregate queries also includes
this additional information.
Closes#10195
interrupt() makes it sound like it's interrupting the compaction, but it's
actually called *on* interrupt, to handle the interrupt scenario.
Let's rename it to on_interrupt().
Signed-off-by: Raphael S. Carvalho <raphaelsc@scylladb.com>
Message-Id: <20220311000128.189840-1-raphaelsc@scylladb.com>
The change is mostly mechanical: update all compactor instances to the
_v2 variant and update all call-sites, of which there is not that many.
As a consequence of this patch, queries -- both single-partition and
range-scans -- now do the v2->v1 conversion in the consumers, instead of
in the compactor.
Add a `consume()` overload for range tombstone changes and convert them
internally to range tombstones, as the underlying reconcilable result
is still v1.
Add a consume() overload which takes a range tombstone change and drops
it just like the existing range tombstone overload does: query results
don't care about range tombstones.
The downstream consumer (mutation_querier) already ignores range
tombstones, so no point forwarding them to it. This makes adding v2
support easier too as range tombstone changes can be similarly dropped.
Changing the capture list of a lambda in
forward_service::execute_on_this_shard from [&] to an explicit one
enables grater readability and prevents potential bugs.
Closes#10191
The services' configuration should be performed with the help of
service-specific config that's filled by the service creator. This
is not the case for gossiper that grabs the db::config and keeps
reference on it throughout its lifetime.
This set brings the gossiper configuration to the described form
by putting the needed config bits onto gossip_config (that already
exists and is partially used for gossiper configuration). And two
live-updateable options need extra care.
tests: unit(dev), dtest.simple_boot_shutdown(dev)
* 'br-gossiper-no-db-config' of https://github.com/xemul/scylla:
gossiper: Remove db::config reference from gossiper
gossiper: Keep live-updateable options on gossiper
gossiper: Keep immutable options on gossip_config
Although its API was long converted to v2, its implementation stayed v1
because the memtable and mutation API were still v1. Now that the
memtable flush returns a v2 reader we can have a second look at
converting this. While the mutation API still uses v1, this can easily
be worked around by using going through `mutation_rebuilder_v2`.
Signed-off-by: Botond Dénes <bdenes@scylladb.com>
Message-Id: <20220302145945.189607-1-bdenes@scylladb.com>
