CI founds [1]:
Exception: Sanitizer assert found for instance ==================
WARNING: ThreadSanitizer: data race (pid=348)
Write of size 8 at 0x7b58000044a0 by main thread:
2 DB::DatabaseReplicated::startupTables(ThreadPoolImpl<ThreadFromGlobalPoolImpl<false>>&, DB::LoadingStrictnessLevel) build_docker/./src/Databases/DatabaseReplicated.cpp:526:16 (clickhouse+0x1ec45092)
3 DB::TablesLoader::startupTables() build_docker/./src/Databases/TablesLoader.cpp:87:26 (clickhouse+0x1f9258ab) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
4 DB::loadMetadata(std::__1::shared_ptr<DB::Context>, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char>> const&) build_docker/./src/Interpreters/loadMetadata.cpp:234:12 (clickhouse+0x1fff3834) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
5 DB::Server::main() build_docker/./programs/server/Server.cpp:1615:9 (clickhouse+0x163e7f78) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
6 Poco::Util::Application::run() build_docker/./base/poco/Util/src/Application.cpp:315:8 (clickhouse+0x257608fe) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
7 DB::Server::run() build_docker/./programs/server/Server.cpp:391:25 (clickhouse+0x163d7d7c) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
8 Poco::Util::ServerApplication::run(int, char**) build_docker/./base/poco/Util/src/ServerApplication.cpp:131:9 (clickhouse+0x25780114) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
9 mainEntryClickHouseServer(int, char**) build_docker/./programs/server/Server.cpp:196:20 (clickhouse+0x163d4c23) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
10 main build_docker/./programs/main.cpp:487:12 (clickhouse+0xdf8c877) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
Previous read of size 8 at 0x7b58000044a0 by thread T27 (mutexes: write M0, write M1):
1 DB::DatabaseReplicated::canExecuteReplicatedMetadataAlter() const build_docker/./src/Databases/DatabaseReplicated.cpp:1303:12 (clickhouse+0x1ec5c5bd)
2 DB::ReplicatedMergeTreeQueue::shouldExecuteLogEntry() const build_docker/./src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp:1471:24 (clickhouse+0x2115fb56) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
3 DB::ReplicatedMergeTreeQueue::selectEntryToProcess(DB::MergeTreeDataMergerMutator&, DB::MergeTreeData&) build_docker/./src/Storages/MergeTree/ReplicatedMergeTreeQueue.cpp:1676:13 (clickhouse+0x21163c58) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
4 DB::StorageReplicatedMergeTree::selectQueueEntry() build_docker/./src/Storages/StorageReplicatedMergeTree.cpp:3240:26 (clickhouse+0x20823db2) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
5 DB::StorageReplicatedMergeTree::scheduleDataProcessingJob(DB::BackgroundJobsAssignee&) build_docker/./src/Storages/StorageReplicatedMergeTree.cpp:3304:65 (clickhouse+0x208240fc) (BuildId: 7d4ce55d33d4c3e3df9fd39b304e67e53eb61a63)
[1]: https://s3.amazonaws.com/clickhouse-test-reports/52395/0b258dda4ee618a4d002e2b5246d68bbd2c77c7e/integration_tests__tsan__[5_6].html
Add ddl_worker_initialized flag to avoid this race.
Note, that it should be enough to check this flag only in
canExecuteReplicatedMetadataAlter() since only it can be run in parallel
with ctor before it had been finished.
v0: initialize ddl before startupTables()
v2: ddl_worker_initialized
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
Co-authored-by: Alexander Tokmakov <tavplubix@clickhouse.com>
By grouping multiple queries into one clickhouse-client invocation,
since each execve of the binary can take ~1 second in debug builds.
But this slightly changes the logic, so be aware.
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
Co-authored-by: Alexander Tokmakov <tavplubix@clickhouse.com>
In debug bulds each client invocation takes ~1 second, and on CI it can
take more if the node is under some load, so let's decrease number of
iterations.
Anyway CI runs each test ~1K times daily, and if there will be something
even this number of iterations should be enough.
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
Use one clickhouse-client invocation instead of 300, in debug builds it
is significant - each spawn is ~1 second
Signed-off-by: Azat Khuzhin <a.khuzhin@semrush.com>
