This should fix the following SIGSEGV, that was found on CI [1]:
<Fatal> BaseDaemon: Address: NULL pointer. Access: read. Unknown si_code.
<Fatal> BaseDaemon: 4.4. inlined from ../contrib/rocksdb/utilities/object_registry.cc:19: rocksdb::ObjectLibrary::FindEntry() const
...
<Fatal> BaseDaemon: 7.3. inlined from ../contrib/rocksdb/options/cf_options.cc:678: rocksdb::$_7::operator()()
[1]: https://clickhouse-test-reports.s3.yandex.net/29341/2b2bec3679df7965af908ce3f1e8e17e39bd12fe/integration_tests_flaky_check_(asan).html#fail1
And also I checked manually with TSan binary, and here is a data race
reported by TSan:
WARNING: ThreadSanitizer: data race (pid=3356)
Read of size 8 at 0x7b0c0008cca8 by thread T40:
2 rocksdb::ObjectLibrary::FindEntry() const obj-x86_64-linux-gnu/../contrib/rocksdb/utilities/object_registry.cc:18:27 (clickhouse-tsan+0x1b839a6c)
...
6 rocksdb::$_7::operator()() const obj-x86_64-linux-gnu/../contrib/rocksdb/options/cf_options.cc:676:32 (clickhouse-tsan+0x1b6bfa63)
...
28 rocksdb::GetColumnFamilyOptionsFromMap() obj-x86_64-linux-gnu/../contrib/rocksdb/options/options_helper.cc:727:10 (clickhouse-tsan+0x1b6fffd2)
29 DB::StorageEmbeddedRocksDB::initDb() obj-x86_64-linux-gnu/../src/Storages/RocksDB/StorageEmbeddedRocksDB.cpp:359:26 (clickhouse-tsan+0x14195e31)
...
Previous write of size 8 at 0x7b0c0008cca8 by thread T41:
...
9 rocksdb::ObjectLibrary::AddEntry() obj-x86_64-linux-gnu/../contrib/rocksdb/utilities/object_registry.cc:31:19 (clickhouse-tsan+0x1b8392fc)
...
11 rocksdb::RegisterTableFactories()::$_0::operator()() const obj-x86_64-linux-gnu/../contrib/rocksdb/table/table_factory.cc:23:14 (clickhouse-tsan+0x1b7ea94c)
...
43 rocksdb::GetColumnFamilyOptionsFromMap() obj-x86_64-linux-gnu/../contrib/rocksdb/options/options_helper.cc:727:10 (clickhouse-tsan+0x1b6fffd2)
44 DB::StorageEmbeddedRocksDB::initDb() obj-x86_64-linux-gnu/../src/Storages/RocksDB/StorageEmbeddedRocksDB.cpp:359:26 (clickhouse-tsan+0x14195e31)
Refs: https://github.com/ClickHouse-Extras/rocksdb/pull/13Fixes: #29341
If ReplicatedAccessStorage startup was not executing or if it
failed before completing (for instance when ZooKeeper was not configured),
its destructor would call shutdown and try to join a missing thread.
