ApsaraDB for ClickHouse:Release details of ApsaraDB for ClickHouse Enterprise Edition V24.2.16476

New features

• Logs are added for splitBlockIntoParts.

• Additional profiling helpers are added for Keeper.

• Missing settings are added in schema inference cache.

Performance improvement

• The response interval for resource scaling of serverless clusters is shortened to prevent queries from being interrupted due to delayed scaling.

• The following parameters are added to ensure that the vertical merge algorithm can be used in some scenarios to improve the merge performance:

vertical_merge_algorithm_min_rows_to_activate=0
vertical_merge_algorithm_min_bytes_to_activate=134217728
vertical_merge_algorithm_min_columns_to_activate=0

• You can create only table engines that do not store data on local disks when you use the Replicated database engines. This avoids various issues caused by incorrect configurations and operations.

• The digest_enabled parameter of Keeper is set to 1 to enable real-time data consistency check.

• The async_replication parameter of Keeper is set to 1 to improve the performance of data replication across multiple nodes when read/write consistency is ensured.

• The queries per second (QPS) and bandwidth required to access Object Storage Service (OSS) are reduced for the same business load.

Fixed issues

• The check_not_exists and create_if_not_exists parameters of Keeper are set to 1 to avoid unexpected errors that occur when the ClickHouse server performs operations.

• The following issue is fixed: SQL queries terminate prematurely due to inaccurate memory statistics.

• The data race issue in S3::ClientCache is fixed. For more information, see Fix data race in S3::ClientCache.

• The following issue is fixed: In rare cases, ApsaraDB for ClickHouse considers parts as broken due to some unexpected projections on disks.

• The missing sparse removal is added in MergeJoin to fix a bug that occurs in Join operations. For more information, see Add missing sparse removal in MergeJoin.

• You can cancel queries on PostgreSQL tables. For more information, see PostgreSQL source support cancel query.

• The buffer overflow issue in unbin implementation is fixed.

• The issue with signed integers in SumIfToCountIfVisitor is fixed.

• The following issue is fixed: The not-ready set error occurs when a subquery with IN is used in the constraint. For more information, see Fixing Not-ready set in constraint.

• The incorrect optimization issue that occurs when the IS NULL condition exists is fixed. For more information, see Better handling of join conditions involving IS NULL checks.

• The subqueries with IN in DDLLoadingDependencyVisitor are ignored.

• The following issue is fixed: A session is closed after a hardware error from ZooKeeper is received. For more information, see Finalize session on hardware error from ZooKeeper.

• Long processing requests in Keeper are properly printed.

• The jemalloc assertion issue caused by non-monotonic CLOCK_MONOTONIC_COARSE is fixed. For more information, see Fix jemalloc assertion due to non-monotonic CLOCK_MONOTONIC_COARSE.

• The crash issue caused by maxIntersections is fixed.

• The chains of injective functions are removed recursively by using UniqInjectiveFunctionsEliminationPass. For more information, see UniqInjectiveFunctionsEliminationPass recursively removing chain of injective functions.

• An error is returned if a connection cannot be established with any of the replicas chosen for query execution.

• The jemalloc with a profiler is built.

• The following issue is fixed: When a session is closed by Keeper, subsequent connections may fail.

• Sufficient global thread pools are available when Keeper is started.

• The memory statistics error caused by the file system cache is fixed.