2025 Changelog

ClickHouse release v25.4.2.31-stable (e2ce00b6885) FIXME as compared to v25.4.1.2934-stable (589918f385f)

Performance Improvement

• Backported in #79662: Introduced threshold (regulated by setting parallel_hash_join_threshold) to fall back to the hash algorithm when the size of the right table is below the threshold. #76185 (Nikita Taranov).
• Backported in #79728: Fix performance of the distributed INSERT SELECT with parallel replicas. #79441 (Azat Khuzhin).

Improvement

• Backported in #79479: Fix possible endless loop when listing blobs from Azure/S3 blob storage. #79425 (Alexander Gololobov).

Bug Fix (user-visible misbehavior in an official stable release)

• Backported in #79502: ```sql CREATE TABLE t0 ( key Int32, value Int32 ) ENGINE=MergeTree() PRIMARY KEY key PARTITION BY key % 2;. #78593 (Vlad).
• Backported in #79435: Fixes a bug where arrayUnion() returned extra (incorrect) values on tables that had batch inserts. Fixes #79157. #79158 (Peter Nguyen).
• Backported in #79607: Fix loading of plain_rewritable disks containing data. #79439 (Julia Kartseva).
• Backported in #79525: Fixed error Block structure mismatch in several cases with tables with sparse columns. #79491 (Anton Popov).
• Backported in #79616: Fix result duplication when reading from remote with both use_hedged_requests and allow_experimental_parallel_reading_from_replicas disabled. #79599 (Eduard Karacharov).

NOT FOR CHANGELOG / INSIGNIFICANT

• Backported in #79694: Lazy Materialization with parallel replicas. #79401 (Igor Nikonov).
• Backported in #79580: Fix a typo in threadstack::getsize. #79533 (Sema Checherinda).
• Backported in #79595: fix data race in poco::JSON::stringify. #79574 (Han Fei).
• Backported in #79686: Make #76185 backportable. #79665 (Nikita Taranov).
• Backported in #79705: refine the exception message when keeper reaches soft memory limit. #79670 (Han Fei).
• Backport test_storage_rabbitmq/test_failed_connection from master into 25.4. #79713 (Pablo Marcos).