v.25.3
Backward Incompatible Change
2Bug Fix (user-visible misbehavior in an official stable release)
48- Fix stuck while processing pending batch
- Improved datetime conversion during index analysis
- Fix race between UUID and table names
- Fix data loss
- Renamed field 'marks' to 'marks\_bytes' in system.data_skipping_indices
- Fix dynamic filesystem cache resize handling unexpected
- Fix used_flag
- Fix a logical error
- Do not request interactive basic auth
- Fix THERE\_IS\_NO\_COLUMN exception
- Subpath inside the table directory is chosen
- Fix an error Not found column
- Add performance tests for null shortcircuit and
- Flush output write buffers before finalizing them
- Add support for MongoDB's binary UUID
- Fix exception handling
- Fix lgamma function behavior
- Fix reverse key analysis
- Fix pretty print of Bool values
- Fix possible crash because of bad JSON
- Previously, multiIf may return different types
- Fix incorrect serialization of constant nullable keys
- Fix sorting of BFloat16 values
- Fix JSON with Variant subcolumn by adding
- Fix crash
- Allow creating EmbeddedRocksDB table with subcolumn
- Fix illegal comparison
- Fix crash during Kafka table creation
- Support JSON and subcolumns
- Fix exceptions stack unwinding on MacOS
- Fix reading 'null' subcolumn
- Fix bloom filter
- We should only check the restriction
- Not a bug: SELECT toBFloat16(-0.0) == toBFloat16
- Fix posbile
- Fix name for partition with a Bool
- Fix comparison between tuples with nullable elements
- Fix crash
- Fix async_insert
- Remove by planer
- Fix several LOGICAL\_ERRORs around setting alias
- Fix error processing during file segment write
- Make DatabaseIceberg use correct metadata file provided
- Query cache now assumes
- Fix system
- Fix a logical error
- Second arguments of type Nullable
- Regularly check if merges and mutations were
Build/Testing/Packaging Improvement
2Experimental Feature
4New Feature
14- JSON data type is production-ready
- Introduce SSH protocol
- Replace table functions
- New implementation of the Userspace Page Cache
- Add concurrent_threads_scheduler server
- Add estimateCompressionRatio aggregate
- Add function arraySymmetricDifference
- Allow to explicitly specify metadata file
- Add keccak256 hash
- Add three new
- Support LowCardinality(Decimal) data types
- FilterTransformPassedRows and FilterTransformPassedBytes profile events will show
- Support the histogram metric type
- Support switching on explicit values
Improvement
21- Support atomic rename when TRUNCATE is used
- It's no longer possible
- Disable parallel replicas by default when analyzer
- Add the ability to define a list
- Respect column insensitive column matching
- Parameters for the codec Gorilla will now
- Implemented parsing enhancements for certain data lakes
- Remove trace\_id
- Encryption (the attribute encrypted_by) can now be
- Add some dynamic warning messages that can
- Add PROFILES a, DROP ALL PROFILES because
- Various enhancements for SYNC REPLICA
- Use correct fallback when multipart copy
- Remove after session_timeout_ms (default: 45 seconds)
- Fix validation of S3 request settings
- System tables like server_settings or settings have
- Add ProfileEvents::QueryPreempted, which has a similar logic
- Previously, a Replicated database could print credentials
- Allow ALTER TABLE DROP PARTITION
- Backup/restore setting allow_s3_native_copy now supports value three
- Support aws session token and environment credentials
Performance Improvement
13- Improve query performance using repeated conditions
- Actively evict data from the cache
- Replace Int256 and UInt256 with clang builtin
- (e.g. empty array column) data parts can
- Improve min/max performance
- Query compilation (setting compile_expressions) now considers
- Optimize arraySort
- Disable filesystem_cache_prefer_bigger_buffer_size when the cache is used
- Allow slightly better code generation
- Faster ClickHouse servers shutdown
- Avoid excess allocation
- Update zstd
- Reduce memory usage during prefetches of JSON