4.0.4

Release Date: January 16, 2026

Improvements

• Supports Parallel Prepare for Operators and Drivers, and single-node batch fragment deployment to improve query scheduling performance. #63956
• Optimized deltaRows calculation with lazy evaluation for large partition tables. #66381
• Optimized Flat JSON processing with sequential iteration and improved path derivation. #66941 #66850
• Supports releasing Spill Operator memory earlier to reduce memory usage in group execution. #66669
• Optimized the logic to reduce string comparison overhead. #66570
• Improved skew detection in GroupByCountDistinctDataSkewEliminateRule and SkewJoinOptimizeRule to support histogram and NULL-based strategies. #66640 #67100
• Enhanced Column ownership management in Chunk using Move semantics to reduce Copy-On-Write overhead. #66805
• For shared-data clusters, added FE TableSchemaService and updated MetaScanNode to support Fast Schema Evolution v2 schema retrieval. #66142 #66970
• Supports multi-warehouse Backend resource statistics and parallelism (DOP) calculation for better resource isolation. #66632
• Supports configuring Iceberg split size via StarRocks session variable connector_huge_file_size. #67044
• Supports label-formatted statistics in QueryDumpDeserializer. #66656
• Added an FE configuration lake_enable_fullvacuum (Default: false) to allow disabling Full Vacuum in shared-data clusters. #63859
• Upgraded lz4 dependency to v1.10.0. #67045
• Added fallback logic for sample-type cardinality estimation when row count is 0. #65599
• Validated Strict Weak Ordering property for lambda comparator in array_sort. #66951
• Optimized error messages when fetching external table metadata (Delta/Hive/Hudi/Iceberg) fails, showing root causes. #66916
• Supports dumping pipeline status on query timeout and cancelling with TIMEOUT state in FE. #66540
• Displays matched rule index in SQL blacklist error messages. #66618
• Added labels to column statistics in EXPLAIN output. #65899
• Filtered out "cancel fragment" logs for normal query completions (for example, LIMIT reached). #66506
• Reduced Backend heartbeat failure logs when the warehouse is suspended. #66733
• Supports IF EXISTS in the ALTER STORAGE VOLUME syntax. #66691

Bug Fix

The following issues have been fixed:

• Incorrect DISTINCT and GROUP BY results under Low Cardinality optimization due to missing withLocalShuffle. #66768
• Rewrite error for JSON v2 functions with Lambda expressions. #66550
• Incorrect application of Partition Join in Null-aware Left Anti Join within correlated subqueries. #67038
• Incorrect row count calculation in the Meta Scan rewrite rule. #66852
• Nullable property mismatched in Union Node when rewriting Meta Scan by statistics. #67051
• BE crash caused by optimization logic for Ranking window functions when PARTITION BY and ORDER BY are missing. #67094
• Potential wrong results in Group Execution Join with window functions. #66441
• Incorrect results from PartitionColumnMinMaxRewriteRule under specific filter conditions.