New Features

Query & Execution

• Support Python UDF/UDAF/UDTF (#63387)
• Support remote scan bytes breaker in workload policy (#64649)
• Add username-based backend workload policy support (#60559)
• Add is_valid_utf8 function (#62515)

Cloud Native

• Add table-level event-driven warm up (#63832)

Load

• Support zstd stream load compression (#64711)

Index & Search

• Add ANN TopN small-candidate fallback to brute-force search (#64555)

Improvement

Query & Execution

• Treat inactive_file as available memory to avoid false query cancellations in cgroup environments (#64347)
• Set default profile level to 2 to print more detail query profile by default (#64378)
• Add configurable return mode for insert publish timeout in ETL scenarios (#64583)

Load

• Make from-to streaming task timeout progress-aware (#64301)
• Add zero-row hint for Kafka read_committed load and delay zero-row retries (#64585)
• Refresh routine load lag metrics more timely (#63654)

Cloud Native

• Add rate limit for BE to Meta-Service RPC (#64396)
• Add virtual compute group switch metric (#63036)

Bugfix

Index & Search

• Handle truncated sparse path stats when reading Variant (#64205)
• Preserve JSON object when casting JSONB to Variant (#63792)
• Return raw string for element_at on scalar-string Variant (#64103)
• Escape JSONB path member control characters (#63517)

Query & Execution

• Fix SimplifyAggGroupBy should verify injectivity (#64335)
• Fix partition TopN optimization requiring window function partition key to be subset of co-located keys (#65073)
• Fix sorted merge continuing when sender queue is ready (#65004)
• Fix TopN merge with child prefix order keys (#64685)
• Fix nullable aggregate visitor dispatch (#64885)
• Fix incorrect aggregate merge with duplicate aliases (#65025)
• Fix timeout checker stopping after an exception (#65040)
• Restrict auto salt join rewrite to avoid incorrect results (#64518)
• Fix handling of generated columns in delete partial update (#64884)
• Guard Count(*) child access with arity check in PushDownAggThroughJoinOnPkFk (#64848)
• Fix cloud query retry setting version_cache_ttl temporarily to 0 on error -230 (#63721)
• Fix recursive CTE blocks not sent to every scan instance (#64964)
• Fix SHOW PROCESSLIST returning unexpected result in FULL mode (#64631)
• Fix isNull parser placement under primaryExpression (#63619)
• Fix subquery eliminating null-aware anti-join incorrectly (#64639)
• Skip TopN lazy materialization when light_schema_change=false (#64441)
• Fix copy into select not binding file column placeholders (#64591)
• Clamp merged limit of MERGE_TOP_N by parent offset (#64306)
• Prevent cast project pushdown through UNION DISTINCT (#64080)
• Disable row-store lazy fetch for shared base columns (#62864)
• Fix self-deadlock in FragmentMgr on query context map clear (#64552)
• Fix month nullable datetimev2 literal binding (#64459)
• Remove incorrect topN-to-MAX optimizer rewrite (#63519)
• Fix local runtime filter merge deadlock (#65102)
• Return error status from materialization operator row-ID fetcher (#62513)
• Normalize aggregate order-by pushdown (#64787)
• Fix NereidsCoordinator not created in proxy flow due to missing parsedStatement (#64363)
• Reject more than 1 argument in COUNT DISTINCT window function (#64783)
• Avoid incorrect convert_tz partition pruning across DST transitions (#63853)
• Fix MAKE_SET constant folding not clearing prior result (#64907)
• Prevent SIGFPE in int_divide for signed minimum value (#64828)
• Fix nth_value for upper-bounded windows (#64864)
• Return NaN for avg_weighted when sum of weights is zero (#64333)
• Stabilize conjunct cost ordering (#64637)
• Stop extra operator work after cancellation (#64077)
• Fix array_first/array_last boolean cast (#64847)
• Validate array_sort lambda arity (#64825)
• Limit retention param count to 32 to avoid BE heap overflow (#64521)
• Fix concat_ws nullable array handling (#64703)
• Fix split_by_delimiter missing backslash escape handling (#61995)
• Normalize v1 date string cast result (#64575)
• Fix SIGSEGV in bvar::take_sample under high EPS due to AgentCombiner/TLS race (#64040)
• Validate sequence pattern event numbers (#64930)
• Catch standard exceptions in pipeline scheduler (#65019)
• Catch block serialization exceptions to avoid coredump (#64852)
• Validate task executor scan handles (#65054)
• Initialize thread context on AsyncIO worker threads (#64846)
• Fix sliced FixedSizeBinary Arrow string reads (#64829)
• Avoid mutable nullable CRC32C hashing (#64944)
• Preserve DateTimeV2 scale in OUTFILE CSV export (#64344)
• Reject invalid IPv4 default value at CREATE TABLE time (#62906)
• Harden Arrow Flight split source error path to avoid BE crash on external scan (#64797)

Storage & Compaction

• Fix group commit block queue unavailability (#63722)
• Fix lost rows when prepared statement reuses plan with same load_id in group commit (#64362)
• Fix shared delta writer state lifetime (#64504)
• Fix tablet writer map lookup race in load channel (#64604)
• Prevent EMPTY_CUMULATIVE / BASE-CUMU races on same tablet in cloud compaction (#64619)
• Persist update time for sub-transaction commit (#64739)
• Make S3 rate limiter config take effect dynamically in cloud mode (#64554)
• Use snapshot read for table version to avoid transaction conflict (#64647)
• Recycle empty rowsets without resource ID (#64630)
• Fix double assignment in recycler_service that caused recycler malfunction (#64168)
• Avoid false-positive leaked delete bitmaps for unexpired job tmp rowsets (#64313)
• Refresh base tablet before schema change V1 (#64312)
• Forbid restoring table with light_schema_change=false in cloud mode (#62914)
• Rewrite table properties and partition info during cloud restore (#64466)
• Make sync file cache clear use safe removal (#64578)
• Disable sync file cache clear in HTTP API (#64321)
• Exclude warmup reads from file cache hit ratio metrics (#63394)
• Force drop partition in INSERT OVERWRITE (#62510)
• Add null-safety to getBaseViewsOneLevel for MTMV backward compatibility (#64412)
• Fix MTMV mutating excluded trigger tables (#62984)
• Avoid finalized pipeline task submit crash (#64953)

Load

• Serialize audit loader batch assembly to fix race (#65107)
• Lock routine load task renew on submit failure (#65007)
• Fix stream load IPv6 host parsing (#64147)
• Replace tablet writer close polling with event wakeup (#64221)
• Fix empty statistics for forwarded INSERT (#64439)
• Fix load_to_single_tablet routing for auto partition (#64356)
• Record per-query SET_VAR hint session variables in audit log (#64569)
• Fix AVRO JNI reader coredump with null pointer during TVF function (#64699)

Lakehouse

• Fix NPE in Iceberg COUNT() pushdown when snapshot summary omits total- counters (#64648)
• Fix Iceberg binary columns written with wrong Arrow types (#64949)
• Add LZ4 compression support for Iceberg/Hive Parquet/ORC writers (#64723)
• Allow disabling REST catalog view operations (#63320)
• Preserve empty text records when reading Hive text files (#64671)
• Estimate MaxCompute write block size from Arrow buffers instead of per-row serialization (#64612)
• Push SQL Server/Oracle boolean predicates as 1/0 instead of TRUE/FALSE in JDBC (#64757)
• Fetch table comment PROP from Iceberg table metadata (#64263)
• Avoid blocking external meta cache refresh on slow miss load (#64705)

Security & Authentication

• Add auth check for manager node and query error REST APIs (#65080)