StarRocks version 3.5

warning

Upgrade Notes

JDK 17 or later is required from StarRocks v3.5.0 onwards.
- To upgrade a cluster from v3.4 or earlier, you must upgrade the version of JDK that StarRocks depends, and remove the options that are incompatible with JDK 17 in the configuration item JAVA_OPTS in the FE configuration file fe.conf, for example, options that involve CMS and GC. The default value of JAVA_OPTS in the v3.5 configuration file is recommended.
- For clusters using external catalogs, you need to add --add-opens=java.base/java.util=ALL-UNNAMED to the JAVA_OPTS configuration item in the BE configuration file be.conf.
- For clusters using Java UDFs, you need to add --add-opens=java.base/java.nio=ALL-UNNAMED --add-opens=java.base/sun.nio.ch=ALL-UNNAMED to the JAVA_OPTS configuration item in the BE configuration file be.conf.
- In addition, as of v3.5.0, StarRocks no longer provides JVM configurations for specific JDK versions. All versions of JDK use JAVA_OPTS.
It is recommended to upgrade the cluster to v3.4.10 or later before upgrading it to v3.5. Otherwise, you must manually disable low cardinality optimization during the gray-scale upgrade by executing the following statement:
```
SET GLOBAL cbo_enable_low_cardinality_optimize=false;
```

Downgrade Notes

After upgrading StarRocks to v3.5, DO NOT downgrade it directly to v3.4.0 ～ v3.4.5, otherwise it will cause metadata incompatibility. You must downgrade the cluster to v3.4.6 or later to prevent the issue.
After upgrading StarRocks to v3.5.2 or later, DO NOT downgrade it to v3.5.0 & v3.5.1, otherwise it will cause FE crash.

3.5.21

Release date: August 28, 2026

Behavior Changes

Reverted the 3.5.20 change that cached Iceberg REST vended-credential tables and kept their credentials fresh: on branch-3.5, that caching caused INSERT OVERWRITE transactions against a slow Iceberg REST catalog to stay COMMITTED but not VISIBLE for several seconds, because a REST metadata refresh could now run while the planner held a lock that transaction publishing also needs. The caching behavior remains in place on the 4.x line. #77039
GIN (inverted) indexes on Primary Key tables now read from the correct segment after a column-mode partial update, instead of serving stale index data from the unmodified base segment. #76271
Audit logs for statements forwarded to the leader FE now record the leader-resolved, fully qualified table relations (with CTEs excluded), instead of the follower's unresolved names. #76387
ARRAY/MAP constructor expressions now reject a result whose flattened size exceeds 4 GB instead of silently wrapping around and returning corrupted values. #76419
Iceberg REST catalogs using OAuth2 client-credentials now self-heal their background token-refresh session after a prolonged failure, instead of leaving the catalog permanently unable to refresh its access token. #76457
External scan contexts (for example, an abandoned Spark/Flink connector read) now properly cancel their pipeline fragments when reaped as expired, instead of leaving them running. #76535
External scan plans (Spark/Flink connector reads) now set query_delivery_timeout, so their QueryContext no longer waits indefinitely for fragments that will never arrive. #76536
array_difference() on integer arrays no longer overflows in 32-bit precision before widening to BIGINT, fixing incorrect results when the true difference falls outside the INT range. #76569
Division expressions with a non-constant divisor (for example, 10 DIV c) are no longer treated as monotonic, fixing incorrect ZoneMap-based pruning that could return an empty or wrong result. #76744
The configured compression codec now applies to the synthetic null/offset sub-columns of flat JSON, ARRAY, MAP, and STRUCT columns, instead of always writing them as uncompressed raw pages. #76949
Load quorum selection no longer picks a replica in DECOMMISSION state as the load primary. #77035
Optimizer rules that rebuild a logical window operator now preserve its inputIsBinary flag, keeping the binary-input merge optimization for ranking-window pre-aggregation intact. #77058

Improvements

Materialized views no longer force a full-partition refresh when manually set from INACTIVE to ACTIVE; only the metadata version map is cleared during a schema change. #57371
Improved error messages for large-column-capacity-limit checks by removing internal diagnostics (raw pointers, operator dumps) from the user-facing error, and fixed a typo in the shared status string. #76303

Bug fixes

The following issues have been fixed:

Guarded against cyclic view definitions: an ALTER VIEW that closes a reference cycle (v1 -> v2 -> v1) is now rejected with a clear error instead of causing a later SELECT to recurse forever and crash with StackOverflowError. #75033
Forbade pushing an aggregate down through a CASE expression that has a non-null constant ELSE branch, fixing a planning abort (IllegalStateException) that could occur once the rule fired. #75037
CTEAnchor now prunes correctly when its child is a ValueOperator. #64491
Fixed predicate conversion for Paimon: an AND-combined predicate now retains its convertible not-null branch instead of the whole conversion returning null. #66038
The Arrow Flight prepared-statement schema no longer reports every view column as nullable regardless of its actual NOT NULL definition, and a related regression that produced wrong nullability for GROUP BY ROLLUP/CUBE/GROUPING SETS key columns has also been fixed. #75684 #76149
Fixed two statistics bugs: an OR-predicate statistics estimate that always clamped the merged nullsFraction to 1 regardless of the real value, and a column-statistics cache load failure under ERROR_IF_OVERFLOW when a column's persisted min/max is an empty string. #75864 #76684
Fixed a bRPC stub cache timer leak that leaked memory over time. #75973
UNNEST output struct pruning now uses its input array's subfield group instead of the output's own, fixing a mismatch between the struct type BE materializes and the one FE declares. #76002
Fixed Arrow Flight Prepared Statement forwarding sending the wrong action-type string when a request is forwarded to a different FE, which caused every ADBC client using prepared statements behind a load balancer to fail. #76310
Hive getTable() now reconnects before falling back to get_table_req(), fixing an intermittent out of sequence response / Unknown table error when querying an Iceberg table through a Hive metastore catalog. #76456
Made catalog-drop existence checks atomic under the write lock, fixing a check-then-act race that could persist a redundant drop record when two drops of the same catalog ran concurrently. #76778
Fixed PipeObservable emitting a source event instead of a sink event on a deferred sink notification, which could leave a driver blocked on OUTPUT_FULL unresponsive. #76782
dictionary_get() no longer rejects a non-NULL key when its input column's cached has_null flag is stale. #76881
Added the missing arrow-compression module for Arrow Flight SQL, restoring LZ4/ZSTD codec support for compressed Arrow IPC clients. #76921
Fixed an OOM in streaming pre-aggregation under memory pressure with spill enabled. #76702
Set operators are no longer placed in colocate execution groups, fixing a hang caused by their branches being terminated by a plain local-exchange sink instead of a grouped-execution sink. #77025
Stopped the query cache from storing incomplete per-tablet results for an aggregation with a LIMIT, which could return wrong results from the cache. #77066
An insert-overwrite failure is no longer journaled against a table that was concurrently dropped, fixing an FE crash on journal replay. #77212
Checkpoint-thread-created thread pools are no longer registered in the metrics registry. #77367
Stopped shipping test-scope jars in the java-extensions reader libraries. #77752
Resolved tablet backend IDs outside the TabletInvertedIndex write lock, fixing an FE deadlock between materialized-view refresh/insert-overwrite commit and tablet force-delete. #78102
Several BE/CN crashes: a crash during graceful shutdown when a pending brpc closure ran after SinkBuffer was destroyed; a null-pointer crash in SparseRangeIterator::has_more() on an empty physical-split tablet; a bad_weak_ptr abort from an unscheduled global runtime-filter timer in PipelineDriver's destructor; a SIGSEGV in the native Parquet reader on an incomplete nested lake schema; a heap-use-after-free with a multi-character CSV delimiter that straddles a buffer expansion; and a heap-buffer-overflow building the error message for a raw JSON value in SimdJsonConverter. #73202 #75985 #76252 #76455 #76718 #76752
Several dependency CVEs: excluded an unused avro-ipc dependency that bundled a vulnerable jQuery 1.4.2; bumped Netty to 4.1.136.Final; excluded vulnerable Jetty jars and bumped pgjdbc to 42.7.12; bumped Apache Thrift to 0.24.0; and bumped Apache HttpCore to 5.4.3. #76270 #76555 #76783 #76922 #77753

3.5.20

Release date: July 23, 2026

Behavior Changes

CREATE DATABASE IF NOT EXISTS on Iceberg REST catalogs now succeeds silently when the database already exists, instead of raising an error. #75017
Iceberg REST catalogs with vended credentials now cache Table objects and keep their credentials refreshed on access, instead of bypassing the cache and re-fetching from the REST catalog/Lake Formation on every getTable() call, which could trigger AWS Rate exceeded errors. #75431
GIN inverted-index-accelerated NOT MATCH predicates no longer return rows with a NULL value, matching SQL three-valued-logic semantics. #75578

Improvements

Added the FE metric txn_max_committed_pending_publish_ms, a per-database gauge reporting the longest time a committed transaction has been pending publish, to help diagnose stuck or lagging version publishing. #75025
Enforced the query memory limit when a column is upgraded (widened) during window-function aggregation in Analytor, instead of letting it grow unbounded. #75821
Removed useless per-rowid seeks in the array-column offsets-only read path used by array_length()/cardinality(). #75861

Bug fixes

The following issues have been fixed:

Several wrong-result issues: EliminateSortColumnWithEqualityPredicateRule dropping the global LIMIT under concurrency; SplitJoinORToUnionRule producing duplicate rows for a null-safe-equal (<=>) JOIN ON p1 OR p2; JIT codegen truncating LARGEINT literals >= 2^64 to 64 bits; array_map/transform silently dropping NULL rows when all non-null input arrays were empty; nested dictionary expressions rebuilt inconsistently across exchange fragments causing dict-decode failures; and a LIKE pattern with the _ wildcard returning wrong rows on a GIN inverted index. #74983 #75038 #75137 #75141 #75246 #75551
Join-reorder column pruning could drop a column still referenced by a predicate, causing a missing statistic of col planning error, and JoinTuningGuide could lose predicateCommonOperators when rebuilding a join, failing plan validation. #74791 #75773
Sync materialized view/rollup rewrite could lose a rollup column when a query aggregated the same base column twice (e.g. min(c) and max(c)), and async materialized view rewrite could serve stale results after an Iceberg base table's rollback_to_snapshot. #75528 #75924
PARTITION-TOP-N could rewrite its partition-by column to a dictionary slot that no longer existed, failing with a slot_id not found error. #75956
An NPE collecting view tables when a SECURITY INVOKER view's stored definition contains a CTE. #74813
Three FE metadata-lock correctness races around DROP PERSISTENT INDEX, RestoreJob post-restore handling, and related unlocked paths. #74968
A race between FE EOS-cancel and BE stage-2 deploy could mark a fully successful query as canceled. #75009
ApplyTuningGuideRule could throw UnsupportedOperationException when an earlier rewrite produced an OptExpression with an immutable input list. #70785
BE/CN crashes: a null driver_executor when a cancel RPC arrives before pipeline start; a use-after-free in the spill partition-sort-sink cancel path; a heap-use-after-free in OrderedPartitionExchanger for a skew-hinted window function at DOP>1; an NLJoin crash from a build-side column nullability mismatch; a StructColumn field-count mismatch in UNNEST output; a crash loop reading a flat-JSON column that changed from NOT NULL to nullable during compaction; an uncaught memory-allocation exception in NLJoinProbeOperator; a crash in primary-key auto-increment partial-update apply; and a crash rewriting predicates inside an array_map lambda during scan-predicate pushdown. #75030 #75140 #75279 #75343 #75445 #75680 #75788 #76119 #76380
histogram() crashed (or silently mis-bucketed values) on a non-positive bucket_num instead of raising a clear error, and bar() could grow an unbounded string for a negative or huge width argument, exhausting BE memory. #75041 #75143
A query using unnest over array columns could exceed query_mem_limit and get the BE OOM-killed instead of failing just that query. #75179
A second-order SQL injection in the information_schema.task_runs TASK_NAME/QUERY_ID predicate lookup. #75520
SHOW CREATE ROUTINE LOAD could emit a spurious leading comma before the first load-desc clause, and an unescaped jsonpaths value, producing non-runnable DDL. #75522 #75755
Shared-data (lake) SHOW PARTITIONS and information_schema.partitions_meta reported every physical partition's bucket count as the table-level default instead of its own bucket count. #75734
Several dependency CVEs by upgrading jackson-databind and Netty. #75373 #76555
Batched TabletInvertedIndex write-lock acquisition in markTabletsForceDelete, reducing lock churn when force-deleting many tablets at once. #75616
Batched tablet inverted-index writes in the insert-overwrite path. #75923
Skipped an unnecessary remote clear_parent_path call when a load spill never used remote storage. #76224
A null-padding size mismatch for missing columns in ParquetScanner so padded rows match the actual per-batch chunk size instead of the whole Parquet/Arrow batch size. #75981
Vulnerable, stale transitive dependencies (old BouncyCastle, OkHttp 2.x, Tomcat, and others) that previously shipped alongside their fixed counterparts. #76097

3.5.19

Release date: June 26, 2026

Behavior Changes

parse_json now respects ALLOW_THROW_EXCEPTION: malformed JSON fails the query instead of silently producing NULL, mirroring the earlier get_json_string change. #74976
FILES() and Broker Load now honor the Parquet isAdjustedToUTC=false flag for INT64 timestamps, so such timestamps are no longer shifted by the session time zone. #73674
SHOW FUNCTIONS now surfaces the isolation property (shared or isolated) for Java UDFs and UDAFs. #75255
The non-reserved keywords FLOOR and CEIL are now allowed as column names. #75241

Improvements

Supports loading Arrow LARGE_LIST and FIXED_SIZE_LIST columns into JSON columns via FILES() and Broker Load. #73714 #73718
Added an opt-in statistics_large_string_column_merge_threshold to isolate wide CHAR/VARCHAR columns into dedicated statistics collection. #73258
Optimized base64_to_bitmap for constant inputs and hardened it against invalid base64-encoded bitmaps. #74684
Added new metrics for lake vacuum batch size and retry counts, and gauges for CatalogRecycleBin size. #74112 #74440
Supports auditing a statement twice. #73896

Bug fixes

The following issues have been fixed:

Several wrong-result and planning issues: low-cardinality dictionary translation for expressions where f(NULL) is not NULL; a Multiple entries with same key error from common-subexpression elimination of commutative AND/OR; an AGGREGATE has mismatch types error; a compound predicate with an always-false nested branch under UNION returning no rows; and an off-by-one that dropped a row in RANK TopN at a chunk boundary. #69376 #72823 #74159 #74218 #75045
Materialized view rewrite issues that could return incorrect results: aggregate MVs with a HAVING clause rewriting queries without (or with weaker) HAVING, and avg(DISTINCT x) being rewritten through a sum/count MV. #73610 #75071
Incorrect window-function results when enable_push_down_pre_agg_with_rank split a window count into local pre-aggregation and global analytic merge, and an empty window operator generated after pushing down distinct aggregation. #74453 #74810
Partition TopN losing a child operator's output column, and silently swallowing sort or pre-aggregation errors and returning wrong or partial results. #72848 #74693
Iceberg equality-delete rows with NULL identity-column values were not applied. #67321
A spurious strict-mode cast overflow error raised from undefined data in NULL rows. #74903
Decimal scale could be lost when a column is entirely NULL. #73789
BE crashes in to_base64 (stack overflow), JSON load of nested types via partial append, local partition TopN with a non-nullable aggregate result, partitioned join (out-of-bounds from inaccurate memory accounting), runtime profile serialization (counter min/max race), JIT compilation failure (use-after-free of LLVMContext), invalid JIT IR for CASE WHEN with mixed float/integer types, and partial column updates under schema drift. #70623 #73715 #69752 #74315 #72904 #74396 #74382 #74005
An out-of-bounds read and potential oversized allocation in split, split_part, and str_to_map when the input ends with a truncated UTF-8 byte. #75068
A memory leak from the UDAF context cache and inflated query-pool memory accounting in OlapTableSink. #74025 #73807
Unexpected backend process restarts. #74424
Materialized view issues: a slot-nullability crash for MVs defined with FULL OUTER JOIN under late materialization, an NPE refreshing nested MVs, a duplicated warehouse property in SHOW CREATE MATERIALIZED VIEW, and a vector ANN query polluting a shared table schema and breaking unrelated statements. #72621 #73644 #69418 #74785
Querying Paimon tables whose DATE partition column contains NULL values. #73950
Reading Hudi MOR tables with char/varchar columns when hudi_mor_force_jni_reader is enabled. #58521
Nested INT96 timestamps (inside ARRAY, MAP, or STRUCT) were shifted by the session time zone during FILES()/Broker Load. #74868
Incorrect bytes-read statistics in the audit log for connector scans, and incremental connector scan ranges being assigned to driver sequences absent from the deployed fragment, which could drop part of the scan. #73799 #74674
Meta scan could fail after schema changes such as ADD COLUMN, which could fail background statistics collection. #72901
Slow broker RPCs held the per-job Routine Load write lock and blocked admin RPCs and SHOW ROUTINE LOAD. #73591
ALTER ROUTINE LOAD persisted an invalid statement for reserved-keyword table names, which could drop the load clause on FE restart. #74188
GRANT/REVOKE on the public role did not invalidate cached merged privileges, leaving stale authorization. #73717
A race allowing concurrent operations to observe torn state during table and materialized-view RENAME and SWAP, and a data race on MaterializedIndexMeta schema-update tracking. #74100 #74412
Database-level UDFs were missing on FE followers after RESTORE ... AS <new_db>. #74313
Queries could become unkillable when a coordinator held its lock during external resource cleanup. #72830
A permanent version hole on non-primary-key replicas could cause queries to fail with version not found. #74408
Force-killed SUBMIT TASK runs disappeared from task-run history (and session-prefixed task-run timeouts are now honored), and an illegal running-to-running edit log could wedge subsequent task runs. #74146 #73882
ADMIN SHOW REPLICA STATUS emitted a misaligned row for missing replicas, which could hang or disconnect the client. #74393
CatalogRecycleBin halted all deletions in shared-data mode when cluster snapshots kept failing, causing unbounded FE memory growth. #74379
An NPE in statistics calculation when a partition is dropped concurrently, and zero row counts written into partition statistics after INSERT OVERWRITE corrupting cardinality estimates. #73711 #74801
Colocate tablets with all replicas on dead BEs were reported as healthy when tablet_sched_disable_colocate_balance is enabled. #73550
An IllegalMonitorStateException from a lock mismatch in the tablet checker could abort a checker round. #74596
Reduced lock contention by narrowing several full-database WRITE locks to table-scoped locks in shared-nothing mode, and skipped unnecessary locking in TabletInvertedIndex.deleteTablets for empty input. #74523 #73955
A race between transaction begin and autovacuum could delete a still-needed transaction log and permanently wedge publishing in shared-data mode; decorrelated jitter was also added to lake vacuum retry backoff. #74906 #74108
Added lake_vacuum_enable_task_timeout to abort BE vacuum tasks once the FE caller's timeout elapses. #74694
A crash from a missing null check when reading gtid during a data-rewriting schema change. #74855
A thread-name race produced noisy warnings during BE data directory load. #73862
An IllegalStateException during parallel profile collection for external-table queries when enable_profile is on. #74746
ALTER TABLE ... MODIFY COLUMN ... AFTER a nonexistent column raised an internal NPE instead of a clean error. #75073
Query hangs and operator stalls from missing notifications when a distinct aggregate source finishes and on missed operator state transitions, and sort merge provider errors now propagate to the fragment context. #74055 #74557 #73337
FE dropped the connection (client ERROR 2013) instead of returning a proper error when connecting to a database the user lacks privileges on. #70072
Prepared statements with a microsecond DATETIME parameter failed with Invalid date type: DECIMAL(6,0). #74141
The audit log recorded ReturnRows=0 for SELECT ... INTO OUTFILE. #74467
DATETIME_PRECISION was always NULL in information_schema.COLUMNS, which broke type mapping for some MySQL-protocol clients. #74623
CTAS did not preserve an explicitly declared VARCHAR(N) length in the new table's schema. #73498
enable_statistic_collect_on_first_load now allows a table-level setting to override the global configuration. #74794
A typo in the azure_adls2_oauth2_client_endpoint configuration field name. #74581
Assertion name lookup in assert-num-rows. #74178
Several dependency CVEs by upgrading libthrift, Tomcat, and Netty, and excluding a vulnerable jline transitive dependency. #73243 #73797 #74668 #75066

3.5.18

Release date: June 5, 2026

Behavior Changes

SHOW statements are now allowed inside explicit transactions. #72954
get_json_string now respects ALLOW_THROW_EXCEPTION when handling JSON parsing errors. #73199
IGNORE NULLS is now preserved in view definitions when the window function argument is an expression. #69971
Ranger row filter and masking policies are now correctly applied to Hive views and to base tables expanded from Hive view definitions. #73265
Hive partition statistics are no longer automatically refreshed per partition. Existing cached stats are preserved while a table-level asynchronous refresh updates the cache in batches. #73563

Improvements

Supports caching Java UDAF class-level initialization so shared UDAFs can reuse loaded classes and generated stubs across aggregator and window-function instances. #72038
Supports Paimon time types and improves Paimon materialized view handling. #58292
Added an Avro schema cache for shadowed PartitionData during partition load. #72215
Added a configurable FE write timeout mysql_send_packet_timeout_ms for the MySQL result send path to prevent indefinitely blocked result sending to slow clients. #73646
Optimized CatalogRecycleBin adjusted recycle timestamp lookup. #72128
Reduced metadata and lock overhead in load balancing, compaction scheduling, consistency checks, and StarMgr metadata synchronization paths. #73555 #72218 #72178 #72108
Improved diagnostics for filesystem copy failures and Parquet broker load errors by surfacing the underlying cause and file/column/row context. #73414 #73236
Reduced external catalog and information schema metadata overhead by deferring JDBC REMARKS fetching, avoiding redundant Paimon snapshot lookups, and pushing down table_name predicates for information_schema.tables_config. #73488 #72892 #73210
Simplified the scalar-function merge implementation by using merge() directly. #69575

Bug fixes

The following issues have been fixed:

Empty ALTER TABLE statements could be parsed as OPTIMIZE clauses, and replaying malformed OPTIMIZE jobs could clear a table's default distribution. #73352
Decimal-valued unit counters in runtime profiles could cause query progress parsing failures and noisy FE warnings. #73683
Concurrent SegmentFlushTask race in DeltaWriter::commit() and loss of merge_condition during normal rowset commit. #73371 #72542
Crashes, hangs, or unsafe cleanup in SinkBuffer graceful exit, PipelineTimerTask, runtime filter workers, spillable hash join probe, information_schema.warehouse_queries, lake vacuum, HTTP connection unregister paths, and query queue timeout handling. #73202 #73082 #72058 #72626 #72397 #72019 #73088 #72006 #65802
Materialized view issues involving JDBC SQL Server tables, lost index properties, cached plan context memory leaks, Paimon tables, and incorrect shuffle distribution after MV rewrite. #72962 #69187 #72300 #58292 #71075
Query planning and rewrite issues in Spark connector external scans, INSERT OVERWRITE re-planning, aggregation spill with small LIMIT, and generated columns produced by UNNEST. #73225 #72832 #72705 #72027
Paimon Primary Key columns could be incorrectly marked as non-nullable when querying external catalogs. #71660
Primary Key and tablet metadata issues including partial tablet schema short-key mismatch, rowset metadata cache warmup deadlock, disk data cache expansion failure, Azure filesystem client cache issues in Starlet, and colocate-heavy cluster-balance performance issues in StarOS. #70586 #71459 #58206 #73145 #72391
Locker rollback and unlock-order issues during partial intensive-lock acquisition. #72789 #72423
Dependency CVEs and broker dependency regressions. #72905 #72797 #72184 #72191
JNI local-reference leaks in JDBC scanner initialization. #72913
Arrow dictionary values in Parquet scanner and Apache Parquet namespace ambiguity during scanner builds. #71855 #72284
NPE in Iceberg getPartitionLastUpdatedTime when the snapshot is expired. #68925

3.5.17

Release date: May 13, 2026

Behavior Changes

SHOW CREATE TABLE and DESC now show Primary Keys for Paimon tables. #70535
Disallowed INSERT into insert-only ACID Hive tables in Hive catalogs. #71460
START_TIME and END_TIME in Profile are now displayed using the session time zone. #71429

Improvements

Supports csv.enclose and csv.escape in INSERT INTO FILES CSV export. #71589
Added query relation information (directly queried tables and viewa) to audit logs. #71596
Made the FE configuration star_mgr_meta_sync_interval_sec runtime mutable. #71675
Reduced metadata and lock overhead in table metadata and row-count paths. #72053 #72042 #71672
Improved build and dependency hygiene by merging the broker builder into the FE build and removing WildFly OpenSSL. #71823 #71908

Bug fixes

The following issues have been fixed:

Wrong results for local-shuffle aggregate queries with OFFSET. #71997
Incorrect Join output properties after Exchange shuffle columns are pruned. #72003
Several dependency CVE issues. #71762 #71914
Oracle JDBC NLS format handling issue. #71412
Missing Iceberg column statistics in manifest data file cache. #71913
Missing Hive partition directory before INSERT OVERWRITE commit. #71810
Aggregate-join-pushdown materialized view rewrite and min/max optimization issues on Iceberg base tables. #71856 #71863
Race conditions in ConnectorSinkPassthroughExchanger and LoadChannel::get_load_replica_status. #71848 #71843
Credential redaction issue in INSERT FILES operations. #71245
Incorrect reverse(DecimalV3) results. #71834
Missing JNI exception handling checks in Java UDF code. #71734
Incorrect short-circuit checks in EventScheduler. #71740
Incorrect Arrow Flight column name for empty result sets. #71534
Batch publish deadlock caused by partition version gaps. #71483
Repeated Apply attachments in scalar-subquery plans. #71155

3.5.16

Release date: April 20, 2026

Improvements

Added clearer warning logs for swallowed exceptions in WarehouseManager. #71215
Supports pausing Routine Load jobs on non-retryable errors. #71161
Added thread names to the utility that prints all thread stacks. #69366
Supports constant folding for regexp_replace in FE. #70804
Supports showing column comments for PostgreSQL external tables, and added an option to let information_schema.tables fetch full metadata such as comments from external catalogs. #70520 #70197
Added automatic query plan dumping on query exceptions. #70387
Improved cloud-native tablet metadata fetch and repair efficiency. #70492 #70386
Added batch tablet deletion in FE to reduce write lock contention. #70052
Added logs for update compaction suspension, and added Iceberg metadata-table and time-travel query metrics. #70538 #70825 #70788

Bug fixes

The following issues have been fixed:

be_tablets.DATA_SIZE reports rowset column data bytes inaccurately. #70735
An outdated Maven repository for broker builds. #71533
Incorrect thread-pool resizing when updating parallel_clone_task_per_path. #71484
Several use-after-free issues. #71513 #71276 #71083 #62917 #69926 #69968
Resource group user classifier validation is not aligned with CREATE USER. #71470
“no queryable replica” issues on follower FEs by syncing StarMgr journal replay. #71263
Multiple dependency CVEs. #71256 #71017 #70862
VARCHAR length is not preserved after reduce-cast with global variables. #70269
Special-character escaping in equality predicates on information_schema.tables. #71273
UpdateTabletSchemaTask signature collisions across alter jobs. #71242
Issue with shared-object mutation in PushDownAggregateRewriter for CASE WHEN and IF expressions. #71309
Stopped inactive materialized view schedulers correctly and added missing leader checks in TaskManager scheduler callbacks. #71265 #71156
NaN row-count estimation for MCV-only histograms. #71241
Packaging issues caused by a missing s3-transfer-manager dependency in the AWS SDK. #71230
Thread-local ConnectContext pollution after leader forwarding. #71141
Orphaned delvec entries when write and compaction transactions are published in the same batch. #71001 #71049 #71107
Missing partition predicates in short-circuit point lookups. #71124
Potential hash-table data loss during aggregation spill set_finishing. #70851
Query-progress HTTP loopback records from current_queries. #71032
Primary Key tablet rowset metadata loss caused by a GC race during disk re-migration (A→B→A). #70727
DB read-lock leaks in SharedDataStorageVolumeMgr. #70987
Incorrect NullColumn sharing in NullableColumn, BinaryColumn, and locate(). #66037 #70957
Race conditions in global runtime-filter readiness checks and corrected metric overflow caused by ACCUMULATED macro truncation. #70920 #70889
Generated-column is not displayed in DESC and SHOW CREATE TABLE. #70037
An issue with load spill directory cleanup timing, an ASAN crash in memory table spiller workgroup handling, and CN crashes when scanning empty tablets with physical split enabled. #70778 #64379 #70281
Incorrect Content-Length handling when proxy_pass_request_body is off. #70821
Issues with connector scan retry state handling and multiple schema pushdown issues for INSERT INTO BY NAME ... FROM FILES(). #70762 #70774 #70622 #70621
Invalid conjunct pushdown in MySQL and JDBC scan nodes that caused BE predicate type errors. #70694
Incomplete and partially written Iceberg manifest cache entries, and bypassed catalog caching when vended credentials are enabled. #70675 #70652 #69434
Ubuntu runtime dependency issues by installing libssl-dev. #70688
User authentication strings are not masked in audit logs and SQL redaction. #70360
External materialized view refresh issues for Iceberg-like connectors. #70589 #70523
array_map crashes on null literal arrays and BE crashes when a child iterator is exhausted in MaskMergeIterator. #70629 #70539
starlet configuration updates were incorrectly captured through std::call_once. #70482
Robustness issue with Iceberg materialized view refresh when snapshot timestamps are non-monotonic. #70382
Issues that forced materialized view refresh is not supported, and duplicated partition names during materialized view refresh. #70381 #70354
Incorrect distribution-column handling across partitions in OlapTableSink. #70310
File-existence checks are not cached across tablet metadata versions during missing-file validation. #70364
Issue with dataVersion handling during RESTORE, and incorrect materialized view rewrite logic. #70373 #69751
Negative query_pool memory accounting during ingestion, and high FE OOM risk under high concurrency. #70228 #68444
Incorrect SLF4J parameterized logging. #70330
AuditEventProcessor exiting on OutOfMemoryException. #70206
Adjusted handling of column-mode partial updates for Primary Key tables; the initial corruption fix was reverted in this release cycle for follow-up work. #69652
Equality and deduplication issues in InformationFunction. #70464
brpc connection retries when exceptions are wrapped in NoSuchElementException. #70203
Lock-free materialized view rewrite does not fallback to live metadata. #70475
Issue with JoinHashTable::merge_ht() that it does not skip dummy rows for expression-based join-key columns. #70465

3.5.15

Release Date: March 26, 2026

Behavior Changes

Improved sql_mode handling: when DIVISION_BY_ZERO or FAIL_PARSE_DATE mode is set, division by zero and date parse failures in str_to_date/str2date now return an error instead of being silently ignored. #70004
When sql_mode is set to FORBID_INVALID_DATE, invalid dates in INSERT VALUES clauses are now correctly rejected instead of being bypassed. #69803
Expression partition generated columns are now hidden from DESC and SHOW CREATE TABLE output. #69793
Client ID is no longer included in audit logs. #69383
The FORCE option for REFRESH EXTERNAL TABLE has been reverted and is no longer supported. #70428
Aligned the backslash escape sequence usage in LIKE predicates with MySQL. Users need to specify four backslashes \\\\ in sequence in a LIKE predicate to match a literal backslash \ in the data, and eight backslashes \\\\\\\\ in the predicate to match two consecutive literal backslashes \\.

Improvements

Allowed disabling split and reverse scan ranges for descending TopN by setting desc_hint_split_range to 0 or less. #70307
information_schema now shows comments for external catalog tables. #70197
Added EXPLAIN and EXPLAIN ANALYZE support for INSERT statements in Trino dialect. #70174
Added configurable parameters for CatalogRecycleBin to control recycle bin behavior. #69838
Improved ADMIN REPAIR TABLE and SHOW TABLET STATUS to provide better repair and status information. #69656
Blacklisted queries are now excluded from error metrics. #69621
Added support for SHOW TABLET STATUS for cloud-native tablets in shared-data deployments. #69616
Reduced overhead of Primary Key tablet statistics collection in shared-data clusters. #69548
Added support for dynamic configuration of the execution state report thread pool size. #69142

Bug Fixes

Fixed the following bugs:

Data version not set when restoring a tablet. #70373
Table comment not set when creating a Hive table. #70318
Constant folding with double precision arithmetic producing INF instead of returning an error. #70309
Iceberg materialized view refresh failing when snapshot timestamps are non-monotonic. #70382
toIcebergTable function using common instead of comment in property mapping. #70267
Root user not correctly bypassing Ranger permission checks in all scenarios. #70254
AuditEventProcessor thread exiting unexpectedly when an OutOfMemoryException occurs. #70206
Out-of-bounds access in cal_new_base_version during schema change publish. #70132
Partition predicates pruned unexpectedly due to type mismatch in boundary comparison. #70097
str_to_date losing microsecond precision in BE runtime. #70068
Crash in join spill process when set_callback_function is called. #70030
DCHECK failure in DeltaWriter::close() when called from a bthread context. #69960
Use-after-free race condition in AsyncDeltaWriter close/finish lifecycle. #69940
Journal replay not awaited in changeCatalogDb on follower FE, causing consistency issues. #69834
Race condition causing missed write transaction finished editlog. #69899
Several known CVEs addressed. #69863
Incorrect LIKE pattern matching with backslash escape sequences. #69775
Expression analysis failing after renaming a partition column. #69771
Use-after-free crash in AsyncDeltaWriter::close. #69770
Potential bugs in PartitionColumnMinMaxRewriteRule caused by incorrect Partition.hasStorageData results. #69751
Duplicated CSV compression suffix in file sink output file names. #69749
Lake capture_tablet_and_rowsets operation accessible without experimental config flag. #69748
Corrupted cache for Primary Key SST tables. #69693
Use-after-free in AsyncFlushOutputStream. #69688
Incorrect retention clock reset and incomplete scan in disableRecoverPartitionWithSameName. #69677
Tablet info not fetched correctly based on run mode in SchemaBeTabletsScanner. #69645
Incorrect minimum partition pruning with shadow partitions. #69641
Different transactions publishing the same version after graceful exit. #69639
Iterator undefined behavior in get_column_values when rssid is not found. #69617
KILL ANALYZE statement sometimes not stopping a running ANALYZE TABLE operation. #69592
Materialized view force refresh bugs for partition tables. #69488

3.5.14

Release Date: March 5, 2026

Improvements

Added SST read/write failure metrics for Primary Key index in Lake tables. #69513
Added a counter metric for "segment file not found" errors. #69543
Extracted range predicates from scalar-subquery containing convert_tz. #69055
Supports complex type for Paimon tables. #66784
Deferred remote load Spill Directory removal. #68803
Supports repairing cloud-native tables. #67108
Supports inserting ARRAY type to Hive table in CSV format. #67355

Bug Fixes

The following issues have been fixed:

Unexpected behavior caused by exceptions of RowGroupWriter. #69568
Sort key not including newly added key columns after schema change on Aggregate Key/Unique Key tables. #69529
Mertic value g_publish_version_failed_tasks does not reflect the real situation during the resource_busy state. #69526
Rowset files are removed when moving Primary Key tablets to trash. #69438
Lock leak in addPartitions caused by name-based table lookup after concurrent SWAP. #69284
isInternalCancelError used equals instead of startsWith. #69523
Pipeline blocks or crashes when _writer->Close() throws an exception other than ParquetStatusException. #69492
A Hadoop-client lib bug. #69503
Success is mistakenly returned while write operations fails. #69473
CVE-2025-67721. #69138
Issue with RuntimeFilter with low-cardinality optimization in share-data clusters. #64669
Materialized view tablet meta inconsistency between FE leader and follower. #69428
Rollup handler's active transaction ID was not considered in computeMinActiveTxnId. #69285
Arrow Flight Proxy issue with multiple FE. #68300
Concurrency bug of function field. #69315
DROP FUNCTION IF EXISTS ignored ifExists flag. #69216
Lacking case-insensitive username normalization for LDAP authentication. #67966
Certain kinds of partitions cannot be written. #68221
Projection loss in materialized view rewrite due to shared mutable state. #69063
Issue with case-insensitive partition lookup in query table copy. #69173
All-null value handling bug in synchronous materialized views. #69136
mv onReload issues when visiting external catalogs. #68926
DISTINCT ORDER BY alias issues for duplicated constants. #69014
Incorrect query results after modifying CHAR column length in shared-data clusters. #68808
Issue with Azure ABFS/WASB FileSystem cache key. #68901
Incorrect predicate rewrite for OUTER JOIN with constant-side column reference. #67072
IllegalArgumentException comparator transitivity violation. #68743
Issue caused by the query lifetime being shorter than the fragment in report_fragment. #67219
Low-cardinality rewrite NPE caused by shared DecodeInfo. #68799
Missing pcu_upt_cnt metric. #68845
JSON-flatten array/object conflict on identical paths. #68804
ClonExpr nullable bug. #68800

3.5.13

Release Date: February 13, 2026

Improvements

Added an FE configuration enable_table_metrics_collect to control the collection of table-level metrics. #68691
Supports setting the default Warehouse for Merge Commit at user level. #68616

Bug fixes

The following issues have been fixed:

Issue with source partition checking in replication transactions. #68883
Used labels were not identified when labels were specified in BEGIN TRANSACTION. #68660
JOIN ON clause bug with CTE scope. #68809
Overlapping range partitions can be created when an explicit lower bound is provided. #68255
Incorrect parser logic when SQL dialect downgrades from Trino to StarRocks. #68725
Issue with pruning projection columns. #68242
Issue with subquery scope check. #68415
Unmatched type cast in the function analyzer. #66749
Incorrect candidate materialized view selection logic. #68571
The Thrift accept thread exits on exception. #68644
Inaccurate Iceberg data file size estimation. #68787
Lake table memory leak issue. #68678
Deadlock when the HMS connection pool is full. #68033
Iceberg delete column nullability issue. #68649
Materialized views hold large external tables. #68171
Iceberg table cache memory limit issue. #67769
Wrong timeout parameter is used for PocoHttpClient. #68765
BE compile failure with Clang. #68805
Materialized view was reloaded multiple times during startup. #62351
CVE-2025-27821. #68529
Variadic functions return incorrect date values in certain scenarios. #67947

3.5.12

Release Date: January 22, 2026

Improvements

Added a cleaner for BrpcStubCache to clean up unused connections. #61417
Supports batch processing for statistics delete (for dropped tables) and Edit Log write requests. #67896
Preserves SQL comments in Audit Logs when encryption is required. #63298
Added the warehouse_name label to the materialized view metrics. #67715
Improved identifier wrapping for JDBC table and column names. #67853
Added the CLIENT_FACTORY property to the Iceberg JDBC catalog. #67613

Bug fixes

The following issues have been fixed:

Variadic functions return wrong dates when mixing DATE and DATETIME types. #67947
NormalizePredicateRule oscillation on non-deterministic expressions. #67923
Low cardinality bugs with the Lambda function. #67843
Subfield expression does not collect children subfields. #67850
NPE in RBO Join reorder when child statistics are missing. #67693
BE crash due to MemTable finalize failed. #67787
Temporary partitions are not cleaned up after FE restart for dynamic overwrite. #67629
Inaccurate I/O statistics of Compaction. #67524
Incorrect logic in physical partition comparison across clusters during replication transaction. #67616
Issue with SQL Server and Oracle identifier symbol handling. #67965
NPE in the Iceberg metadata table query due to missing configuration propagation. #67151
Issue with the f``iles() schema detection for empty Parquet or ORC files. #67762
Inaccurate value of metrics in Profile caused by UNION ALL on Hive tables. #67912
Lacking support for data retrieval from Arrow Flight proxy for FE queries. #67794
SIGSEGV crash during automatic partition creation caused by a race condition in OlapTableSink::is_full(). #67566

3.5.11

Release date: January 5, 2026

Improvements

Supports Arrow Flight data retrieval from inaccessible nodes. #66348
Logs the cause (including the triggering process information) in the SIGTERM handler. #66737
Added an FE configuration enable_statistic_collect_on_update to control whether UPDATE statements can trigger automatic statistics collection. #66794
Supports configuring networkaddress.cache.ttl. #66723
Improve the “no rows imported” error message. #66624 #66535
Optimized deltaRows with lazy evaluation for large partition tables. #66381
Optimized materialized view rewrite performance. #66623
Supports single-tablet ResultSink optimization in shared-data clusters. #66517
rewrite``_``simple``_``agg``_``to``_``meta``_``scan is enabled by default. #64698
Supports pushing down GROUP BY expressions and materialized view rewrite. #66507
Add overloaded newMessage methods to improve materialized view logs. #66367

Bug Fixes

The following issues have been fixed:

A Publish Compaction crash when the input rowset is not found. #67154
Significant CPU overhead and lock contention caused by repetitive invocation of update_segment_cache_size when querying tables with a large number of columns. #66714
MulticastSinkOperator stuck in the OUTPUT_FULL state. #67153
A “column not found” issue in the skew join hint. #66929
The growth of all tablets continues unabated, and the sum of pending and running tablets is not the total number of tablets. #66718
Transactions in the Compaction map built during Leader startup cannot be accessed by CompactionScheduler and will never be removed from the map. #66533
Delta Lake table refresh does not take effect. #67156
CN crash at queries against non-partitioned Iceberg tables with DATE predicates. #66864
Statements in Profiles cannot be correctly displayed when multiple statements are submitted. #67097
Missing dictionary information during collection because Meta Reader does not support reading from Delta column group files. #66995
Potential Java heap OOM in Java UDAF. #67025
BE crash due to the incorrect logic of ranking window optimization without PARTITION BY and ORDER BY. #67081
Misleading log level for timezone cache miss. #66817
Crash and incorrect results caused by the incorrect can_use_bf checking when merging runtime filters. #67021
Issue about pushing down runtime bitset filter with other OR predicates. #66996
Patch critical fix from lz4. #67053
AsyncTaskQueue deadlock issue. #66791
Cache inconsistency in ObjectColumn. #66957
RewriteUnnestBitmapRule causes wrong output column types. #66855
Data races and data loss when there are WRITE or FLUSH tasks after FINISH tasks in the Delta Writer. #66943
Invalid load channel and misleading internal errors caused by reopened load channels that were previously aborted. #66793
Bugs of Arrow Flight SQL. #65889
Issues when querying renamed columns with MetaScan. #66819
Hash column is not removed before flushing chunk in partitionwise spillable aggregation when skew elimination is off. #66839
BOOLEAN type default values were not correctly handled when stored as string literals. #66818
decimal2decimal cast unexpectedly returns the input column as the result directly. #66773
NPE in query planning during schema change. #66811
LocalTabletsChannel and LakeTabletsChannel deadlock. #66748
publish_version log shows empty txn_ids with new FE. #66732
Incorrect behavior of the FE configuration statistic_collect_query_timeout. #66363
UPDATE statements do not support statistics collection. #66443
Case rewrite errors related to low cardinality. #66724
Statistics query failure when the column list is empty. #66138
Usage/record mismatch when switching warehouse via hint. #66677
ANALYZE TABLE statements lack ExecTimeout. #66361
array_map returns wrong results from constant unary expressions. #66514
Foreign key constraints are lost after FE restart. #66474
max(not null string) on empty table throws std::length_error. #66554
Concurrency issue between Primary Key index Compaction and Apply. #66282
Improper behavior of EXPLAIN <query>. #66542
Issue when sinking DECIMAL128 to Iceberg table column. #66071
JSON length check issue for JSON → CHAR/VARCHAR when the target length equals the minimum. #66628
An expression children count error. #66511

3.5.10

Release date: December 15, 2025

Improvements

Supports dumping plan node IDs in BE crash logs to speed up locating problematic operators. #66454
Optimized scans on the system views in information_schema to reduce the overhead. #66200
Added two histogram metrics (slow_lock_held_time_ms and slow_lock_wait_time_ms) to provide better observability for slow lock scenarios and distinguish between long-held locks and high lock contention. #66027
Optimized replica lock handling in tablet report and clone flows by switching the lock from database level to table level, reducing lock contention and improving scheduling efficiency. #61939
Avoided outputting columns in BE storage, and pushed down predicate computation to BE storage. #60462
Improved query profile accuracy when deploying scan ranges in background threads. #62223
Improved profile accounting when deploying additional tasks, so CPU time is not repetitively counted. #62186
Added more detailed error messages when a referenced partition does not exist, making failures easier to diagnose. #65674
Made sample-type cardinality estimation more robust in corner cases to improve row-count estimates. #65599
Added a partition filter when loading statistics to prevent INSERT OVERWRITE from reading stale partition statistics. #65578
Splited pipeline CPU execution_time metrics into separate series for queries and loads, improving observability by workload type. #65535
Supported enable_statistic_collect_on_first_load at table granularity for finer-grained control over statistics collection on the first load. #65463
Renamed the S3-dependent unit test from PocoClientTest to an S3-specific name to better reflect its dependency and intent. #65524

Bug Fixes

The following issues have been fixed:

libhdfs crashes when StarRocks is started with an incompatible JDK. #65882
Incorrect query results caused by PartitionColumnMinMaxRewriteRule. #66356
Rewrite issues due to the materialized view metadata not refreshed when resolving materialized views by AST keys. #66472
The trim function crashes or produces wrong results when trimming specific Unicode whitespace characters. #66428, #66477
Failures in load metadata and SQL execution that still referenced a deleted warehouse. #66436
Wrong results when group execution Join is combined with window functions. #66441
A possible FE null pointer in resetDecommStatForSingleReplicaTabletUnlocked. #66034
Missing Join runtime filter pushdown optimization in shared-data clusters for LakeDataSource. #66354
Parameters are inconsistent for runtime filter transmit options (timeouts, HTTP RPC limits, etc.) because they are not forwarded to receivers. #66393
Automatic partition creation fails when partition values already exist. #66167
Inaccurate scan statistics in audit logs when predicates have high selectivity. #66280
Incorrect query results because non-deterministic functions are pushed below operators. #66323
Exponential growth in the number of expressions caused by CASE WHEN. #66324
Materialized view compensation bugs when the same table appears multiple times in a query with different partition predicates. #66369
BE becomes unresponsive when using fork in subprocesses. #66334
CVE-2025-66566 and CVE-2025-12183. #66453, #66362
Errors caused by nested CTE reuse. #65800
Issues due to the lack of validation on conflicting schema-change clauses. #66208
Improper rowset GC behavior when rowset commit fails. #66301
A potential use-after-free when counting down pipelines. #65940
The W``arehouse field is NULL in information_schema.loads for Stream Load. #66202
Issues with materialized view creation when the referenced view has the same name as its base table. #66274
The global dictionary is not updated correctly under some cases. #66194
Incorrect query profile logging for queries forwarded from Follower nodes. #64395
BE crash when caching SELECT results and reordering schema. #65850
Shadow partitions are dropped when dropping partitions by expression. #66171
DROP tasks run when a CLONE task exists for the same tablet. #65780
Stability and observability issues because RocksDB log file options were not properly set. #66166
Incorrect materialized view compensation that could produce NULL results. #66216
BE reported as alive even after receiving SIGSEGV. #66212
Bugs in Iceberg scans. #65658
Regression coverage and stability issues for Iceberg view SQL test cases. #66126
Unexpected behavior because set_collector is invoked repetitively. #66199
Ingestion failures when column-mode partial updates are used together with conditional updates. #66139
Temporary partition value conflicts under concurrent transactions. #66025
An Iceberg table cache bug where Guava LocalCache could retain stale entries even when cache.size() == 0, causing refresh to be ineffective and queries to return outdated tables. #65917
Incorrect format placeholder in LargeInPredicateException, causing the actual number of LargeInPredicate occurrences to be incorrectly reported in the error message. #66152
NullPointerException in ConnectScheduler’s timeout checker when connectContext is null. #66136
Crashes caused by unhandled exceptions thrown from threadpool tasks. #66114
Data loss when pushing down DISTINCT LIMIT in certain plans. #66109
multi_distinct_count not updating distinct_size after the underlying hash set is converted to a two-level hash set, which could lead to incorrect distinct counts. #65916
A race condition when an exec group submits the next driver which could trigger Check failed: !driver->is_in_blocked() and abort the BE process. #66099
INSERT failures when running ALTER TABLE ADD COLUMN with a default value concurrently with INSERT, due to mismatched types for the newly added column’s default expression. #65968
An issue where MemoryScratchSinkOperator could remain in pending_finish after RecordBatchQueue was shut down when SparkSQL exited early, causing the pipeline to hang. #66041
A core dump when reading Parquet files that contain empty row groups. #65928
Recursive calls and potential stack overflow at high DOP because the event scheduler’s readiness check is complicated. #66016
Asynchronous materialized view refresh skips updates when the Iceberg base table contains expired snapshots. #65969
Potential issues in predicate reuse and rewrite because the optimizer relies solely on hashCode to distinguish differences in predicates. #65999
In the refresh of an asynchronous materialized view with multi-level partitioned base tables, only the parent partition metadata was checked, while sub-partition updates were skipped. #65596
Statistics collection issues where AVG(ARRAY_LENGTH(...)) could return NULL for empty result sets. #65788
Runtime profile counters are not correctly updating or clearing their min/max values during incremental updates on both BE and FE. #65869
Incorrect logic to obtain the image journal ID when creating a cluster snapshot to ensure the snapshot uses the correct log position. #65970
Results are misreported when the cleanup fails due to incorrect status checking logic in file cleanup error handling. #65709
A possible infinite loop in certain plans isVariable() in DictMappingOperator. #65743
Failures and missing audit/profile data because ConnectContext is not passed into scan-range deployment threads. #63544
A use-after-free issue in the local Primary Key index manager when the storage engine is stopped. #65534
Statistics collection issues for INSERT OVERWRITE with dynamic overwrite. #65657
Concurrency issues caused by coarse-grained locks in DiskAndTabletLoadReBalancer. #65557
Slow locks cannot be detected and reported correctly for the lack of slow-lock detection for critical locks. #65559
NullPointerException when replaying upsert transaction state after the target database has been dropped. #65595
Stale statistics were used because outdated partition statistics are not dropped after statistics collection triggered by INSERT OVERWRITE. #65586
Data race in partition ID allocation that could lead to ID conflicts under concurrency. #65608
Missing tablet IDs when retrieving initial tablet metadata. #65550
Incorrect record information for PREPARE/EXECUTE statements in audit and profile logs. #65448
Potential crashes because the non–thread-safe has_output function is called from multiple threads. #65514
MemTable finalize tasks cannot be properly tracked because the memtable_finalize_task_total counter metric is lacking. #65548
Query ID collisions in Arrow Flight, causing multiple queries no longer share the same query ID. #65558
Lock conflicts for TabletChecker.doCheck() with other operations. #65237
Scan behavior is inconsistent between shared-data and shared-nothing clusters, causing query semantics to differ. #61100

3.5.9

Release date: November 26, 2025

Improvements

Added transaction latency metrics to FE for observing timing across transaction stages. #64948
Supports overwriting S3 unpartitioned Hive tables to simplify full-table rewrites in data lake scenarios. #65340
Introduced CacheOptions to provide finer-grained control over tablet metadata caching. #65222
Supports sample statistics collection for INSERT OVERWRITE to ensure statistics stay consistent with the latest data. #65363
Optimized the statistics collection strategy after INSERT OVERWRITE to avoid missing or incorrect statistics due to asynchronous tablet reports. #65327
Introduced a retention period for partitions dropped or replaced by INSERT OVERWRITE or materialized view refresh operations, keeping them in the recycle bin for a while to improve recoverability. #64779

Bug Fixes

The following issues have been fixed:

Lock contention and concurrency issues related to LocalMetastore.truncateTable(). #65191
Lock contention and replica check performance issues related to TabletChecker. #65312
Incorrect error logging when changing user via HTTP SQL. #65371
Checksum failures caused by DelVec CRC32 upgrade compatibility issues. #65442
Tablet metadata load failures caused by RocksDB iteration timeout. #65146
When the internal flat_path string is empty because the JSON hyper extraction path is $ or all paths are skipped, calling substr will throw an exception and cause BE crash. #65260
The PREPARED flag in fragment execution is not correctly set. #65423
Inaccurate write and flush metrics caused by duplicated load profile counters. #65252
When multiple HTTP requests reuse the same TCP connection, if a non‑ExecuteSQL request arrives after an ExecuteSQL request, the HttpConnectContext cannot be unregistered at channel close, causing HTTP context leaks. #65203
MySQL 8.0 schema introspection errors (Fixed by adding session variables default_authentication_plugin and authentication_policy). #65330
SHOW ANALYZE STATUS errors caused by unnecessary statistics collection for temporary partitions created after partition overwrite operations. #65298
Global Runtime Filter race in the Event Scheduler. #65200
Data Cache is aggressively disabled because the minimum Data Cache disk size constraint is too large. #64909
An aarch64 build issue related to the gold linker automatic fallback. #65156

3.5.8

Release date: November 10, 2025

Improvements

Upgraded Arrow to 19.0.1 to support the Parquet legacy list to include nested, complex files. #64238
FILES() supports legacy Parquet LIST encodings. #64160
Automatically determine the Partial Update mode based on the session variable and the number of inserted columns. #62091
Applied low-cardinality optimization on analytic operators above table functions. #63378
Added configurable table lock timeout to finishTransaction to avoid blocking. #63981
Shared-data clusters support table-level scan metrics attribution. #62832
Window functions LEAD/LAG/FIRST_VALUE/LAST_VALUE now accept ARRAY type arguments. #63547
Supports constant folding for several array functions to improve predicate pushdown and join simplification. #63692
Supports batched API to optimize tabletNum retrieval for a given node via SHOW PROC /backends/{id}. Added an FE configuration item enable_collect_tablet_num_in_show_proc_backend_disk_path (Default: true). #64013
Ensured INSERT ... SELECT reads the freshest metadata by refreshing external tables before planning. #64026
Added capacity_limit_reached checks to table functions, NL-join probe, and hash-join probe to avoid constructing overflowing columns. #64009
Added FE configuration item collect_stats_io_tasks_per_connector_operator (Default: 4) for setting the maximum number of tasks to collect statistics for external tables. #64016
Updated the default partition size for sample collection from 1000 to 300. #64022
Increased lock table slots to 256 and added rid to slow-lock logs. #63945
Improved robustness of Gson deserialization in the presence of legacy data. #63555
Reduced metadata lock scope for FILES() schema pushdown to cut lock contention and planning latency. #63796
Added Task Run execute timeout checker by introducing an FE configuration item task_runs_timeout_second, and refined cancellation logics for overdue runs. #63842
Ensured REFRESH MATERIALIZED VIEW ... FORCE always refreshes target partitions (even in inconsistent or corrupted cases). #63844

Bug Fixes

The following issues have been fixed:

An exception when parsing the Nullable (Decimal) type of ClickHouse. #64195
An issue with tablet migration and Primary Key index lookup concurrency. #64164
Lack of FINISHED status in materialized view refresh. #64191
Schema Change Publish does not retry in shared-data clusters. #64093
Wrong row count statistics on Primary Key tables in Data Lake. #64007
When tablet creation times out in shared-data clusters, node information cannot be returned. #63963
Corrupted Lake DataCache cannot be cleared. #63182
Window function with IGNORE NULLS flags can not be consolidated with its counterpart without iIGNORE NULLS flag. #63958
Table compaction cannot be scheduled again after FE restart if the compaction was previously aborted. #63881
Tasks fail to be scheduled if FE restarts frequently. #63966
An issue with GCS error codes. #64066
Instability issue with StarMgr gRPC executor. #63828
Deadlock when creating an exclusive work group. #63893
Cache for Iceberg tables is not properly invalidated. #63971
Wrong results for sorted aggregation in shared-data clusters. #63849
ASAN error in PartitionedSpillerWriter::_remove_partition. #63903
BE crash when failing to get splits from morsel queue. #62753
A bug with aggregate push-down type cast in materialized view rewrite. #63875
NPE when removing expired load jobs in FE. #63820
Partitioned Spill crash when removing partitions. #63825
Materialized view rewrite throws IllegalStateException under certain plans. #63655
NPE when creating a partitioned materialized view. #63830

3.5.7

Release date: October 21, 2025

Improvements

Improved memory statistics accuracy for Scan operators by introducing retry backoff under heavy memory contention scenarios. #63788
Optimized materialized view bucketing inference by leveraging existing tablet distribution to prevent excessive bucket creation. #63367
Revised the Iceberg table caching mechanism to enhance consistency and reduce cache invalidation risks during frequent metadata updates. #63388
Added the querySource field to QueryDetail and AuditEvent for better traceability of query origins across APIs and schedulers. #63480
Enhanced Persistent Index diagnostics by printing detailed context when duplicate keys are detected in MemTable writes. #63560
Reduced lock contention in materialized view operations by refining lock granularity and sequencing in concurrent scenarios. #63481

Bug Fixes

The following issues have been fixed:

Materialized view rewrite failures caused by type mismatch. #63659
regexp_extract_all has wrong behavior and lacks support for pos=0. #63626
Degraded scan performance caused by the profitless simplification of CASE WHEN with complex functions. #63732
Incorrect DCG data reading when partial updates switch from column mode to row mode. #61529
A potential deadlock during initialization of ExceptionStackContext. #63776
Crashes in Parquet numeric conversion for ARM architecture machines. #63294
An issue caused by the aggregate intermediate type uses ARRAY<NULL_TYPE>. #63371
Stability issue caused by incorrect overflow detection when casting LARGEINT to DECIMAL128 at sign-edge cases (for example, INT128_MIN) #63559
LZ4 compression and decompression errors cannot be perceived. #63629
ClassCastException when querying tables partitioned by FROM_UNIXTIME on INT-type columns. #63684
Tablets cannot be repaired after a balance-triggered migration when the only valid source replica is marked DECOMMISSION. #62942
Profiles lost SQL statements and Planner Trace when the PREPARE statement is used. #63519
The extract_number, extract_bool, and extract_string functions are not exception-safe. #63575
Shutdown tablets cannot be garbage-collected properly. #63595
Profiles showing SQL as omit for returns of the PREPARE/EXECUTE statements. #62988
date_trunc partition pruning with combined predicates that mistakenly produced EMPTYSET. #63464
Crashes in release builds due to the CHECK in NullableColumn. #63553

3.5.6

Release date: September 22, 2025

Improvements

A decommissioned BE will be forcibly dropped when all its tablets are in the recycle bin, to avoid the decommission being blocked by those tablets. #62781
Vacuum metrics will be updated when Vacuum succeeds. #62540
Added thread pool metrics to the fragment instance execution state report, including active threads, queue count, and running threads. #63067
Supports S3 path-style access in shared-data clusters to improve compatibility with MinIO and other S3-compatible storage systems. You can enable this feature by setting aws.s3.enable_path_style_access to true when creating a storage volume. #62591
Supports resetting the starting point of the AUTO_INCREMENT value via ALTER TABLE`` <table_name>`` AUTO_INCREMENT`` = 10000;. #62767
Supports using Distinguished Name (DN) in Group Provider for group matching, improving the user group solution for LDAP/Microsoft Active Directory environments. #62711
Supports Azure Workload Identity authentication for Azure Data Lake Storage Gen2. #62754
Added transaction error messages to the information_schema.``loads view to aid failure diagnosis. #61364
Supports reusing common expressions for complex CASE WHEN expressions in Scan predicates to reduce repetitive computation. #62779
Uses the REFRESH (instead of ALTER) privilege on the materialized view to execute REFRESH statements. #62636
Disabled low-cardinality optimization on Lake tables by default to avoid potential issues. #62586
Enabled tablet balancing between workers by default in shared-data clusters. #62661
Supports reusing expressions in outer-join WHERE predicates to reduce repetitive computation. #62139
Added Clone metrics in FE. #62421
Added Clone metrics in BE. #62479
Added an FE configuration item enable_statistic_cache_refresh_after_write to disable statistics-cache lazy refresh by default. #62518
Masked credential information in SUBMIT TASK for better security. #62311
json_extract in the Trino dialect returns a JSON type. #59718
Supports ARRAY type in null_or_empty. #62207
Adjusted the size limit for the Iceberg manifest cache. #61966
Added a remote file-cache limit for Hive. #62288

Bug Fixes

The following issues have been fixed:

Secondary replicas hang indefinitely due to negative timeout values, which cause incorrect timestamp comparisons. #62805
PublishTask may be blocked when TransactionState is REPLICATION. #61664
Incorrect repair mechanism for Hive tables that have been dropped and recreated during materialized view refresh. #63072
Incorrect execution plans were generated after the materialized view aggregation push‑down rewrite. #63060
ANALYZE PROFILE failures caused by PlanTuningGuide producing unrecognized strings (null explainString) in the query profiles. #63024
Inappropriate return type of hour_from_unixtime and incorrect rewrite rule of CAST. #63006
NPE in Iceberg manifest cache under data races. #63043
Shared-data clusters lack support for colocation in materialized views. #62941
Iceberg table Scan Exception during Scan Range deployment. #62994
Incorrect execution plans were generated for view-based rewrite. #62918
Errors and disrupted tasks due to Compute Nodes are not gracefully shut down on exit. #62916
NPE when Stream Load execution status updates. #62921
An issue with statistics when the column name and the name in the PARTITION BY clause differ in case. #62953
Wrong results are returned when the LEAST function is used as a predicate. #62826
Invalid ProjectOperator above the table-pruning frontier CTEConsumer. #62914
Redundant replica handling after Clone. #62542
Failed to collect Stream Load profiles. #62802
Ineffective disk rebalancing caused by improper BE selection. #62776
A potential NPE crash in LocalTabletsChannel when a missing tablet_id leads to a null delta writer. #62861
KILL ANALYZE does not take effect. #62842
SQL syntax errors in histogram stats when MCV values contain single quotes. #62853
Incorrect output format of metrics for Prometheus. #62742
NPE when querying information_schema.analyze_status after the database is dropped. #62796
CVE-2025-58056. #62801
When SHOW CREATE ROUTINE LOAD is executed, wrong results are returned because the database is considered null if not specified. #62745
Data loss caused by incorrectly skipping CSV headers in files(). #62719
NPE when replaying batch-transaction upserts. #62715
Publish being incorrectly reported as successful during graceful shutdown in shared-nothing clusters. #62417
Crash in asynchronous delta writer due to a null pointer. #62626
Materialized view refresh is skipped because the materialized view version map is not cleared after a failed restore job. #62634
Issues caused by case-sensitive partition column validation in the materialized view analyzer. #62598
Duplicate IDs for statements with syntax errors. #62258
StatisticsExecutor status is overridden due to redundant state assignment in CancelableAnalyzeTask. #62538
Incorrect error messages produced by statistics collection. #62533
Premature throttling caused by insufficient default maximum connections for external users. #62523
A potential NPE in materialized view backup and restore operations. #62514
Incorrect http_workers_num metric. #62457
The runtime filter fails to locate the corresponding execution group during construction. #62465
Tedious results on Scan Node caused by simplifying CASE WHEN with complex functions. #62505
gmtime is not thread-safe. #60483
An issue with getting Hive partitions with escaped strings. #59032

3.5.5

Release date: September 5, 2025

Improvements

Added a new system variable enable_drop_table_check_mv_dependency (default: false). When set to true, if the object to be dropped is referenced by a downstream materialized view, the system prevents the execution of DROP TABLE / DROP VIEW / DROP MATERIALIZED VIEW. The error message lists the dependent materialized views and suggests checking the sys.object_dependencies view for details. #61584
Logs now include the Linux distribution and CPU architecture of the build, to facilitate issue reproduction and troubleshooting. Log format: ... build <hash> distro <id> arch <arch>. #62017
Persisted per-Tablet index and incremental column group file sizes are now cached, replacing on-demand directory scans. This accelerates Tablet status reporting in BE and reduces latency under high I/O scenarios. #61901
Downgraded several high-frequency INFO logs in FE and BE to VLOG, and aggregated task submission logs, significantly reducing redundant storage-related logs and log volume under heavy load. #62121
Improved query performance for External Catalog metadata through information_schema by pushing table filters before calling getTable, avoiding per-table RPCs. #62404

Bug Fixes

The following issues have been fixed:

NullPointerException when fetching partition-level column statistics during the Plan stage due to missing data. #61935
Fixed Parquet write issues with non-empty NULL arrays, and corrected SPLIT(NULL, …) behavior to consistently return NULL, preventing data corruption and runtime errors. #61999
Failure when creating materialized views using CASE WHEN expressions due to incompatible VARCHAR type returns (fixed by ensuring consistency before and after refresh, and introducing a new FE configuration transform_type_prefer_string_for_varchar to prefer STRING and avoid length mismatch). #61996
Statistics for nested CTEs could not be computed outside of memo when enable_rbo_table_prune was false. #62070
In Audit Logs, inaccurate Scan Rows results for INSERT INTO SELECT statements. #61381
ExceptionInInitializerError/NullPointerException during initialization caused FE startup failure when Query Queue v2 was enabled. #62161
BE crash when LakePersistentIndex initialization failed and _memtable cleanup was triggered. #62279
Permission issues during materialized view refresh due to creator roles not being activated (fixed by adding FE configuration mv_use_creator_based_authorization. When set to false, materialized views are refreshed as root, for compatibility with LDAP-authenticated clusters). #62396
Materialized view refresh failures caused by case-sensitive List partition table names (fixed by enforcing case-insensitive uniqueness checks on partition names, aligning with OLAP table semantics). #62389

3.5.4

Release Date: August 22, 2025

Improvements

Added logs to clarify the reason that tablets cannot be repaired.  #61959
Optimized DROP PARTITION information in logs.  #61787
Assigned a large but configurable row count to tables with unknown stats for statistical estimation.  #61332
Added balance statistic according to label location.  #61905
Added colocate group balance statistics to improve cluster monitoring. #61736
Skipped the Publish waiting phase when the number of healthy replicas exceeds the default replica count. #61820
Included the tablet information collection time in the tablet report. #61643
Supports writing Starlet files with tags.  #61605
Supports viewing cluster balance statistics via SHOW PROC.  #61578
Bumped librdkafka to 2.11.0 to support Kafka 4.0 and removed deprecated configurations.  #61698
Added prepared_timeout configuration to Stream Load Transaction Interface.  #61539
Upgraded StarOS to v3.5‑rc3.  #61685

Bug Fixes

The following issues have been fixed:

Incorrect Dict version of random distribution tables. #61933
Incorrect query context in context conditions. #61929
Publish failures caused by synchronous Publish for shadow tablets during ALTER operations. #61887
CVE‑2025‑55163 issue.  #62041
Memory leak in real-time data ingestion from Apache Kafka.  #61698
Incorrect count of rebuild files in the lake persistent index. #61859
Statistics collection on generated expression columns causes cross-database query errors. #61829
Query Cache misaligns in shared-nothing clusters, causing inconsistent results. #61783
High memory usage in CatalogRecycleBin due to retaining deleted partition information.#61582
SQL Server JDBC connections fail when the timeout exceeds 65,535 milliseconds. #61719
Security Integration fails to encrypt passwords, exposing sensitive information. #60666
MIN() and MAX() functions on Iceberg partition columns return NULL unexpectedly.  #61858
Other predicates of Join containing non‑push‑down subfields were incorrectly rewritten.  #61868
QueryContext cancellation can lead to a use‑after‑free situation.  #61897
CBO’s table pruning overlooks other predicates.  #61881
Partial Updates in COLUMN_UPSERT_MODE may overwrite auto-increment columns with zero.  #61341
JDBC TIME type conversion uses an incorrect timezone offset that leads to wrong time values. #61783
max_filter_ratio was not being serialized in Routine Load jobs. #61755
Precision loss in the now(precision) function in Stream Load. #61721
Cancelling a query may result in a “query id not found” error. #61667
LDAP authentication may miss PartialResultException, causing incomplete query results. #60667
Paimon Timestamp timezone conversion issue when the query condition contains DATETIME. #60473

3.5.3

Release Date: August 11, 2025

Improvements

Lake Compaction adds Segment write time statistics. #60891
Disable inline mode for Data Cache writes to avoid performance degradation. #60530
Iceberg metadata scan supports shared file I/O. #61012
Support termination of all PENDING ANALYZE tasks. #61118
Force reuse when there are too many CTE nodes to avoid excessive optimization time. #60983
Added BALANCE type to cluster balance results. #61081
Optimized materialized view rewrite for external tables. #61037
Default value of system variable enable_materialized_view_agg_pushdown_rewrite is changed to true, enabling aggregation pushdown for materialized view queries by default. #60976
Optimized partition statistics lock competition. #61041

Bug Fixes

The following issues have been fixed:

Inconsistent Chunk column size after column pruning. #61271
Synchronous execution of partition statistics loading may cause deadlocks. #61300
Crash when array_map processes constant array columns. #61309
Setting an auto-increment column to NULL results in the system mistakenly rejecting valid data within the same Chunk. #61255
The actual number of JDBC connections may exceed the jdbc_connection_pool_size limit. #61038
FQDN mode did not use IP addresses as cache map keys. #61203
Array column cloning error during array comparison. #61036
Deploying serialized thread pool blockage led to query performance degradation. #61150
OK hbResponse not synchronized after heartbeat retry counter reset. #61249
Incorrect result for the hour_from_unixtime function. #61206
Conflicts between ALTER TABLE jobs and partition creation. #60890
Cache does not take effect after upgrading from v3.3 to v3.4 or later. #60973
Vector index metric hit_count is not set. #61102
Stream Load transactions fail to find the coordinator node. #60154
BE crashes when loading OOM partitions. #60778
INSERT OVERWRITE failed on manually created partitions. #60750
Partition creation failed when partition names matched case-insensitively but had different values. #60909
The system does not support PostgreSQL UUID type. #61021
Case sensitivity issue with column names when loading Parquet data via FILES(). #61059

3.5.2

Release Date: July 18, 2025

Improvements

Collected NDV (number of distinct values) statistics for ARRAY columns to improve query plan accuracy. #60623
Disabled replica balancing for Colocate tables and tablet scheduling in Shared-data clusters to reduce unnecessary log output. #60737
Optimized Catalog access workflow: FE now delays accessing external data sources asynchronously at startup to prevent hanging due to external service unavailability. #60614
Added session variable enable_predicate_expr_reuse to control predicate pushdown. #60603
Supports a retry mechanism when fetching Kafka partition information fails. #60513
Removed the restriction requiring exact mapping of partition columns between materialized views and base tables. #60565
Supports building Runtime In-Filters to enhance aggregation performance by filtering data during aggregation. #59288

Bug Fixes

Fixed the following issues:

COUNT DISTINCT queries crash due to low-cardinality optimization for multiple columns. #60664
Incorrect matching of global UDFs when multiple functions share the same name. #60550
Null pointer exception (NPE) issue during Stream Load import. #60755
Null pointer exception (NPE) issue when starting FE during a recovery from a cluster snapshot. #60604
BE crash caused by column mode mismatch when processing short-circuit queries with out-of-order values. #60466
Session variables set via PROPERTIES in SUBMIT TASK statements did not take effect. #60584
Incorrect results for SELECT min/max queries under specific conditions. #60601
Incorrect bucket pruning when the left side of a predicate is a function, leading to incorrect query results. #60467
Crash for queries against a non-existent query_id via Arrow Flight SQL. #60497

Behavior Changes

The default value of lake_compaction_allow_partial_success is set to true. Compaction operations can now be marked as successful even if partially completed, preventing blockage of subsequent compaction tasks. #60643

3.5.1

Release Date: July 1, 2025

New Features

[Experimental] Starting from v3.5.1, StarRocks introduces a high-performance data transfer channel based on the Apache Arrow Flight SQL protocol, comprehensively optimizing the data import channel and significantly improving transfer efficiency. This solution establishes a fully columnar data transfer pipeline from the StarRocks columnar execution engine to the client, eliminating the frequent row-column conversions and serialization overhead typically seen in traditional JDBC and ODBC interfaces, and achieving true zero-copy, low-latency, and high-throughput data transfer capabilities. #57956
Java Scalar UDFs (user-defined functions) now support ARRAY and MAP types as input parameters. #55356
Cross-node data cache sharing: Enables nodes to share cached external table data of data lakes across compute nodes via the network. If a local cache miss occurs, the system first attempts to fetch data from the caches of other nodes within the same cluster. Only if all caches miss will it re-fetch data from remote storage. This feature effectively reduces performance jitter caused by cache invalidation during elastic scaling and ensures stable query performance. A new FE configuration parameter enable_trace_historical_node controls this behavior (Default: false). #57083
Storage Volume adds native support for Google Cloud Storage (GCS): You can now use GCS as a backend storage volume and manage and access GCS resources through the native SDK. #58815

Improvements

Optimized error messages when creating Hive external tables fails. #60076
Optimized count(1) query performance using the file_record_count in Iceberg metadata. #60022
Refined the Compaction scheduling logic to avoid delayed scheduling when all subtasks succeed. #59998
Added JAVA_OPTS="--add-opens=java.base/java.util=ALL-UNNAMED" to BE and CN after upgrading to JDK 17. #59947
Supports modifying the kafka_broker_list property via the ALTER ROUTINE LOAD command when Kafka Broker endpoints change. #59787
Supports reducing build dependencies of the Docker base image through parameters. #59772
Supports accessing Azure using Managed Identity authentication. #59657
Improved error messages when querying external data via Files() function with duplicate path column names. #59597
Optimized LIMIT pushdown logic. #59265

Bug Fixes

Fixed the following issues:

Partition pruning issue when queries include Max and Min aggregations and empty partitions. #60162
Incorrect query results when rewriting queries with materialized views due to missing NULL partitions. #60087
Refresh errors on Iceberg external tables when using partition expressions based on str2date. #60089
Incorrect partition range when creating temporary partitions using the START END syntax. #60014
Incorrect display of Routine Load metrics on non-leader FE nodes. #59985
BE/CN crashes when executing queries containing COUNT(*) window functions. #60003
Stream Load failures when the target table name contains Chinese characters. #59722
Overall loading failures to triple-replica tables when loading to a secondary replica fails. #59762
Missing parameters in SHOW CREATE VIEW output. #59714

Behavior Changes

Some FE metrics include the is_leader label. #59883

3.5.0

Release Date: June 13, 2025

Shared-data Enhancement

Shared-data clusters support generated columns. #53526
Cloud-native Primary Key tables in shared-data clusters support rebuilding specific indexes. The performance of the indexes is also optimized. #53971 #54178
Optimized the execution logic of large-scale data loading operations to avoid generating too many small files in Rowset due to memory limitations. During the import, the system will merge the temporary data blocks to reduce the generation of small files, which improves the query performance after the import and also reduces the subsequent Compaction operations to improve the system resource utilization. #53954

Data Lake Analytics

[Beta] Supports creating Iceberg views in the Iceberg Catalog with Hive Metastore integration. And supports adding or modifying the dialect of the Iceberg view using the ALTER VIEW statement for better syntax compatibility with external systems. #56120
Supports nested namespace for Iceberg REST Catalog. #58016
Supports using IcebergAwsClientFactory to create AWS clients in Iceberg REST Catalog to offer vended credentials. #58296
Parquet Reader supports filtering data with Bloom Filter. #56445
Supports automatically creating global dictionaries for low-cardinality columns in Parquet-formatted Hive/Iceberg tables during queries. #55167

Performance Improvement and Query Optimization

Statistics optimization:
- Supports Table Sample. Improved statistics accuracy and query performance by sampling data blocks in physical files. #52787
- Supports recording the predicate columns in queries for targeted statistics collection. #53204
- Supports partition-level cardinality estimation. The system reuses the system-defined view _statistics_.column_statistics to record the NDV of each partition. #51513
- Supports multi-column Joint NDV collection to optimize the query plan generated by CBO in the scenario where columns correlate with each other. #56481 #56715 #56766 #56836
- Supports using histograms to estimate the Join node cardinality and in_predicate selectivity, thus improving the estimation accuracy in data skew. #57874 #57639
- Optimized Query Feedback. Queries with the identical structure but different parameter values will be categorized as the same type and share the same tuning guide for plan execution optimization. #58306
Supports Runtime Bitset Filter as an alternative for optimization to Bloom Filter in specific scenarios. #57157
Supports pushing down Join Runtime Filter to the storage layer. #55124
Supports Pipeline Event Scheduler. #54259

Partition Management

Supports using ALTER TABLE to merge expression partitions based on time functions for optimized storage efficiency and query performance. #56840
Supports partition Time-to-live (TTL) for List-partitioned tables and materialized views. And supports the property partition_retention_condition in tables and materialized views to allow users to set data retention strategies for list partitions, thus achieving more flexible partition deletion strategies. #53117
Supports using ALTER TABLE to delete partitions specified by common partition expressions, allowing users to flexibly delete partitions in batches. #53118

Cluster Management

Upgraded FE compile target from Java 11 to Java 17 for better system stability and performance. #53617 #57030

Security and Authentication

Supports secure connections encrypted by SSL based on the MySQL protocol. #54877
Enhanced authentication using external systems:
- Supports creating StarRocks users with OAuth 2.0 and JSON Web Token (JWT).
- Supports Security Integration to simplify the authentication process with external systems. Security Integration supports LDAP, OAuth 2.0, and JWT. #55846
Supports Group Provider to obtain the user group information from external authentication services. The group information can then be used in authentication and authorization. Group Provider supports acquiring group information from LDAP, operating systems, or files. Users can query the user group they belong to using the function current_group(). #56670

Materialized Views

Supports creating materialized views with multiple partition columns to allow users to partition the data with a more flexible strategy. #52576
Supports setting query_rewrite_consistency to force_mv to force the system to use the materialized view for query rewrite, thus keeping performance stability at the cost of data timeliness to a certain extent. #53819

Loading and Unloading

Supports pausing Routine Load jobs on JSON parse errors by setting the property pause_on_json_parse_error to true. #56062
[Beta] Supports transactions with multiple SQL statements (currently, only INSERT is supported). Users can start, apply, or undo a transaction to guarantee the ACID (atomicity, consistency, isolation, and durability) properties of multiple loading operations. #53978

Functions

Introduced the system variable lower_upper_support_utf8 on the session and global level, enhancing the support for UTF-8 strings (especially non-ASCII characters) in case conversion functions such as upper() and lower(). #56192
Added new functions:

3.5.21​

Behavior Changes​

Improvements​

Bug fixes​

3.5.20​

Behavior Changes​

Improvements​

Bug fixes​

3.5.19​

Behavior Changes​

Improvements​

Bug fixes​

3.5.18​

Behavior Changes​

Improvements​

Bug fixes​

3.5.17​

Behavior Changes​

Improvements​

Bug fixes​

3.5.16​

Improvements​

Bug fixes​

3.5.15​

Behavior Changes​

Improvements​

Bug Fixes​

3.5.14​

Improvements​

Bug Fixes​

3.5.13​

Improvements​

Bug fixes​

3.5.12​

Improvements​

Bug fixes​

3.5.11​

Improvements​

Bug Fixes​

3.5.10​

Improvements​

Bug Fixes​

3.5.9​

Improvements​

Bug Fixes​

3.5.8​

Improvements​

Bug Fixes​

3.5.7​

Improvements​

Bug Fixes​

3.5.6​

Improvements​

Bug Fixes​

3.5.5​

Improvements​

Bug Fixes​

3.5.4​

Improvements​

Bug Fixes​

3.5.3​

Improvements​

Bug Fixes​

3.5.2​

Improvements​

Bug Fixes​

Behavior Changes​

3.5.1​

New Features​

Improvements​

Bug Fixes​

Behavior Changes​

3.5.0​

Shared-data Enhancement​

Data Lake Analytics​

Performance Improvement and Query Optimization​

Partition Management​

Cluster Management​

Security and Authentication​

Materialized Views​

3.5.21

Behavior Changes

Improvements

Bug fixes

3.5.20

Behavior Changes

Improvements

Bug fixes

3.5.19

Behavior Changes

Improvements

Bug fixes

3.5.18

Behavior Changes

Improvements

Bug fixes

3.5.17

Behavior Changes

Improvements

Bug fixes

3.5.16

Improvements

Bug fixes

3.5.15

Behavior Changes

Improvements

Bug Fixes

3.5.14

Improvements

Bug Fixes

3.5.13

Improvements

Bug fixes

3.5.12

Improvements

Bug fixes

3.5.11

Improvements

Bug Fixes

3.5.10

Improvements

Bug Fixes

3.5.9

Improvements

Bug Fixes

3.5.8

Improvements

Bug Fixes

3.5.7

Improvements

Bug Fixes

3.5.6

Improvements

Bug Fixes

3.5.5

Improvements

Bug Fixes

3.5.4

Improvements

Bug Fixes

3.5.3

Improvements

Bug Fixes

3.5.2

Improvements

Bug Fixes

Behavior Changes

3.5.1

New Features

Improvements

Bug Fixes

Behavior Changes

3.5.0

Shared-data Enhancement

Data Lake Analytics

Performance Improvement and Query Optimization

Partition Management

Cluster Management

Security and Authentication

Materialized Views