Kernel release notes

Version

Release date

Description

References

26.1.1

July 31, 2026

This version incorporates the major features and fixes from Apache Doris 4.1.2.

[New features]

• Added recursive common table expressions through WITH RECURSIVE for hierarchical and graph queries.

• Added the UNNEST syntax to expand arrays and other nested values into rows.

• Added the json_each and json_each_text table-valued functions to expand JSON objects into key-value rows.

• Added the array_cross_product function to compute the Cartesian product of multiple arrays.

• Added the murmur_hash3_128 scalar function, the exponential_moving_average aggregate function, and DataSketches HLL aggregate functions.

• Added current_database as an alias for the existing database() function.

• Added disk-based IVF vector indexes for vector search workloads where the index exceeds available memory.

• Added DOC mode for Variant columns that stores raw documents during ingestion and defers sub-column extraction to reduce write costs for wide JSON documents.

• Added NestedGroup support for Variant columns and the NESTED search operator to query arrays of nested objects without flattening them first.

• Added conditional caching to reuse repeated filter results on unchanged table data.

• Added row-level UPDATE, DELETE, and MERGE INTO support for Iceberg V2 and V3 tables, along with Iceberg V3 delete vectors and row lineage reads.

• Added Iceberg JDBC catalog, sorted writes, static partition overwrite, and the publish_changes procedure.

• Added Paimon JDBC catalog and SQL operations for creating and dropping Paimon databases and tables.

• Added CREATE TABLE, DROP TABLE, and INSERT INTO support for the MaxCompute external catalog.

• Added Parquet metadata table-valued functions to inspect row groups, column statistics, and other file metadata through SQL.

• Added INSERT INTO support for file table-valued functions to export query results to HDFS or S3.

• Added the cdc_stream table-valued function to read MySQL and PostgreSQL change streams and use them as data sources for streaming jobs.

• Added OIDC authentication for MySQL clients, including authentication integration, role mappings, and the information_schema.role_mappings table.

• Added AWS MSK IAM authentication for Kafka Routine Load jobs.

[Improvements]

• Improved stability of large joins, aggregations, and sorts by allowing intermediate data to spill across multiple tiers when memory is constrained.

• Reduced latency and memory usage for complex join and aggregation queries by aggregating rows earlier and improving plans for duplicate and simple aggregations.

• Improved ORDER BY ... LIMIT queries on indexed columns through index-backed lazy materialization, and improved execution of CASE WHEN, explode, and posexplode.

• Improved throughput and memory efficiency for large table scans by adaptively adjusting scan batch sizes.

• Improved repeated Parquet queries by reusing decompressed data pages, and improved Paimon query performance through the native reader.

• Extended MySQL and PostgreSQL CDC streaming jobs with custom table mappings, snapshot-only sync, PostgreSQL schema change handling, column exclusion, and additional PostgreSQL data types.

• Added explicit start offsets, compute group selection, structured failure reasons, and per-job latency and progress metrics for streaming jobs.

• Improved full snapshot CDC reliability for large or data-skewed tables by preventing in-progress tasks from being timed out and retried.

• Changed window_funnel to use the more memory-efficient v2 implementation by default, while keeping the previous implementation as window_funnel_v1. In FIXED mode, rows that do not match any funnel condition no longer break the sequence, so affected queries may return longer funnel chains after upgrade.

• Extended multiple aggregate and array functions to accept TIMESTAMPTZ values, and extended max_by and min_by to support complex types.

• Exposed unified runtime statistics and task progress for running queries.

• Added compute groups to MTMV refresh task output so that refresh placement can be inspected directly.

• Reduced memory pressure during row-store ingestion and parsing of sparse Variant documents.

• Added a default limit of 768 buckets per partition. CREATE TABLE and ADD PARTITION now reject partitions that exceed this limit.

[Fixed issues]

• Fixed an issue where *_diff functions, datediff on zero dates, and convert_tz during daylight saving time transitions produced inconsistent results between compile-time constant folding and runtime execution.

• Fixed an issue where coalesce returned NULL when a nullable float argument contained an unused NaN value and subsequent arguments provided the expected non-NULL result.

• Fixed an issue where runtime filters produced incorrect results when pushed through the nullable side of an outer join.

• Fixed an issue where query plans reused pre-aggregation state across different join branches, potentially producing incorrect plans when branches required different aggregation behavior.

• Fixed an IndexOutOfBoundsException that occurred when querying SELECT * views created on external tables after the source table had columns added and metadata refreshed.

• Fixed an issue where tokenize returned only one row when its first argument was a constant and the query scanned multiple input rows.

• Fixed an issue where json_contains treated numerically equal JSON values (such as 1 and 1.0) as distinct, and where duplicate array candidates produced incorrect containment results.

• Fixed an issue where enable_insert_strict silently altered INSERT statement casting behavior. Insert row filtering and enable_strict_cast now operate independently.

• Fixed an issue where COM_RESET_CONNECTION cleared the selected database and partially reset other session states, potentially breaking pooled MySQL connections after reset.

• Fixed a meet invalid precision: real_precision=0 error when querying legacy DecimalV2 data written by Doris versions earlier than 2.1.0.

• Fixed an issue where MATCH rejected aliased Variant dot-subcolumns (such as msg.trace_id) and where nested Variant searches bound to the wrong inverted index.

• Fixed an issue where array subscripts and element_at returned NULL when applied to Variant sub-paths that the optimizer had pruned.

• Fixed an issue where sparse Variant paths inferred incorrect element types for mixed-type arrays and lost TIMESTAMPTZ values during serialization.

• Fixed a Compaction TAPE_ERROR failure when nullable NestedGroup Variant data had visible nested behavior as NULL or empty.

• Fixed low or zero recall for IVF and PQ vector indexes when training used only a subset of segments, and ANN range search state leaking across segments. Segments with insufficient training rows now skip index creation instead of failing writes.

• Fixed an issue where window_funnel v2 returned incorrect longest chains in DEDUPLICATION and FIXED modes when subsequent starting events produced longer valid sequences.

• Fixed an issue where MTMV skipped required refresh work after excluded_trigger_tables changes, and partitioned MTMV checks failed when the caller provided an immutable excluded table set.

• Fixed an issue where PostgreSQL CDC jobs with multiple tables and offset=initial permanently lost changes during the snapshot phase, and where same-named tables in matching patterns caused Duplicate key failures.

• Fixed an issue where cdc_stream offsets did not survive FE checkpoints, potentially causing recovered jobs to resume from incorrect source positions.

• Fixed a broken pipe error in Stream Load when redirect request bodies were not consumed, and a backend selection failure when backend ID differences exceeded integer range.

• Fixed an issue where JSON ingestion mapped STRUCT sub-fields by position instead of by name when input field order differed from the table schema.

• Fixed an issue where workload group queries were resumed while their memory reservations were still unavailable, potentially bypassing expected pause behavior.

• Fixed an issue where client queries hung until socket timeout when FE leader switching caused the final result packet to be lost.

• Fixed a null pointer failure in transactional INSERT INTO ... VALUES in multi-compute-group deployments when a backend from another compute group was selected.

• Fixed an issue where a failed ADD PARTITION left stale partition metadata when partial catalog updates had completed before rollback began.

• Fixed an issue where vertical compaction ran out of memory on large rowsets due to incorrect available buffer capacity calculation.

• Fixed an issue where migrated Iceberg tables without explicit write format properties were read with incorrect file formats, and Iceberg writes ignored the configured object-store.path data location.

• Fixed incorrect COUNT DISTINCT results on Iceberg V3 row lineage tables and missing required Iceberg field IDs in position delete file writes.

• Fixed Iceberg write failures when partition values contained characters requiring URL encoding, and large sorted Iceberg write failures after spilling intermediate data.

• Fixed an issue where the SQL Server JDBC catalog omitted built-in schemas in SHOW DATABASES when using newer Microsoft JDBC drivers.

• Fixed a No suitable driver found failure when Paimon JDBC system table queries were run through the native Paimon scan path.

Apache Doris 4.1.2

Version

Release date

Description

References

5.0.4

(26.0.4)

July 5, 2026

This version incorporates the major features and fixes from Apache Doris 4.0.6.

[New features]

• Added the mmhash3_u64_v2 hash function.

• Added the json_object_flatten scalar function to flatten nested JSON objects.

• Added lambda comparator support for array_sort, allowing arrays to be sorted with custom comparison logic.

• Added support for Iceberg REST catalog and S3 Tables IAM role authentication.

• Added the ability to track Compaction tasks through system tables.

[Improvements]

• Changed the default trigger threshold for time-series Compaction from 2,000 files to 1,000 files, enabling earlier merging of time-series data.

• Reduced memory pressure during load and write workloads by adaptively adjusting write buffer sizes, allowing online tuning of flush thread pools, and releasing unused write memory earlier.

• Improved Compaction isolation in compute-storage separated mode, reducing resource contention between user workloads and background Compaction.

• Reduced optimizer planning overhead for complex queries and improved row count estimation after partition pruning.

• Reduced statistics collection overhead by skipping excessively long string columns.

• Improved the accuracy of audit logs and metrics for partition near-limit reporting, histogram labels, internal query states, and prepared statement QPS.

[Fixes]

• Fixed an issue where INTERSECT and EXCEPT queries lost NULL rows when predicates were pulled up during planning.

• Fixed an issue where count(null) was incorrectly treated as count(*), and where count queries with MATCH_ALL expressions failed due to NoSuchElementException.

• Fixed an issue where materialized view rewrite returned incorrect results when partition pruning removed always-true predicates, and where rewrite merged partitions that were not needed by the query.

• Fixed an issue where queries with multiple input fragments that received both broadcast and hash-partitioned inputs in the same fragment failed with Rows mismatched! Data may be lost.

• Fixed errors in pushing projections into UNION or reusing CTE producer statistics in queries.

• Fixed an issue where the no_use_cbo_rule hint was silently ignored.

• Fixed multiple TIMESTAMPTZ and datetime issues, including LEAD and LAG preserving incorrect types, daylight saving time (DST) gap and fold handling, elapsed time calculations, and TopN runtime predicate support.

• Fixed an issue where converting scientific notation strings to Decimal produced incorrect results.

• Fixed incorrect results from allow_zero_date and changed from_olap_string to return NULL instead of throwing an exception when datetime parsing fails.

• Fixed integer type inference and parameter parsing issues with user variables.

• Fixed an issue where ALTER VIEW definitions containing COMMENT were not synchronized to follower FEs. This issue only affects multi-FE deployments.

• Fixed an issue where views based on Variant columns returned incorrect results when the view definition lost Variant subfield references.

• Fixed invalid alias rewriting in view definitions and missing colUniqueId for view columns under lazy materialization.

• Fixed an issue where Variant Compaction failed on keyless tables with Variant columns containing uid 0.

• Fixed an issue where Variant subcolumns were lost after row-store partial column updates, and an issue with reading legacy single-part point key subcolumn paths.

• Fixed an issue where duplicate Variant JSON paths retained the wrong occurrence position. The first occurrence is now retained.

• Fixed an issue where Broker Load only loaded the first file when multiple file paths were specified.

• Fixed an issue where PostgreSQL CDC streaming jobs silently dropped DML after task restarts.

• Fixed issues with offset, statistics, and property recovery for streaming jobs after FE checkpoint or restart.

• Fixed Routine Load failures caused by null Kafka metadata state and BE restarts during coordination.

• Fixed CSV parsing issues when files with UTF-8 BOM used enclose characters.

• Fixed an issue where INSERT job statistics disappeared from SHOW LOAD after FE restart, and a memory leak in InsertLoadJob when jobs were in PENDING state.

• Fixed an issue where JDBC catalog type mapping was incorrect when the Doris-compatible MySQL endpoint's version_comment did not contain "Doris".

• Fixed an issue where TVF column aliases were lost in cross-JDBC catalog queries.

• Fixed an issue where Hive DATE values were shifted due to timezone handling in external readers.

• Fixed an issue where predicate filters were skipped when external file scans mixed native and JNI readers.

• Fixed incompatibility between SHOW PARTITIONS and external catalogs or the partitions TVF.

• Fixed an out-of-bounds crash when generating partition columns for Iceberg or Paimon tables.

• Fixed Parquet read and write issues involving INT96 timestamp output, Page V2 encoding, and conditional checks.

• Fixed an issue where TVFs failed when the generated Thrift message exceeded the configured size limit.

• Fixed multiple cloud schema change failures involving local rowset replacement, version gaps, and delete bitmap capture.

• Fixed a race condition between initial dynamic partition setup and concurrent CREATE MATERIALIZED VIEW.

• Fixed an issue where CANCEL ALTER with an empty job list was treated as canceling all rollup jobs.

• Fixed an issue where SHOW TABLET failed when no database was selected, where SHOW BACKENDS field order was mismatched, and where partition cache version was missing from SHOW PROC.

• Fixed BE crashes during concurrent OLAP scans, shared hash table Hash Join with runtime filter release, and IO callback cleanup.

• Fixed Arrow Flight SQL query execution issues when returning results from BE, client IP authentication issues, and Arrow string size handling issues.

Apache Doris 4.0.6

5.0.3

(26.0.3)

April 22, 2026

[New features]

Supports all features of Apache Doris 4.0.5.

Apache Doris 4.0.5

5.0.2-beta

(26.0.2-beta)

April 1, 2026

[New features]

Supports all features of Apache Doris 4.0.4.

Apache Doris 4.0.4

5.0.0-beta

December 31, 2025

[New features]

Supports all features of Apache Doris 4.0.0.

• AI Functions library

• Search functions

• Vector index

Apache Doris 4.0.0

Version

Release date

Description

References

4.1.8

July 22, 2026

[New features]

• Statistics collection now skips string columns whose single-row length exceeds 1,024 bytes by default, preventing large text columns from slowing down or blocking ANALYZE. The task still completes, and the skip reason is visible in the analysis results.

• Iceberg REST Catalog and S3Tables Catalog can now access AWS resources through IAM roles, temporary credentials, or credentials providers, reducing the need to manually maintain long-term access keys.

• OBS parallel file systems can now be accessed through the native OBS file system, preventing external table or object storage read/write failures in scenarios not supported by the S3-compatible protocol.

• Routine Load now supports per-task observability of consumed rows, error rows, throughput, progress, lag, and abort counts, making it easier to identify latency or anomalies in individual load tasks.

• Added metrics for asynchronous materialized views and query rewrite, enabling observation of MV refresh task status and MV rewrite duration during the query planning phase.

• The MTMV refresh task list now displays which compute group a task is running on, making it easier to troubleshoot resource group selection and refresh failures.

• Automatic partitioning and dynamic partitioning now allow more partitions by default and expose warning metrics when the partition count exceeds 80% of the limit, helping you detect approaching partition limits earlier.

[Improvements]

• Auto-bucketing tables now create a minimum of 3 buckets by default, mitigating the insufficient parallelism caused by having only 1 bucket for small tables or underestimated estimates.

• The Cloud write path now reduces the number of empty rowset metadata entries and object storage files by default, lowering metadata and object storage overhead after high-frequency small-batch writes.

• File Cache now reads cached files directly by default and adds capacity limits for the background LRU update queue and shadow queue, reducing cache hit read path overhead and preventing excessive memory usage from LRU events.

• After configuring File Cache TTL, subsequent writes no longer refresh the cache expiration time for the entire tablet, making the cache lifecycle more consistent with table or tablet creation time semantics.

• When the same table warm-up task is submitted repeatedly, duplicate pending jobs are now reused or rejected. Asynchronous rowset warmup no longer blocks queries or subsequent scheduling, and warm-up metrics are more accurate.

• Routine Load using Kafka read_committed now reduces idle retries when encountering uncommitted transactions, and lag reporting is more timely.

• Stream Load Group Commit and Broker Load write paths now reduce unnecessary initialization and parameter passing, lowering import overhead and preventing unsupported scenarios from incorrectly using sink node memtables.

• BE memory usage and rowset metadata size are reduced in high-concurrency import, row-store write, Cloud packed file write, and large rowset metadata scenarios.

• When object storage returns SlowDown rate-limiting errors, the system now fails fast by default, preventing excessive automatic retries from amplifying object storage pressure.

• During query troubleshooting, scan nodes, query summaries, and internal query failure states can now be more easily correlated through plan, profile, and audit logs.

• Prometheus histogram metric labels are no longer lost, allowing related latency metrics to be correctly aggregated and alerted by their original label dimensions.

• DataWorks lineage logs are now automatically cleaned based on configuration, preventing log files from accumulating over long-running periods.

[Fixes]

• Fixed an issue where residual partition metadata after a failed new partition creation on Cloud tables caused subsequent partition creation, query, or metadata sync anomalies.

• Fixed an issue where high-concurrency transaction commits or high-concurrency ADD/DROP PARTITION on Cloud tables were more likely to encounter KV_TXN_CONFLICT.

• Fixed an issue where Cloud recycling of dropped tablets containing empty segment rowsets got stuck, preventing tablet/index metadata from being cleaned up for extended periods.

• Fixed an issue where Cloud recycling deleted tablet KV entries even when some objects failed to delete, potentially leaving orphaned objects or preventing correct retry of subsequent recycling.

• Fixed an issue where version gaps or legacy compaction rowsets after Cloud schema change affected queries, delete bitmaps, or subsequent compaction.

• Fixed an issue where low-write tablets could not trigger time-series compaction by time threshold for extended periods after BE restart or migration.

• Fixed issues with File Cache cleanup during restart, direct cleanup, or abnormal directory states that caused statistical anomalies, residual cache items, or cleanup failures. The HTTP cleanup interface is now uniformly executed asynchronously.

• Fixed an issue where slow or chunked Stream Load clients encountered broken pipe during FE redirect, and prevented redirect URL rewriting from causing path or parameter anomalies with special characters.

• Fixed an issue where Stream Load or BE selection logic could not select the correct BE or caused import failures when BE ID differences were large.

• Fixed an issue where transaction writes in Cloud multi-compute group scenarios were routed to incorrect backends, causing INSERT or transaction import failures.

• Fixed an issue where parallel exports caused file deletion or export failures when multiple BEs cleaned the target path simultaneously. Local file system exports now reject unsupported delete_existing_files.

• Fixed an issue where writes failed immediately after Insert Overwrite auto-created replacement partitions because the partitions were not yet routable.

• Fixed an issue where row and byte statistics were duplicated during automatic partition table imports, causing metrics or displayed values to be inflated.

• Fixed an issue where queries containing outer join, mark anti join, pre-aggregation join, or random/unique functions could return incorrect results after specific rewrites.

• Fixed an issue where combined queries involving TopN, aggregation, datetime predicates, alias functions, user variables, and null checks produced incorrect results or parsing failures in specific scenarios.

• Fixed an issue where materialized view rewrite could select incorrect partitions or return incorrect results in outer join or partition compensation scenarios.

• Fixed an issue where view definitions were incorrectly rewritten when creating or querying views containing aggregate aliases, star expansion, or Variant subfields, causing query failures or incorrect results.

• Fixed result errors or unsupported semantics being accepted in point queries, multi-layer projections, IPv4/IPv6 literal comparisons, and complex type literal comparisons.

• Fixed an issue where JSONB functions could cause BE crashes when encountering invalid input or unknown types.

• Fixed an issue where Variant content was lost after row-store partial updates, and where reading Variant subfields with dots written by older versions failed after upgrade.

• Fixed an issue where Variant duplicate JSON path behavior was unstable, and where JSONB/Variant was incorrectly used as hash distribution columns, causing subsequent write or query anomalies.

• Fixed an issue where TLS/mTLS configuration required a CA, where PEM multi-certificate chain loading was incomplete, or where SNI host check failed when accessing HTTPS via IP.

• Fixed an issue where queries using CTE could bypass Ranger column-level permission checks.

• Fixed an issue where LDAP authentication failed when usernames or passwords contained special characters, and where LDAP connection exceptions could block new connections.

• Fixed an issue where Arrow Flight SQL could not obtain the real remote IP for client IP-based authentication or access control.

• Fixed an issue where queries or INSERT INTO ... SELECT executed via SU/proxy and forwarded to master obtained {location=invalid} when the SU target user lacked resource_tags.location or UserProperty, causing default tag replicas to be filtered and reporting tablet ... has no queryable replicas.

• Fixed an issue where skip_catalog_priv_check had an overly broad scope, causing some catalog operations to bypass permission checks.

• Fixed an issue where the KeyCache background thread accessed freed resources during BE shutdown or restart in TDE scenarios, causing crashes.

• Fixed an issue where connection pool reuse of MySQL connections left residual user variables, temporary tables, prepared statements, or transaction states.

• Fixed an issue where SQL parsing failed when table names, aliases, or CTE names used "su".

• In Cloud mode, executing the unsupported ADMIN SET REPLICA VERSION now directly returns an unsupported error, preventing incorrect modification of replica version information that does not apply to Cloud.

• Fixed an issue where SHOW TABLET failed when USE db was not executed, and where SHOW BACKENDS result columns were misaligned with headers.

• Fixed an issue where queries could still select replicas marked for deletion after the user deleted them, causing query failures or instability.

• Fixed an issue where Iceberg RestCatalog could not use temporary credentials when accessing S3-compatible storage such as OSS, OBS, COS, or MinIO.

• Fixed an issue where schema retrieval failed with newer SQLServer JDBC drivers, causing external catalog query anomalies.

• Fixed an issue where Hive external table row count cache was incomplete, and where external file systems that were cache-evicted were still used, causing query or transaction failures.

• Fixed an issue where string or binary column decoding was incorrect, or where group reader failed when reading some Parquet external files.

• Fixed an issue where concurrent scans of the same tablet during lightweight schema change or rowset state changes could cause BE crashes.

• Fixed an issue where INSERT error message formatting was abnormal, and where FE exceptions occurred when Routine Load retrieved Kafka metadata with null values.

• The select @@version return value no longer contains the "selectdb" or "velodb" strings.

Apache Doris 3.1.4

4.1.7

May 26, 2026

[Fixes]

• Cache and cache warm-up: Improves the stability of cache warm-up and remote Peer Read, reducing the impact of scaling, hot/cold data tiering, and remote data reads on import and query performance. Enhances observability of the warm-up progress.

• File cache: Enhances the stability of the file cache during cleanup, error retries, and partial hit scenarios. This reduces the risk of cache pollution, error amplification, BE crashes, and cache space expansion.

• Compaction: Enhances observability and troubleshooting for compaction tasks. Improves the retry mechanism after conflicts and increases stability in schema change and Merge-on-Write (MOW) scenarios.

• Data writing: Improves the stability of write operations such as Stream Load, Broker Load, Group Commit, and INSERT/DELETE. Enhances the reliability of task status after error recovery and restarts.

• Write resource control: Reduces peak memory usage during large-volume writes and high-concurrency flush scenarios, enhancing write stability and resource control.

• Query correctness: Fixes multiple query accuracy issues, preventing incorrect results or errors in scenarios involving complex SQL, set operations, short-circuit queries, and function calculations.

• External data sources and object storage: Improves compatibility for connecting to object storage and external data sources. Enhances configuration adaptation and read stability for scenarios involving S3, OSS, Azure, JDBC, Hive, ORC, and CSV.

• Security and authentication: Improves cloud-based certificate authentication, encrypted communication, permission control, and license management. Enhances security configuration usability and reduces the risk of sensitive information exposure.

• Cloud metadata and transactions: Enhances the stability of cloud metadata, transactions, and version synchronization. Improves the reliability of transaction aborts, version reporting, issue diagnosis, and inter-service request tracing.

• Backup, recovery, and snapshots: Improves the stability of backup, recovery, and snapshot processes. Reduces view recovery errors and snapshot generation failures.

• DDL and metadata changes: Improves stability in metadata change scenarios, including view alterations, automatic partitioning, and materialized view creation.

• Observability and diagnostics: Enhances diagnostic capabilities for scenarios involving cache warm-up, compaction, Routine Load, compute groups, BDBJE, and Kubernetes deployments, providing more complete and intuitive information for troubleshooting online issues.

Apache Doris 3.1.4

4.1.6

May 9, 2026

[Fixes]

• Improves the stability of packed file uploads to prevent data write failures caused by exceptions during writes to object storage.

Apache Doris 3.1.4

4.1.5

April 20, 2026

[Fixes]

• Enhanced cross-compute group read cache capability: On a cache hit, cross-group read requests can bypass remote object storage access, which reduces network overhead from origin-pulls. This version also achieves production-grade stability for the Colocate table Peer Read mechanism, improving overall query latency and cross-node bandwidth utilization.

• Improved Profile/bvar metrics system: Adds Peer Read hit rate, cross-group routing topology, and latency breakdown dimensions for precise performance bottleneck identification. Exposes direct query and dynamic intervention interfaces for Peer Cache candidate node status, improving cache topology management.

• Optimized file cache concurrency model: Enhances state machine stability under heavy query loads and high-concurrency scenarios, suppresses resource contention, and reduces abnormal state triggers and performance jitter.

• Hardened BE node fault tolerance: Eliminates potential memory leak and thread deadlock risks under high-concurrency file cache access and abnormal boundary conditions, which reduces the risk of unexpected backend process crashes.

• Fine-grained read-side cache lifecycle management: Introduces per-job dynamic cache retention policies for INSERT INTO SELECT tasks, preventing persistent cache pollution from long-tail queries.

• Improved large-scale sorting performance: Optimizes sort engine memory scheduling logic and adds a session-level variable to control full sort buffer to dynamically adjust full sort memory thresholds based on workload characteristics.

• Accelerated schema change version reclamation: Optimizes the metadata version cleanup mechanism, shortens invalid version retention periods, mitigates version accumulation, and reduces I/O and compute overhead in subsequent compaction stages.

• Converged INSERT OVERWRITE cleanup logic: Adopts a deterministic partition path overwrite strategy to eliminate legacy path discrepancies and ensure strong consistency between metadata and physical storage state.

• Strengthened FE communication and transaction boundary fault tolerance: Optimizes inter-node RPC timeout backoff and retry mechanisms, and unifies error code semantics for cloud-native distributed transaction exception scenarios, making client interactions more predictable and retry-friendly.

• Implemented tiered log noise reduction: Filters redundant debug information and optimizes critical event aggregation logic, ensuring log output during troubleshooting focuses precisely on core exception chains and system state transitions.

• Adjusted Tablet scheduling policy baseline: Disables the "active Tablet priority scheduling" feature by default after upgrades, shifting system load balancing to a conservative and stable mode and reducing hotspot skew risk under extreme loads.

• Optimized Recycler lightweight scheduling: Reduces the default execution weight and metadata scan granularity of the background recycler, which alleviates lock contention and memory overhead at the metadata layer in high-watermark clusters.

Apache Doris 3.1.4

4.1.4

April 15, 2026

[Fixes]

• Enhances cloud compaction throughput and wide-table organization capabilities, with more conservative default policies to reduce additional jitter from new features after an upgrade.

• Improves stability during scaling, hot/cold tiering, and cache warm-up phases, with more conservative balancing logic to reduce hotspot scheduling anomalies and task accumulation risks.

• Improves the robustness of file cache hit-and-recovery logic, reducing cache pollution, missed retries, and error amplification in query and import scenarios.

• Reduces the risk of stuck transactions, version inconsistencies, and upgrade interruptions, improving upgrade stability.

• Reduces memory pressure in large-scale clusters, with safer recovery paths to lower the risk of OOM errors, accidental data deletion, or increased FDB pressure from recycling in exceptional cases.

• Improves availability with special network, certificate, authentication, and endpoint configurations for Azure/S3, reducing external storage connection issues.

• Improves stability for connections to Paimon, Hudi, Iceberg, and Ozone, reducing external catalog hangs, read errors, and compatibility issues.

• Improves the accuracy of various query plans and function results. Fixes result anomalies in some set operations, Nereids optimizations, and time zone-related queries.

• Enhances stability for import, overwrite, schema change, and Routine Load scenarios, reducing write failures and unintended task cancellations.

• Improves SQL compatibility and permission checking accuracy, enabling more stable connections through proxies or in scenarios with complex request headers.

• Adds DataWorks lineage integration and fixes compatibility issues with the 3.1 branch, reducing plugin integration risks.

• Allows configurable Pinyin tokenization for Chinese search scenarios, providing more flexible search capabilities.

• You can now directly use KMS to manage transparent encryption master keys in Alibaba Cloud environments.

• Enhances stability for semi-structured data scenarios using Variant/Struct types, supporting more sub-columns and reducing compaction and type conversion anomalies.

• Covers more global time zone scenarios and improves export format flexibility.

• Enhances the observability of cloud transactions, recycling, Routine Load, and interactions with the meta service.

Apache Doris 3.1.4

4.1.3

March 11, 2026

[New features]

Supports all features of Apache Doris 3.1.4.

[Fixes]

• Adds a small file merging feature that reduces object storage API calls, lowering storage costs and improving import performance. This feature is disabled by default.

• Improves transaction commit performance in high-concurrency import scenarios, reducing import latency.

• Fixes multiple issues that could cause query failures, BE crashes, or memory leaks.

• Improves metadata query performance, optimizes the cluster startup and upgrade experience, and enables automatic job recovery after compute group switchovers.

• Supports multi-version FDB clients to improve deployment compatibility.

• Fixes multiple file cache crashes and capacity inaccuracies to improve cache stability.

• Fixes a risk of accidental data deletion by the data recycler, improving recycling security.

• Fixes several import issues, including partial updates in Routine Load, path columns in Broker Load, and format-based imports with Copy Into.

• Fixes multiple optimizer issues that could cause query hangs, incorrect results, or execution failures.

• Fixes issues related to materialized view rewrite selection. Disables materialized views with conditions on the value column of Merge-on-Write (MOW) tables to prevent data errors.

• Supports more flexible AWS authentication methods. Fixes issues with hanging external table queries and connections to S3-compatible storage such as MinIO and COS.

• Fixes partition and cache-related issues for Iceberg, Hive, and Paimon. Improves Hive partition pruning performance.

• Fixes a memory leak in JDBC catalog. Adds support for the Elasticsearch (ES) flatten type.

• Optimizes import memory usage with automatic backpressure in high-memory scenarios to prevent OOM errors.

• Fixes data correctness issues related to compaction and schema change.

• Enhances storage credential security to prevent sensitive information leaks and Server-Side Request Forgery (SSRF) attacks.

• Fixes issues with SQL parsing compatibility, SQL blacklist failure on restart, missing monitoring metrics, and data type compatibility.

• Adds the array_flatten and QUANTILE_STATE_TO/FROM_BASE64 functions.

• Audit Log now supports SQL fingerprints and encryption for easier analysis and security compliance.

• Improves JDK 17 compatibility.

Apache Doris 3.1.4

4.1.2

January 21, 2026

[Fixes]

• Optimizes view creation logic in specific scenarios to improve the SQL user experience.

• Optimizes the temporary resource management mechanism to improve long-term system stability.

• Optimizes the cache capacity management strategy to improve resource utilization efficiency.

• Optimizes the execution performance and stability of Routine Load in update scenarios.

• Optimizes the transaction recovery process to enhance system robustness.

• Adds transaction-related monitoring metrics to enhance observability.

• Optimizes the FE service encoding configuration to improve system compatibility and stability.

• Optimizes task scheduling and resource recovery efficiency in cloud environments.

Apache Doris 3.1.3

4.1.1

December 17, 2025

[New features]

Supports all features of Apache Doris 3.1.3.

Apache Doris 3.1.3

4.1.0-beta

November 1, 2025

[New features]

Supports all features of Apache Doris 3.1.1.

Apache Doris 3.1.1

Version

Release date

Description

References

4.0.11

March 25, 2026

[Enhancements and fixes]

• Improves stability when querying CSV files with quoted headers by using S3 TVFs, preventing field parsing errors.

• Enhances S3 TVF compatibility with more CSV escape formats, reducing preprocessing costs.

• Increases the success rate of importing compressed JSON data and prevents Stream Load failures.

• Improves the stability of exporting data to OSS, preventing backup and export failures caused by authentication errors.

• Ensures that semantically identical SQL queries return consistent results, which reduces the risk of misjudgment in production environments.

• Fixes errors in some date and time-related SQL queries.

• Improves cluster stability in high-concurrency and complex import scenarios, reducing BE crashes.

• Reduces the risk of batch query timeouts in long-running clusters.

• Fixes an issue where some complex queries triggered BE crashes, which improves query stability.

• Improves the stability of queries that involve remote reads and the file cache and reduces BE core dumps.

• Improves FE stability under high traffic, which allows it to support larger business workloads.

• Prevents the unknown inverted index format 0 error during queries and improves query availability.

• Improves stability in complex write scenarios, such as INSERT OVERWRITE.

• Reduces the risk of abnormal FE restarts and complete service outages.

• Allows TVF queries on system libraries to automatically wake up the cluster, which reduces no available backends errors.

• Increases the success rate of instance type changes and scale-ins and reduces the time that operations remain stuck.

• Accelerates the rebalancing process after a scale-out, which enables faster business recovery.

• Clarifies the meaning of monitoring metrics to improve the assessment of remote scanning workloads.

• Optimizes the out-of-the-box default configuration for high-concurrency and import scenarios in production environments.

Apache Doris 3.0.8

4.0.10

January 11, 2026

[Fixes]

• Fixes an issue where incorrect URL information was truncated during an INSERT import.

• Fixes a GROUPING_PREFIX error that occurred in specific aggregate query scenarios.

• Fixes occasional failures in insert overwrite operations caused by rowset conflicts.

• Fixes an issue where materialized view creation failed.

• Fixes an execution plan error that occurred because a Java User-Defined Function (UDF) prevented consecutive project operations from merging.

• Fixes an error in the pipeline parallelism calculation during TVF import.

• Fixes abnormal memory usage during UDF execution.

• Fixes the unknown load_id error during the import process.

• Optimizes the tablet rebalance scheduling mechanism by enabling dynamic adjustment of the scheduling period, which improves efficiency.

• Fixes data skew after a version upgrade.

• Fixes a window function matching error in partition top-N scenarios.

• Fixes an issue where the parameter constant check for lead/lag functions was too strict in certain scenarios.

• Fixes an issue where Routine Load jobs reported errors after an upgrade.

• Fixes an issue where the behavior of the JSON_EXTRACT function changed after an upgrade.

• Fixes an issue with predicate pushdown optimization in scenarios that involve multiple left joins.

• Fixes an issue where the start value of an AUTO INCREMENT column was incorrect.

• Fixes a BE crash caused by column type conversion.

• Optimizes the predicate pushdown strategy for random sampling scenarios.

• Fixes an issue where JSON_EXTRACT returned null when processing an array.

• Fixes an issue where the substring function failed in specific scenarios.

• Fixes an inference error in join reordering that caused query failures.

• Fixes an issue with detecting ambiguous column names in multi-table joins.

• Improves the accuracy of error messages for failed imports.

• Fixes an issue where an overwrite operation performed by using a MaxCompute catalog caused data loss and adds debug logs.

• Fixes a bug that caused the observer to stop when a get version call failed during meta service rate limiting.

• Fixes a null pointer exception that occurred during Routine Load task assignment.

• Fixes BE stability issues.

• Fixes a deadlock in BE JVM metrics collection.

• Fixes an issue where a DLF-based Paimon catalog failed to read from OSS HDFS.

• Fixes an issue where the max_tablet_size log printed a table ID instead of a tablet ID.

Apache Doris 3.0.8

4.0.9

November 25, 2025

[Fixes]

• Fixes an issue where UDFs occasionally reported a NoClassDefFoundError.

• Fixes an issue where queries on Iceberg external tables without partition columns returned incorrect results.

Apache Doris 3.0.8

4.0.8

October 27, 2025

[Fixes]

• Fixes an issue where INSERT jobs failed when they queried an Elasticsearch catalog.

• Fixes an issue where an invalid JSON path caused a core dump.

Apache Doris 3.0.8

4.0.7

September 26, 2025

[New features]

Supports all features of Apache Doris 3.0.8.

Apache Doris 3.0.8

4.0.6

September 15, 2025

[New features]

Supports all features of Apache Doris 3.0.6.

[Enhancements]

• Optimizes the startup process for file cache upgrades to accelerate the upgrade from version 3.0.x to 4.0.x.

• Improves query performance and availability in multi-cluster scenarios.

• Adds status checks during rowset preparation and commit to improve availability.

• Improves the performance of top-N queries.

[Fixes]

Fixes a query performance issue caused by premature tablet cache cleanup.

Apache Doris 3.0.6

4.0.5

July 14, 2025

[New features]

Supports all features of Apache Doris 3.0.4 and Apache Doris 3.0.5.

[Enhancements]

Improves availability by resolving an issue of premature data recycling.

[Fixes]

Fixes an inverted index core dump issue that occurred when you upgrade from version 4.0.4 to 4.0.5.

• Apache Doris 3.0.4

• Apache Doris 3.0.5

4.0.4

December 6, 2024

[New features]

• Supports all features of Apache Doris 3.0.3.

• Supports arrays of the text field type in JDBC catalog.

• Supports the timestamp field type in MaxCompute catalog.

[Enhancements]

• Improves query performance by resolving an issue of high CPU consumption during the segment loading process.

• Reduces fluctuations in query response times by optimizing the file cache policy.

• Optimizes the file cache recycling mechanism.

• Improves information_schema query performance.

• Optimizes the metadata lock mechanism.

• Optimizes the delete bitmap lock mechanism.

• Improves memory management for Compressed Class Space.

• Enhances security by removing an interface that posed a Server-Side Request Forgery (SSRF) risk.

• Improves the performance of Audit Log statistics.

• Improves the performance of window functions in paginated queries.

• Improves the performance of time-series compaction for Merge-on-Write (MOW) tables.

[Fixes]

• Fixes an issue where modifying the disable_auto_compaction property of a MOW table could make table data unavailable.

• Fixes an issue where imports to MOW tables failed during peak hours.

• Fixes several Routine Load issues, including data loss, job stalls, and transaction commit failures.

• Fixes an issue where Routine Load jobs were paused and reported errors after the instance access module was restarted.

• Fixes an issue where Group Commit could fail.

• Fixes an issue where queries scanned more data after an instance was upgraded to kernel version 4.0.4.

• Fixes an issue where point queries that used precompilation did not return a result set.

• Fixes an issue where duplicate values could appear in the primary key field of a table that used the Unique model.

• Fixed an issue where reclaiming stale rowset caused cache misses, which increased query response time.

• Fixes an inconsistency between the results from Change Data Capture (CDC) partial column updates and from Stream Load wide table joins.

• Fixes an issue that caused core dumps on multiple cluster nodes.

• Fixes a metadata exception that occurred when an asynchronous materialized view read data from a table across clusters.

• Fixes an issue with data quality validation when you write to an Auto Partition table by using insert overwrite.

• Fixes an issue where insert overwrite caused a write error.

• Fixes an issue that incorrectly blocked insert overwrite operations on the same table.

• Fixes an issue where an error URL was not provided for insert into select operations that failed due to data quality issues.

• Fixed an issue where a stmtId overflow caused a Prepared Statement exception in high-concurrency point query scenarios.

• Fixes a RUNTIME_ERROR caused by an uninitialized state when you execute the ConvertTz function after you upgrade to version 4.0.4.

• Fixes an issue that caused slow cluster restarts.

• Fixes an issue where the NOT operator did not take effect in short-word matching.

• Fixes an issue with compaction recovery after a schema change.

• Fixes a data import error and a version limit exceeded issue caused by empty partitions after a schema change.

• Fixes a JDBC connection leak issue.

• Fixes a PostgreSQL (PG) catalog connection leak issue.

• Fixes an issue where partition pruning for MaxCompute catalog did not take effect.

• Fixes an issue where the local file cache size dropped sharply due to premature eviction of cached data.

• Fixes an issue where SQL Cache caused incorrect results.

• Fixes a calculation error in the split_part function.

• Fixes a compatibility issue with field naming rules.

• Fixes a query latency issue caused by multiple threads waiting for a spinlock.

• Fixes an issue where cluster nodes went offline abnormally due to transient network jitters.

• Fixes an issue where queries briefly reported a COMPUTE_GROUPS_NO_ALIVE_BE error.

• Fixes an issue where the window_funnel function did not support high-precision timestamp columns.

• Fixes a query planning timeout issue.

• Fixes a core dump issue in explode_variant_array.

• Fixes an issue where queries missed the materialized view.

• Fixes an issue where error log URLs were too long to be accessed.

Apache Doris 3.0.3

4.0.3

October 11, 2024

[New features]

• Supports all features of Apache Doris 3.0.2.

• Supports large transaction imports. This feature is disabled by default.

[Enhancements]

• Disables java_udf by default.

• Optimizes the top n query performance in logging scenarios.

• Improves point query performance through short-circuit evaluation.

• Improves tablet pruning to enhance query performance.

[Fixes]

• Fixes an issue where replaying the catalog could cause a backend instance to fail to start.

• Fixes an issue where the Audit Log component failed after an upgrade from SelectDB Core 3.0.x to 4.0.x.

• Fixes an issue where default compute cluster permissions were restricted after an upgrade to SelectDB Core 4.0.2.

• Fixes an issue where constant folding that uses a Remote Procedure Call (RPC) might cross compute clusters.

• Fixes an issue where a compute cluster could not be found when an SQL statement is forwarded.

• Fixes a core dump issue that is caused by a null pointer in the compaction delete bitmap.

• Fixes an issue where a full compaction could not be triggered for a MOW table when its delete bitmap was too large.

• Fixes an issue where materialized views defined by select * failed to build.

Apache Doris 3.0.2

4.0.2-beta

August 25, 2024

[New features]

• Supports all features of Apache Doris 3.0.1.

[Fixes]

• Fixes an issue where import performance could degrade after a large number of import failures when the same label was used for imports.

• Fixes an issue where a cluster scale-out could cause an instance version upgrade to get stuck.

• Fixes a memory leak issue with the runtime filter.

Apache Doris 3.0.1

4.0.1-beta

July 9, 2024

[New features]

• Supports upgrading SelectDB Core from version 3.0.x (x >= 9) to 4.0.x.

• Supports all features of Apache Doris 3.0.0.

[Fixes]

• Fixes an issue where a backend instance restart was required to restore cluster permissions after an upgrade from SelectDB Core 3.0.9.

• Fixes an issue where inverted index queries could get stuck.

Apache Doris 3.0.0

4.0.0-beta

May 15, 2024

[New features]

• Supports multi-table materialized views (MTMVs), which provide transparent rewriting, automatic refresh, materialization from external to internal tables, and direct queries. Based on this capability, MTMVs can now be used for data warehouse layering, job scheduling, and data processing.

• Introduces the Variant and IP data types and enhances a range of analytical functions to simplify the storage and analysis of complex and semi-structured data.

• Supports auto-incrementing columns, automatic partitioning, and server-side batch request submission (Group Commit), improving efficiency for large-scale and real-time data writes.

• Introduces a high-speed data read interface based on Arrow Flight, which increases data transfer efficiency by up to 100 times. This feature is ideal for data science and other large-scale data read scenarios.

• Supports all features of Apache Doris 2.1.0.

[Enhancements]

• Significantly improves out-of-the-box query performance and delivers good performance without manual tuning. This version also further improves the performance of complex SQL queries, achieving a performance increase of over 100% on the TPC-DS 1 TB benchmark and delivering industry-leading query speeds.

• Significantly improves data lake analytics performance, delivering 4 to 6 times the performance of Trino and Spark. This version also introduces compatibility with multiple SQL dialects, enabling seamless migration from your existing systems to ApsaraDB for SelectDB.

Apache Doris 2.1.0