MatrixOne-v4.2.0

Release date: August 21, 2026
MatrixOne version: v4.2.0

MatrixOne 4.2.0 is a feature and reliability release on top of 4.1. It expands SQL compatibility and external data access, adds native data movement and recovery tooling, improves vector and distributed execution, and hardens transaction visibility, locking, backup, fileservice, bootstrap, and query execution correctness.

Key Improvements

SQL & Compatibility

• DML and Procedural SQL: Add DML RETURNING and PERFORM SELECT support.
• Constraints and Locking Reads: Persist and enforce CHECK constraints and support shared locking reads.
• Recursive CTE and Functions: Improve recursive CTE semantics and memory bounds; add approximate percentile, ONNX inference, JSON, date/time, and compatibility functions.

External Data, Backup & Recovery

• External Connectors: Add Iceberg and MongoDB external table scans.
• Native Data Movement: Add native DUMP TABLE and LOAD TABLE support.
• Offline Recovery Tooling: Add offline checkpoint catalog inspection, CSV export and restore-script generation, plus offline LogService WAL recovery.

Execution, Vector & Data Branch

• Execution Engine: Introduce a unified spill engine and unified shuffle paths, with staged query placement and improved memory accounting.
• Vector Indexes: Add IVF INCLUDE columns and bf16, float16, int8 and uint8 quantization support for IVFFLAT, IVFPQ and CAGRA.
• Data Branch: Materialize diff output as a table and support schema evolution during diff and merge.

Key Bug Fixes

Transaction, Locking & Visibility

• Preserve remote snapshot and logtail visibility across retries and cancellation.
• Exclude uncertain cleanup-only locks from deadlock graphs and enforce consistent base/index lock ordering for irregular-index UPDATEs.
• Fence primary-key checks and table reads on subscription readiness and pending logtail application.

Storage, Backup & Fileservice

• Preserve hidden commit timestamps in rewritten backup tombstones and harden checkpoint/object-copy failure handling.
• Fix disk-cache ownership and cache-key consistency, validate remote cache payloads, honor cancellation while waiting for object-storage slots, and prevent duplicate S3 range fan-out.
• Preserve block filters and zonemap pruning, and defer wide scan columns until after filtering.

Bootstrap & Cluster Reliability

• Gate CN startup on self-visible cluster inventory.
• Harden tenant view upgrades and recreate missing INFORMATION_SCHEMA character-set metadata.
• Bound full-text top-K memory and improve cancellation/resource cleanup across spill and remote execution paths.

What's Changed

• feat: support DML RETURNING by @VioletQwQ-0 in #26613
• feat(sql): persist and enforce CHECK constraints for INSERT by @ck89119 in #26382
• feat(iceberg): add external Iceberg connector by @iamlinjunhong in #25368
• feat: add MongoDB external table scan by @iamlinjunhong in #26424
• feature: support native DUMP TABLE and LOAD TABLE by @aptend in #25989
• feat: unified spill engine across hashjoin, dedupjoin and rightdedupjoin by @aunjgr in #25488
• feat(data-branch): materialize diff output as table by @gouhongshen in #25863
• fix(backup): preserve hidden commitTS in rewritten tombstones by @LeftHandCold in #27003
• fix(lockservice): exclude uncertain locks from deadlock graph by @LeftHandCold in #27129
• fix(plan): lock base rows before irregular update fanout by @LeftHandCold in #27222
• fix(cnservice): gate startup on self-visible cluster inventory by @LeftHandCold in #27167
• fix: harden tenant view upgrades by @aptend in #27314
• fix: recreate missing character sets table by @aptend in #27397
• fix: preserve zonemap pruning for DATETIME/TIMESTAMP comparisons by @aptend in #27097
• perf: defer wide scan columns until after filtering by @aptend in #27099
• fix(fileservice): prevent duplicate range fan-out by @LeftHandCold in #26842

Full Changelog: v4.1.4...v4.2.0