v1.0.0-beta.3

Release date: December 17, 2025

It's recommended to upgrade to v1.0.0-beta.4.

Breaking changes

• Cache Architecture Refactoring: Replaced generic LRU cache with specialized manifest cache and background downloads by @evenyag in #7294, breaking configuration changes:
  • NEW: region_engine.mito.manifest_cache_size (default: 256MB)
  • REMOVED: storage.cache_path, storage.enable_read_cache, storage.cache_capacity

👍 Highlights

🚀 Key New Features

• Vector Index Support:
  • Added foundational types and SQL parsing support for vector index functionality.
• Memory Management:
  • Implemented memory manager for compaction operations.
  • Introduced per file scan metrics for better monitoring.
• Operations & Monitoring:
  • Added ability to suspend frontend and datanode components.
  • Added Grafana PostgreSQL data source query builder support.
  • Implemented manifest file cache for improved performance.

⚡ Performance Improvements

• Optimized DISTINCT operations by treating them as commutative/partially-commutative.
• Enhanced histogram quantile calculations with safe mode for incomplete data.
• Improved part sort behavior on overlapping time windows.
• Added verbose metrics to scanners for better performance tracking.

🐛 Notable Bug Fixes

• Fixed network failure detection with HTTP/2 keep-alive for heartbeat client.
• Resolved PostgreSQL extended query regression with shortcutted statements.
• Fixed PromQL histogram issues with aggregation and offset direction.
• Corrected TLS option validation and merging problems.
• Fixed CPU core calculation that could incorrectly return 0.

🚀 Features

• feat: implement a cache for manifest files by @evenyag in #7326
• feat: add more verbose metrics to scanners by @evenyag in #7336
• feat: decode pk eagerly by @waynexia in #7350
• feat: run histogram quantile in safe mode for incomplete data by @waynexia in #7297
• feat: update dashboard to v0.11.9 by @ZonaHex in #7364
• feat: allow publishing new nightly release when some platforms are absent by @waynexia in #7354
• feat: allow one to many VRL pipeline by @v0y4g3r in #7342
• feat: collect merge and dedup metrics by @evenyag in #7375
• feat: suspend frontend and datanode by @MichaelScofield in #7370
• feat: table/column/flow COMMENT by @waynexia in #7060
• feat: grafana postgresql data source query builder support by @sunng87 in #7379
• feat: mark index outdated by @discord9 in #7383
• feat: mem manager on compaction by @fengjiachun in #7305
• feat: optimize and fix part sort on overlapping time windows by @waynexia in #7387
• feat: per file scan metrics by @evenyag in #7396
• feat: move memory_manager to common crate by @fengjiachun in #7408
• feat: introduce copy_region_from for mito engine by @WenyXu in #7389
• feat: support function aliases and add MySQL-compatible aliases by @killme2008 in #7410
• feat(vector_index): adds the foundational types and SQL parsing support for vector index by @killme2008 in #7366

🐛 Bug Fixes

• fix: reset cached channel on errors with VIP by @fengjiachun in #7335
• fix: configure HTTP/2 keep-alive for heartbeat client to detect network failures faster by @WenyXu in #7344
• fix: regression with shortcutted statement on postgres extended query by @sunng87 in #7340
• fix: use saturating in gc tracker by @discord9 in #7369
• fix: part sort behavior by @waynexia in #7374
• fix: improve network failure detection by @WenyXu in #7382
• fix(procedure): update procedure state correctly during execution and on failure by @WenyXu in #7376
• fix: gc listing op first by @discord9 in #7385
• fix: parse "KEEP FIRING FOR" by @fengys1996 in #7386
• fix: promql histogram with aggregation by @waynexia in #7393
• fix: promql offset direction by @waynexia in #7392
• fix: TLS option validate and merge by @shuiyisong in #7401
• fix: cpu cores got wrongly calculated to 0 by @MichaelScofield in #7405
• fix: use verified recycling method for PostgreSQL connection pool by @WenyXu in #7407
• fix(servers): flight stuck on waiting for first message by @v0y4g3r in #7413
• fix: using anonymous s3 access when ak and sk is not provided by @shuiyisong in #7425

🚜 Refactor

• refactor(servers): allow custom flight service by @v0y4g3r in #7333
• refactor(servers): bulk insert service by @v0y4g3r in #7329
• refactor: remove datafusion data frame wrapper by @waynexia in #7347
• refactor: use versioned index for index file by @discord9 in #7309
• refactor: extract file watcher to common-config by @shuiyisong in #7357
• refactor: remove duplication coverage and code from window sort tests by @waynexia in #7384
• refactor: optimize heartbeat channel and etcd client keepalive settings by @WenyXu in #7390

⚡ Performance

• perf: treat DISTINCT as comm/part-comm by @waynexia in #7348

🧪 Testing

• test: gc worker scheduler mock test by @discord9 in #7292
• test: test_tracker_cleanup skip non linux by @discord9 in #7398

⚙️ Miscellaneous Tasks

• chore: members and committers update by @fengjiachun in #7341
• chore: rm unnecessary warning by @discord9 in #7352
• chore: rename to avoid git reserved name by @discord9 in #7359
• chore(mito): expose some symbols by @v0y4g3r in #7373
• chore: saturating duration since by @discord9 in #7380
• chore(mito): nit avoid clone the batch object on inverted index building by @lyang24 in #7388
• chore: sort histogram sqlness result by @waynexia in #7406
• chore: add is_initialized method for frontend client by @fengys1996 in #7409
• chore: expose symbols by @v0y4g3r in #7417
• chore: change etcd endpoints to array in the test scripts by @daviderli614 in #7419
• chore: bump version to beta.3 by @discord9 in #7423
• chore: feature gate vector_index by @discord9 in #7428

All Contributors

We would like to thank the following contributors from the GreptimeDB community:

@MichaelScofield, @WenyXu, @ZonaHex, @daviderli614, @discord9, @evenyag, @fengjiachun, @fengys1996, @killme2008, @lyang24, @shuiyisong, @sunng87, @v0y4g3r, @waynexia