Recently, we released GreptimeDB v1.0.0-rc.1, introducing region repartition and other key capabilities, marking the feature freeze and the start of the stability validation phase toward v1.0 GA.

This week, we're releasing v1.0.0-rc.2. This version continues to focus on performance optimization, query enhancements, and repartition stability improvements, while also fixing a number of critical issues, paving the way for the upcoming v1.0 GA release.

We strongly recommend upgrading to this version.

Development Stats

Here are the development statistics for rc.2 (as of 2026-03-11):

• 84 PRs merged
• 13 contributors
• Including 1 new contributor

Breakdown of major improvements:

• 32 feature enhancements: TopK dynamic filtering, PostgreSQL COPY support, Flow TQL CTE, Prometheus 3.x compatibility, etc.
• 20 bug fixes: region repartition stability fixes, slow query log fixes, Parquet re-download fixes, etc.
• 5 refactors: vector index metadata refactoring, partition expr version renaming, remote write code cleanup, etc.
• 3 performance optimizations: Prometheus decode optimization, remote write decode optimization, Parquet scan min-max cache acceleration
• 4 test enhancements: repartition fuzz testing, vector index sqlness tests, etc.
• 2 documentation updates
• 18 engineering and dependency improvements: DataFusion/Arrow upgrades, CI and tracing enhancements, etc.

Thanks to all 13 contributors who participated in this release! Special thanks to our 2 community contributors:

• @YZL0v3ZZ (new contributor) — Fixed Metasrv PostgreSQL advisory lock release issue
• @linyihai — Added time format configuration for JSON COPY

If you're interested in building observability databases, come join the GreptimeDB community — contributions of all kinds are welcome.

Highlights

TopK Dynamic Filter Pushdown

RC2 introduces dynamic filter pushdown for TopK queries.

Consider a typical query pattern:

SELECT * FROM table
ORDER BY end_time DESC
LIMIT 10;

When running ORDER BY ... LIMIT on non-time-index columns, RC2 pushes DataFusion's runtime dynamic filters down into Mito region scan, filtering out irrelevant data early based on the TopK threshold during execution. This reduces unnecessary IO and CPU overhead.

In real-world testing:

Query	RC1	RC2
ORDER BY end_time DESC LIMIT 10	~28.86s	~0.21s

That's a 100×+ speedup.

PostgreSQL COPY Support (ADBC Postgres Path)

RC2 adds support for the following on the PostgreSQL interface:

COPY (query) TO STDOUT WITH (format binary)

This enables GreptimeDB's PostgreSQL interface to be compatible with the ADBC (Arrow Database Connectivity) PostgreSQL driver's COPY-optimized export path, significantly improving throughput for data export and analytics workloads.

Other Improvements

• Prometheus ingestion optimization: Prometheus decode and remote-write decode path optimizations, showing ~10% improvement in benchmarks.
• Mito2 Parquet scan acceleration: Introduced min-max caches in the Mito2 engine to speed up Parquet file scanning.
• Partition pruning improvements: Further optimized partition pruning to improve query selectivity and reduce scan ranges.
• Flow TQL CTE support: CREATE FLOW now supports simple TQL CTE (Common Table Expressions) patterns.
• Prometheus 3.x compatibility: Aligned selector, lookback, and matrix selector semantics with Prometheus 3.x behavior.
• PromQL GROUP BY support
• New anomaly detection functions: anomaly_score_zscore, anomaly_score_iqr, anomaly_score_mad
• JSON COPY time format configuration
• Flow statistics support
• CLI scan benchmark tool
• Internal improvements: DataFusion and Arrow dependency upgrades, GC and tracing optimizations

Notable Bug Fixes

• Region repartition stability fixes:
  • Reject writes to deallocating regions during region merge, adding write-path protection
  • Fixed synchronization issue when entering the staging phase to prevent potential data state inconsistency
• Fixed extension range filtering in Mito2 partition pruning
• Fixed incorrect PromQL plan statistics
• Fixed slowlog query panic
• Fixed partition expression handling errors
• Fixed Parquet file re-download issue
• Fixed metadata route address mapping issue

Compatibility Notes

This release contains no breaking changes. You can upgrade directly from RC1.

Closing

For the full changelog, see the GitHub Release.

Thanks to all contributors and users. We'll continue working toward 1.0 GA as planned.