# Prometheus 3.x, Arrow-PG & BulkMemtable Optimization | Greptime Biweekly Report - No. 76 & 77

Source: https://greptime.com/blogs/2026-02-22-greptimedb-biweekly-report

## Summary

*Development period: 2026-01-26 - 2026-02-22*

This is a combined report covering the past four weeks. Here are the highlights:

- Postgres protocol encoding upgraded to `arrow-pg` with proper unsigned integer handling
- BulkMemtable lazy encoding reduces L0 SST files by ~38% in TSBS workloads
- Write operations now validate partition rule versions before staging
- Prometheus 3.x selector and lookback behavior supported
- Metadata-only table updates no longer restart memtables unnecessarily

We encourage users on older versions to upgrade for these fixes and improvements.

Our biggest announcements this period are [v1.0 RC1](/blogs/2026-02-04-greptimedb-v1.0.0-rc1-release) and the [GreptimeDB 2026 Roadmap](/blogs/2026-02-11-greptimedb-roadmap-2026) — if you want to know what's ahead for GreptimeDB in 2026, the roadmap is a must-read!

We've also redesigned the Greptime website with a cleaner look. Check it out at [greptime.com](https://greptime.com/).

## Contributors

Over the past four weeks, contributors merged a total of **58 PRs**. Among them, **3 individual contributors** contributed **3 PRs**.
Welcome to our new contributor: **[@YZL0v3ZZ](https://github.com/YZL0v3ZZ)**!

Thanks to our individual contributors:

- **[@linyihai](https://github.com/linyihai)** *[db#7633](https://github.com/GreptimeTeam/greptimedb/pull/7633)*
- **[@YZL0v3ZZ](https://github.com/YZL0v3ZZ)** *[db#7673](https://github.com/GreptimeTeam/greptimedb/pull/7673)*
- **[@ZonaHex](https://github.com/ZonaHex)** *[db#7707](https://github.com/GreptimeTeam/greptimedb/pull/7707)*

## Highlights of Recent PRs

### [db#7648](https://github.com/GreptimeTeam/greptimedb/pull/7648) [db#7634](https://github.com/GreptimeTeam/greptimedb/pull/7634) Vector index improvements (experimental)

Vector index metadata switched to Protocol Buffers format for better serialization and cross-version compatibility, and sqlness integration tests were added. This feature is still under active development.

### [db#7591](https://github.com/GreptimeTeam/greptimedb/pull/7591) feat: use arrow-pg for postgres data encoding

PostgreSQL client connections now support `IntervalStyle` configuration and handle unsigned integers more accurately by mapping them to wider PostgreSQL types (`uint32` maps to `int8`, `uint64` maps to `numeric`) to prevent overflow issues.

### [db#7617](https://github.com/GreptimeTeam/greptimedb/pull/7617) feat: BulkMemtable compact parts without encoding into Parquet

BulkMemtable now uses lazy encoding to reduce compaction overhead, batching data in memory before converting to Parquet format based on configurable row and byte thresholds. This optimization reduces L0 SST file count by approximately 38% in TSBS workloads, lowering storage amplification and improving write performance.

### [db#7628](https://github.com/GreptimeTeam/greptimedb/pull/7628) feat: partition rule version validation for writes and staging

GreptimeDB now validates partition rule versions during write operations to prevent data corruption when partition schemas change. Writers using outdated partition information will be rejected, ensuring data lands in the correct partitions even during schema transitions.

### [db#7688](https://github.com/GreptimeTeam/greptimedb/pull/7688) feat: adapt prometheus 3.x selector and lookback behavior

PromQL queries now follow Prometheus 3.x selector behavior with start-exclusive instant selectors, and lookback handling is fixed for `end` bounds and matrix selectors to return correct results.

### [db#7695](https://github.com/GreptimeTeam/greptimedb/pull/7695) fix(mito2): introduce PartitionExprChange in staging flow and keep memtables on metadata-only updates

Partition expression changes during region repartitioning no longer rebuild memtables, avoiding unnecessary memory operations and potential data loss. The system now distinguishes between metadata-only partition updates and column schema changes, ensuring safer and more efficient repartitioning workflows.

### [db#7626](https://github.com/GreptimeTeam/greptimedb/pull/7626) refactor: remove the RawTableMeta and RawTableInfo to make codes more concise

Code refactoring removes `RawTableMeta` and `RawTableInfo` internal structures, reducing code complexity without changing user-facing functionality or breaking existing queries and operations.

## Good First Issue

### [Issue#7314](https://github.com/GreptimeTeam/greptimedb/issues/7314) Feature Request: Add built-in daemon mode to keep GreptimeDB running independently of shell sessions

Add a `--daemon` CLI flag to allow GreptimeDB to run as a background process independent of shell sessions, with proper stdout/stderr redirection and PID management.

Keywords: CLI, Daemon, Process Management

Difficulty: Medium

### [Issue#7544](https://github.com/GreptimeTeam/greptimedb/issues/7544) Allow ALTER TABLE to modify the skip_wal option dynamically

Add support for dynamically changing the `skip_wal` table option through `ALTER TABLE` statements, allowing users to enable or disable WAL at runtime.

Keywords: SQL parser, Storage, WAL

Difficulty: Medium
