What problem does it solve?
This Skill helps you diagnose and fix StarRocks compaction backlogs that cause rising query latency, disk space pressure from accumulated rowsets, and stalled or inefficient merges on BE nodes.
Core Features & Use Cases
- Compaction visibility and scoring: Interprets compaction score patterns and reads live task state via SQL (
SHOW PROC '/compactions', SHOW TABLET ... STATUS', information_schema.be_compactions) and BE APIs to identify high-urgency tablets.
- Operational tuning and configuration: Guides BE parameter adjustments for compaction threads and thresholds (cumulative vs base), plus Primary Key Table (PKT) delete vector compaction controls.
- Safe remediation workflows: Covers manual triggers at table/partition/tablet scope (
ALTER TABLE ... COMPACT, BE HTTP endpoints), along with monitoring (Prometheus/Grafana) and troubleshooting for stuck compactions and I/O saturation.
Quick Start
Use this skill to determine whether you need cumulative or base compaction, identify the worst tablets/BE nodes, and decide the safest manual trigger or BE configuration change to reduce compaction score.