AutoVacuum - Tuning and Monitoring

on YouTube ↗47m

" summary

Postgres autovacuum often leaves tables and indexes bloated until a restart or restore masks the problem; a practical fix is autovacuum_cost_delay 20ms→2ms (Postgres 12) which cuts a real-life 18GB table vacuum from 81 minutes to 8, and custom per-table tuning (even down to 3.3 minutes in one test) restores responsiveness. Monitoring uses collectd+graphite+grafana+timescale, tracks pending writes, write efficiency, and idle in transaction = cluster event horizon, and enforces per-table thresholds because vacuum full (and repack) is the only way to remove middle-page bloat.

AI-written summary · about

" chapters (9)

1. 0:00Introduction
2. 3:55Autovacuum tuning necessity
3. 10:20Key autovacuum metrics
4. 15:23Autovacuum settings overview
5. 24:34Advanced monitoring tools
6. 29:48Real-time autovacuum monitoring
7. 34:01Hardware impact on autovacuum
8. 40:27Table tuning strategies
9. 43:19Q&A

" related talks

• 32mOptimizing autovacuum: PostgreSQL’s vacuum cleanerSamay Sharma · Citus Con: An Event for Postgres 2022
• 55mUnderstanding AutovacuumDan Robinson · PGConf Silicon Valley 2016
• 49mUnderstanding and Fixing AutovacuumRobert Haas · PostgreSQL Conference Europe 2023
• 48mThe Good, The Bad, and The Bloated: An Autovacuum StoryJérémie Grauer · PostgreSQL Conference Europe 2025
• 27mMaintaining Large Tables in PostgreSQLSarat Balijepalli · POSETTE: An Event for Postgres 2026
• 25mUnderstanding & Managing Postgres Table BloatChelsea Dole · Citus Con: An Event for Postgres 2023
• 33mPostgreSQL Observed—and ExplainedKaren Jex, Stacey Haysler · PostgreSQL Conference Europe 2024
• 19mVacuuming Enhancements in PostgreSQL 18: Faster, Smarter, More PredictableShashikant Shakya · POSETTE: An Event for Postgres 2026