Vacuum Is Lying To You? | Scaling Postgres 397
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Episode 397 of Scaling Postgres covers several topics. The main focus is on how PostgreSQL's VACUUM has limited effectiveness on B-tree indexes: it removes dead tuple pointers and marks empty pages as reusable, but cannot merge sparse pages, reduce tree depth, or restructure the B-tree. After heavy deletions, index bloat persists and can skew query planner statistics, causing it to prefer sequential scans. The recommended fix is REINDEX CONCURRENTLY rather than more vacuuming. Additional topics include: a beta PG Edge Agentic AI toolkit (MCP server, vectorizer, RAG server, BM25 hybrid search) for using LLMs with any Postgres instance; index corruption risks from OS-level collation changes during upgrades and how logical replication upgrades avoid this; PlanetScale Metal's new $50/month Postgres tier; a PG OIDC validator extension for token caching ahead of OAuth2 in Postgres 18; and PG Edge Anonymizer for PII masking. A consulting corner shares real-world optimization work: query tuning, pre-aggregating data, and replacing LIMIT/OFFSET pagination.