Zero Bloat Postgres Queue | Scaling Postgres 416

PostgreSQL's traditional job queues, reliant on `FOR UPDATE SKIP LOCKED`, often spiral into performance decay due to dead tuples and index bloat under load—what the host calls the 'Q Spiral of Death.' But a new tool called PGQ, built by Nick from Postgres FM, offers a radical alternative: a zero-bloat queue that avoids updates and deletes entirely. Instead of modifying rows, PGQ uses snapshot-based batching and a rotating table system with three tables cycling through truncate operations. This design eliminates dead tuples, maintains consistent performance even at scale, and works on managed cloud Postgres. The tradeoff? Latency: jobs are processed at 50–150ms intervals by default (10 ticks per second), making it unsuitable for sub-millisecond needs but ideal for stable, high-throughput systems. The episode also covers broader ecosystem updates, including PG Backrest’s revival through community funding, PGX Backup as a potential fork, Postgres 19’s expanded multi-exact member support, and Figma’s new connection pooler PG-Keeper built to overcome PG Bouncer’s limitations. Finally, it shares practical disaster response rules and insights into how PostgreSQL committers are selected.