Treating Postgres as a plain queue is why your workers stall and your vacuum…

Treating Postgres as a plain queue is why your workers stall and your vacuum debt piles up. The real mental model is “a queue running under MVCC”: rows are versioned, visibility lags, and deletes don't instantly free space. That changes the wiring — use SKIP LOCKED intentionally, keep tx scopes tiny, and schedule explicit cleanup/vacuum work rather than assuming deletes are instantaneous. The takeaway: design retries, visibility checks, and compaction as first-class parts of your agent loop, not afterthoughts.