PlanetScale warns: keep your Postgres job queue tidy or risk the whole app

It has been reported that PlanetScale engineer Simeon Griggs lays out a blunt, useful metaphor: a healthy queue table is a lot like a healthy digestive system—it's not about glamour, it's about getting rid of waste fast. In a post shared on Hacker News and the PlanetScale blog, Griggs walks through why Postgres is a fine place to run queues, and where teams tend to trip up when they mix high-throughput transient workloads with OLTP, analytics, and other database traffic.

Why Postgres queues are uniquely finicky

Queues are weird beasts. Most rows are transient: insert, read once, delete. That means huge cumulative throughput while table size looks stable. When you run queues alongside other workloads in the same cluster, those lightweight, ultra-frequent transactions can still hog locks, postpone vacuuming, and interfere with backups and analytics. Griggs points to common patterns—SELECT ... FOR UPDATE SKIP LOCKED to claim work, and immediate DELETE on success—and emphasizes that the big risk is long-running transactions. Keep them short. Really short.

Practical takeaways: monitor, index, and choose wisely

PlanetScale’s post includes a simple jobs table and a partial index on pending rows as a working example, and it has been reported that the core advice is straightforward: monitor the cleanup machinery before it gets clogged, use partial indexes and SKIP LOCKED to limit contention, and avoid letting worker transactions linger. There’s also a candid nod to reality: sometimes a dedicated queue service is the right call. The “just use Postgres for everything” meme has teeth, but it’s not unconditional gospel.

Don’t let a misbehaving job queue be the butterfly that flaps and blooms into an outage. Want the resilience and transactional guarantees of Postgres? Fine. But treat your queues as first-class citizens—instrument them, tune VACUUM and autovacuum, and be ready to pull the right tool off the shelf when the workload demands it. Who wants their whole app brought down by a clogged queue? No one.

Sources: planetscale.com, Hacker News