Just how far can the FOR UPDATE SKIP LOCKED Postgres queue go?
A small Postgres job-queue experiment hit CPU saturation before exhausting its connection slots: on one CPU, throughput levelled off around 376 jobs/s with 64 workers, while p99 claim/completion latency rose to roughly 393 ms.
These are single-run results for a specific workload with 100 ms simulated work per job, not a universal queue limit. The sample also leaves crash recovery out: production workers need a way to reclaim abandoned jobs. The Lobsters discussion was unavailable when checked.