Ferchau Gupta
08/18/2026, 2:16 PMMarvin
08/18/2026, 2:16 PMMarvin
08/18/2026, 2:23 PMScheduled to Pending.
- The server serializes that transition; if multiple workers see the same run, only one Pending transition succeeds. The others abort submission.
- Therefore, faster workers or workers with more free capacity may receive more runs—there is no worker-level fairness guarantee.
Capacity is controlled at several levels:
- Worker `--limit`: maximum simultaneous runs for that individual worker.
- Work-queue concurrency limit: maximum active runs from that queue.
- Work-pool concurrency limit: maximum across the pool.
So running several identical workers against one queue is the normal way to scale horizontally and add redundancy.
Sources: Prefect 2.20 work pools and workers, worker polling implementation, server-side state transition