Background jobs
Worker fleet
Five simulated workers over five queues, running the real Queue and WorkerPool.
Queue depth
jobs waiting to run, sampled every 500ms of virtual time
Retry policy
full jitter · four attempts
Rate limit
,
4 concurrent
15s lease
Workers
,
5
Recent jobs
newest first; click a row for the stored payload
Event log
claim, completion, retry and dead-letter events
Recurring jobs
the scheduler enqueues these like any other job
Reliability
Dead-letter queue
A job lands here after its final allowed attempt. Replay resets the attempt counter and puts it back at the head of the queue, an operator decision, not an automatic retry.
Parked jobs
click a row to inspect the error and payload
Event log
same stream as the fleet view
Why jobs die
attempt 4/4
Chaos mode fails 30% of handler runs. With four attempts and full jitter the chance a job exhausts its budget is roughly 0.34 ≈ 0.8%, low, but a fleet processes enough jobs that the DLQ fills anyway.
lease expiry
A worker that stops heartbeating leaves a job in running. The 15-second lease returns it to the queue and spends one attempt, so a job that kills its worker cannot loop forever.
replay
Replay is manual on purpose. Automatic DLQ draining hides the one signal you need: which payloads are permanently broken.