queuelens_depth
Jobs waiting per queue. Useful on its own only when paired with throughput: a depth of 500 is normal for a queue draining 2,000 jobs a minute and an incident for one draining 20.
queuelens_oldest_job_seconds
Age of the oldest waiting job. This is the metric that catches a stalled worker: depth grows slowly, but the oldest job's age grows without bound.
queuelens_processed_total
A counter of completed jobs by queue and status. Rate it to get throughput and failure ratio; the raw counter is only useful for restart detection.
queuelens_failed_total
A counter of failed attempts. A failure ratio above one percent is the alert, not an absolute number, because traffic differs by an order of magnitude between queues.