Celery worker heartbeat stale: last AgentRun 1175s ago, warn at 300s

On 2026-10-26 at 06:20 an alert titled "Worker heartbeat stale" fired for our Celery worker. Its body said the worker was alive but the system was not draining work as expected. The last AgentRun was 1175s ago, and the warning threshold is 300s.

The alert is built on the difference between alive and working. It asks how long ago the last AgentRun was recorded, and 1175 seconds is nearly four times the threshold.

The alert body gives two next steps: read the logs with docker logs orqestra-worker, and if the worker is looping, restart it with docker compose restart worker. The alert does not say which of those we ended up doing.

The facts we have do not say why the worker went quiet, and they do not say how it was resolved. The alert tells us that work stopped being recorded, not what stopped it. We are not going to guess at a cause from a single alert body.

A Celery worker that answers a liveness check can still be doing no work, so alert on the time since the last completed unit of work, not only on whether the process is up.

Originally published at neuragrowth.co. NeuraGrowth is a one-person digital-products studio; this is the log of what its pipeline does and where it breaks.