Worker is not running

Missing cleanup, scheduled production, or collector progress can mean a worker is suspended, unclaimed, or failing.

Inspect worker state

SELECT worker.id, worker.name, worker.system_id, worker.stream_id,
    worker.consumer_group_id, worker.target_instances,
    count(instance.id) FILTER (WHERE instance.expires_at > now()) AS live_instances,
    coalesce(max(instance.attempts) FILTER (WHERE instance.expires_at > now()), 0) AS attempts
FROM {schema}.worker_config AS worker
LEFT JOIN {schema}.worker_instance AS instance ON instance.worker_id = worker.id
WHERE worker.name = '{worker}'
GROUP BY worker.id
ORDER BY worker.id;

Use the warning or alert’s worker name. Match the returned owner ids to the affected resource.

Result	Action
`target_instances` is `0`	The worker is suspended. Follow step 1 if it should run.
Nonzero target, `live_instances` is `0`	No instance currently holds a live claim. Follow step 2.
Live instances with `attempts` above `0`	The worker is running but failing. Follow step 3.
Live instances with `attempts` equal to `0`	A claim exists without a recorded failure streak. Check `SQL0040` for late ticks, or the collector completion measurement for collector progress.
No rows	Check the worker name and installation. Run the resource’s existing registration code if its declaration is missing.

Restore the worker

If stream maintenance was suspended, resume the intended worker.
```
sqlstreams stream janitor unsuspend '{stream}'
```
Use vacuum in place of janitor for stream_vacuum. Vacuum starts suspended by default, so an unused vacuum worker is not itself a fault.

If the system’s manager or metrics_collector row was suspended through an administrative database change, restore its normal one-instance target:
```
UPDATE {schema}.worker_config
SET target_instances = 1
WHERE id = {worker_id}
    AND system_id IS NOT NULL
    AND name IN ('manager', 'metrics_collector')
    AND target_instances = 0
RETURNING id, name, target_instances;
```
No returned row means it did not match those conditions. Re-registration preserves suspension and does not substitute for this change.
If background workers have no live claims, run a manager.
```
sqlstreams manager run
```
For message_consumer or exception_consumer, start the application’s Consume call instead. The standalone manager cannot run application consumer handlers. A suspended parent manager must be resumed first.

If the worker has failures, read its latest tick error. Match worker and owner in either log:

level=WARN msg="could not run worker tick -- backing off" worker={worker} owner={owner} attempts={attempts} delay={delay} error={error}
level=ERROR msg="worker tick backoff curve exhausted -- ticks continue at its cap" code=SQL0036 worker={worker} owner={owner} attempts={attempts} delay={delay} error={error}

Fix the nested database or configuration error. Ticks continue with backoff, so a process restart alone does not correct the cause.

Background managers and workers — How workers start, renew claims, and take over.
Worker logs — Ownership, suspension, and tick failure fields.
Control stream maintenance — Inspect, suspend, or resume a janitor or vacuum worker.

Inspect worker state

Restore the worker

Related topics