Shutdown

The cleanup handlers are executed when the operator exits, either due to a signal (e.g. SIGTERM), by catching an exception, by raising the stop-flag, or by cancelling the operator’s task (for embedded operators):

import kopf
from typing import Any

@kopf.on.cleanup()
async def cleanup_fn(logger: kopf.Logger, **_: Any) -> None:
    pass

The cleanup handlers are not guaranteed to execute fully if they take too long — due to a limited graceful period or non-graceful termination.

Similarly, the cleanup handlers are not executed if the operator is force-killed with no opportunity to react (e.g. by SIGKILL).

Note

If the operator is running in a Kubernetes cluster, there can be timeouts set for graceful termination of a pod (terminationGracePeriodSeconds, the default is 30 seconds).

If the cleanup takes longer than that in total (e.g. due to retries), the activity will not finish completely, as the pod will be SIGKILL’ed by Kubernetes.

Either design the cleanup activities to be as fast as possible, or configure terminationGracePeriodSeconds accordingly.

Kopf itself does not set any implicit timeouts for the cleanup activity, and it can continue forever (unless explicitly limited).