If the thormanager hits an exception whilst starting, it fails to notify the agent and exits cleanly.
This causes the agent to sit indefinitely waiting for an appropriate workunit state change.

Ensure the manager exception in relayed to the workunit, and avoid a clean exit code.
NB: if the thormanager job died for other reasons (e.g. k8s delete it abruptly), then the agent would have noticed and caught the k8s exit status.