Launching just one new agent can fix the issue for the whole cluster. Investigate why the orchestrate greenlet can stop globally!