The controller would recover eventually because it double-checks when seeing the pod status update, but when reacting to events it recovers faster. Kubernetes-commit: 03c34ede467ea16580beee3937f9997070421529