What is happening
etcd detects hash disagreement and can replace a corrupted member from healthy peers when quorum remains. Catastrophic restore recreates cluster membership from a snapshot; Kubernetes restores may require a revision bump so controllers and watches do not rely on stale cached revisions.