You obviously have not operated kubernetes long enough. With large enough scale you'll find the cluster controllers crashing for various reasons, getting out of sync with each other, etcd crashing or getting locked, and a whole bag of bugs.