A dead simple, deterministic threshold alert on the pod restart metric in any monitoring tool could also surface this same issue.

In fact, it happened to me today at work!