Liveness probe
A health check in Kubernetes that determines whether a container is still "alive" or has hung. If it fails, the kubelet restarts the container automatically. The solution to processes that haven't crashed but are stuck in a deadlock.
Mechanic: the kubelet periodically runs a probe — an HTTP GET against an endpoint, a TCP connection attempt, or a command in the container. If it doesn't respond correctly a certain number of times in a row, the container is considered dead and restarted. Use case: an app stuck in a deadlock, a memory leak that made it unresponsive, or a hung thread — the process is alive (didn't crash) but doesn't work, and only a restart helps. Pitfall: an overly aggressive liveness probe (too short a timeout, or one that tests dependencies) can restart healthy containers needlessly and make problems worse (a restart storm under high load) — it should only test "am I hung", not "do my dependencies work". Distinct from readiness (whether it should take traffic) and startup (whether it has finished starting). Related to readiness probe and startup probe.