Liveness and Readiness Probes

Your pod has been running for days. The process is alive: kubectl get pods shows Running, CPU is flat, memory is stable. But requests are silently failing. Not crashing, not erroring loudly. Just hanging. A goroutine leaked, a mutex deadlocked, or a connection pool exhausted and nothing recovered. The application is alive in every metric that Kubernetes tracks by default. And Kubernetes keeps sending traffic to it.

Liveness and readiness probes exist to close that gap. Liveness tells Kubernetes when to restart a container. Readiness tells Kubernetes when to route traffic to it. They look similar in a manifest: both are fields on a container spec, both support the same probe types. But they have completely different semantics, and conflating them is one of the most common misconfigurations in production clusters.