Why Kubernetes Graceful Shutdown Leaves Connections Hanging

Episode notes

Kubernetes Pod graceful shutdown is supposed to let applications drain traffic and close connections before being terminated. But in practice, many teams find that clients still see connection errors, timeouts, or incomplete requests right when a Pod shuts down. In this episode, Lucas and Luna dig into the gap between the Kubernetes graceful shutdown spec and real-world behavior. They walk through the termination sequence step by step — preStop hook, SIGTERM, terminationGracePeriodSeconds — and explain why a Pod often stops receiving traffic before its application is ready to stop. They cover a real incident at a mid-size SaaS company where a rolling update caused a five-second blip of failed requests, and trace the root cause to the timing mismatch between the kube-proxy endpoint update and the Pod's actual shutdown. Lucas shares a concrete fix: a preStop sleep combined with a readiness probe delay. If you manage Kubernetes workloads in production, this episode helps you avoid a class of outages that's easy to miss.