Why Does Your Kubernetes Pod Always Take 30 Seconds to Die?
A pod that takes exactly 30 seconds to terminate is a pod that never heard SIGTERM. Why the kernel refuses to deliver a default-action signal to PID 1, why a shell-form CMD puts sh in that seat and your app behind it, how to check what PID 1 actually is in a running container, and the three fixes: exec form, a real signal handler, or tini.