SIGTERM-triggered graceful shutdown is exactly what systemd sends before SIGKILL when stopping a service, and Kubernetes relies on the same contract to give a pod's main process a chance to drain connections before its termination grace period expires. The `volatile sig_atomic_t` requirement here isn't pedantic: calling non-async-signal-safe functions like `printf` or `malloc` inside a real handler is a documented cause of heap corruption and deadlocks, since a signal can interrupt malloc mid-operation.