When Kubernetes terminates your pod, the orchestrator sends SIGTERM, waits 30 seconds, then SIGKILL. If your service doesn't handle that conversation correctly, you drop in-flight requests, leak connections, lose data, and corrupt distributed locks. The fix is graceful shutdown -- and it's a small, specific set of patterns that almost no team gets right on the first deploy.
Engineering Craft
TypeScript, CI/CD, databases, observability -- the skills that make code production-ready.
