payment_settlement_daily crashes at 2am with Postgres pool exhaustion. Tapifra reads the traceback, diagnoses root cause with Opus 4.7, verifies the fix, and posts it all to Slack. The engineer sees it at 8am. Nobody loses sleep.
SIGTERM sounds simple. It's not. A deep dive into the full pod termination sequence, the kube-proxy race condition nobody talks about, and how to fix dropped requests during rolling updates with a preStop hook.