Every deploy drops 0.3% of requests for fifteen seconds
A rolling deploy produces a reliable burst of 502s. The entrypoint is `sh -c "node server.js"`, the termination grace period is 30 seconds, and the application has no signal handler. The team's proposed fix is to raise the grace period to 120 seconds.
The fix that looks right
Raising the grace period, which is the fix that everyone proposes and that costs nothing to try. It is a change to how long the platform waits for a process that is not listening, so the drops continue at exactly the same rate while deploys get slower — and the slower deploys make the next investigation harder, because now a deploy takes long enough that people start it and go do something else. The second version of the trap is adding client retries and closing the ticket: idempotent GETs stop showing errors, and the non-idempotent writes are now silently duplicated instead of visibly failed.