| ✅ Do                                                               | ❌ Don’t                                                 |
| ------------------------------------------------------------------ | ------------------------------------------------------- |
| Log full stack traces and context (request IDs, payloads, etc.)    | Continue running as if nothing happened                 |
| Report to external error‑tracking services (Sentry, Datadog, etc.) | Swallow errors or leave handlers empty                  |
| Gracefully shut down: close DB, stop accepting new requests        | Call `process.exit(0)` (signals success)                |
| Use a supervisor (e.g., PM2, Docker restart policy)                | Believe you can recover 100% safely in‑process          |
| Convert unhandled rejections into exceptions (`throw reason`)      | Ignore Node.js v15+ default behavior—explicit is better |