Graceful Shutdown / Health / Readiness：Server 不只是「開起來」

Learning outcomes

• 能區分 liveness 與 readiness。
• 能處理 termination signal。
• 能描述 graceful shutdown 順序。
• 能設計 deploy smoke / drain evidence。

1. Signal

process.on(
  "SIGTERM",
  shutdown
);

平台可能先送 termination signal，給 process 一段時間收尾；不要只靠突然 kill。

2. Graceful sequence

SIGTERM
  ↓
mark not ready
  ↓
stop accepting new requests
  ↓
wait in-flight requests
  ↓
close DB pool / resources
  ↓
exit 0

3. Health vs readiness

4. 不要讓 health endpoint 做太重

health check 本身會被頻繁呼叫。過度昂貴的外部依賴查詢可能反而放大故障。

5. Project checkpoint：Course API production lifecycle

startup
  ↓ config validation
  ↓ DB pool ready
  ↓ readiness = true
  ↓ serve traffic

deploy / shutdown
  ↓ readiness = false
  ↓ close listener
  ↓ drain
  ↓ close pool
  ↓ exit

6. Release evidence

• new version health/readiness PASS
• smoke GET /health /courses
• old process drained
• error rate / latency not spiking
• rollback path available

Backend Advanced review