Course outline

01 · HTTP & REST APIs

02 · Databases & Persistence

03 · Authentication & Authorization

04 · API Design & gRPC

05 · Caching, Queues & Background Jobs

06 · Production-Ready Services

07 · Advanced Auth & Tenant Security

08 · Performance & Caching at Scale

09 · Testing & Release Confidence

Stage 3 · Build

Production-Ready Services

Graceful Shutdown

Handle SIGTERM, stop accepting traffic, drain requests, close pools, and respect Kubernetes deadlines.

8 min readBuilding Backend ApplicationsBuild

Why Graceful Shutdown

When Kubernetes terminates a pod, it sends SIGTERM and waits 30 seconds before SIGKILL. Without graceful shutdown, in-flight requests are dropped, database connections are leaked, and users see errors.

Signal Handling

Golistening-for-shutdown-signals.go

26 linesLn 1, Col 1Go

signal.Notify captures SIGINT and SIGTERM. On signal, call srv.Shutdown which stops accepting new connections and waits for in-flight requests to complete.

HTTP Server Shutdown

Goshutdown-with-resource-cleanup.go

37 linesLn 1, Col 1Go

Shutdown order: stop HTTP server first (no new requests), then close database and Redis pools (wait for in-flight queries). This ensures clean resource release.

Draining Connections

Godraining-with-keep-alive-connections.go

24 linesLn 1, Col 1Go

srv.Shutdown waits for active requests and idle keep-alive connections to finish. The timeout prevents waiting forever if a client holds a connection open.

Cleanup Resources

Goresource-cleanup-pattern.go

35 linesLn 1, Col 1Go

Wrap all resources in an App struct. Shutdown each resource in order. Collect all errors. This ensures nothing is leaked even if one resource fails to close.

Kubernetes Deadlines

Kubernetes gives you 30 seconds (default) between SIGTERM and SIGKILL. Your shutdown timeout must be less than this. A common pattern is 25 seconds for shutdown, leaving 5 seconds of buffer.

Kubernetes pod spec


apiVersion: v1
kind: Pod
spec:
  terminationGracePeriodSeconds: 35
  containers:
    - name: api
      lifecycle:
        preStop:
          exec:
            command: ["/bin/sh", "-c", "sleep 5"]
      readinessProbe:
        httpGet:
          path: /readyz
          port: 8080
        initialDelaySeconds: 5
        periodSeconds: 5

preStop hook sleeps 5 seconds to let the load balancer remove the pod from rotation before the application starts shutting down. This prevents requests from being routed to a shutting-down pod.

Match shutdown timeout to Kubernetes

If your shutdown timeout is 30 seconds and Kubernetes terminationGracePeriodSeconds is 30 seconds, Kubernetes will SIGKILL your process before it finishes shutting down. Set your timeout to 25 seconds.

Mark this lesson complete to store local progress and unlock a cleaner resume path the next time you visit.