Stage 3 · Build

Long-Running Daemons & Services

Graceful HTTP Shutdown

http.Server Shutdown, connection draining, readiness flips, and Kubernetes terminationGracePeriodSeconds.

8 min readGo for SREBuild

Shutdown Flow

Graceful HTTP shutdown follows a specific sequence: stop accepting new connections, drain in-flight requests, and close idle connections. http.Server.Shutdown handles this automatically.

Goshutdown-flow.go

30 linesLn 1, Col 1Go

Shutdown stops accepting new connections, waits for in-flight requests to complete, and closes idle connections. The timeout prevents hanging connections from blocking shutdown. Always set a timeout.

Connection Draining

Goconnection-draining-with-tracking.go

47 linesLn 1, Col 1Go

trackingHandler counts in-flight requests. Shutdown flips readiness, waits for connections to drain, then shuts down the server. This provides explicit control over the draining process.

Readiness Flip

Goreadiness-endpoint-with-flip.go

43 linesLn 1, Col 1Go

Flip readiness to not-ready before starting shutdown. This tells Kubernetes to remove the pod from endpoints. Wait for endpoints to propagate. Then drain connections. This prevents new traffic from arriving during shutdown.

HTTP Server Configuration

Goproduction-server-configuration.go

29 linesLn 1, Col 1Go

ReadHeaderTimeout prevents slowloris attacks. WriteTimeout prevents slow responses. IdleTimeout closes idle connections. MaxHeaderBytes limits header size. These timeouts protect your server from resource exhaustion.

Shutdown Coordination

Gomulti-component-shutdown.go

49 linesLn 1, Col 1Go

Multi-component shutdown requires coordination. Each component has its own shutdown timeout. The total timeout must be less than Kubernetes terminationGracePeriodSeconds. Log errors but continue shutting down other components.

Kubernetes Integration

Kubernetes deployment


apiVersion: apps/v1
kind: Deployment
spec:
  template:
    spec:
      terminationGracePeriodSeconds: 60
      containers:
        - name: server
          ports:
            - containerPort: 8080

          # Liveness probe
          livenessProbe:
            httpGet:
              path: /healthz
              port: 8080
            initialDelaySeconds: 10
            periodSeconds: 15

          # Readiness probe
          readinessProbe:
            httpGet:
              path: /readyz
              port: 8080
            initialDelaySeconds: 5
            periodSeconds: 10

          # Startup probe
          startupProbe:
            httpGet:
              path: /started
              port: 8080
            failureThreshold: 30
            periodSeconds: 2

          lifecycle:
            preStop:
              exec:
                # Give time for endpoints to propagate
                command: ["sh", "-c", "sleep 5"]

terminationGracePeriodSeconds must exceed your shutdown timeout. preStop hook sleeps to allow endpoint propagation. readiness probe must fail after shutdown flip. liveness probe must not kill during shutdown.

preStop sleep is critical

Without the preStop sleep, Kubernetes sends SIGTERM immediately after removing the pod from endpoints. But endpoint propagation takes time. The sleep ensures the pod is fully removed from endpoints before shutdown begins.

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