Zero-downtime deployments with Kubernetes rolling updates

Welcome to devops.blog! This is your first post. Replace this content with your own and start sharing your knowledge.

The problem with naive deployments

When you run kubectl set image deployment/my-app app=my-app:v2, Kubernetes starts replacing your pods immediately. Without the right configuration, this means:

• Requests hitting pods that are starting up but not yet ready
• Old pods being killed before new ones pass health checks
• No automatic rollback if the new version is broken

The right way: readiness probes

Tell Kubernetes exactly when a pod is ready to accept traffic:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
spec:
  replicas: 3
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 1        # how many extra pods during rollout
      maxUnavailable: 0  # never kill a pod before a new one is ready
  template:
    spec:
      containers:
      - name: app
        image: my-app:v2
        readinessProbe:
          httpGet:
            path: /healthz
            port: 8080
          initialDelaySeconds: 5
          periodSeconds: 3
          failureThreshold: 3

Monitoring the rollout

# Watch the rollout in real-time
kubectl rollout status deployment/my-app --timeout=5m
 
# Check rollout history
kubectl rollout history deployment/my-app
 
# Instant rollback if something's wrong
kubectl rollout undo deployment/my-app

PodDisruptionBudgets for extra safety

Prevent too many pods from being unavailable simultaneously:

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: my-app-pdb
spec:
  minAvailable: 2        # always keep at least 2 pods running
  selector:
    matchLabels:
      app: my-app

Conclusion

Zero-downtime deployments aren't magic — they require telling Kubernetes:

1. When pods are ready — via readinessProbe
2. How to roll — via RollingUpdate strategy with maxUnavailable: 0
3. How to stay available — via PodDisruptionBudget

Set these up once per deployment and you'll never have a maintenance window again.