Myth 3: Deployment Supports All Pod Restart Policies

Can You Use Any Restart Policy in a Deployment? Many believe that Deployments can be used with any Kubernetes Pod restart policy. The assumption is that you can define Always, OnFailure, or Never as the restartPolicy within a Deployment’s Pod spec.But that's not true.

Why This Myth Exists?

1. Pods support all three restart policies (Always, OnFailure, Never), leading to the assumption that Deployments do as well.
2. Users often assume Pod behavior is the same inside different controllers like Deployments, DaemonSets, or Jobs.
3. The Kubernetes API allows specifying other restart policies in Pod definitions, but Deployments override them!

The Reality:

Deployment = restartPolicy: Always Only! In Kubernetes, Deployments are designed to manage long-running applications. That means: - If a Pod fails, Kubernetes restarts it. - If a node crashes, a new Pod is scheduled on another node. - Any restart policy other than Always is rejected!

Experiment & Validate

Let’s try using restartPolicy: OnFailure inside a Deployment and see what happens:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-deployment
spec:
  replicas: 3
  selector:
    matchLabels:
      app: test-app
  template:
    metadata:
      labels:
        app: test-app
    spec:
      restartPolicy: OnFailure
      containers:
      - name: test-container
        image: busybox
        command: ["sh", "-c", "echo Hello && exit 1"]

What Happens? Deployment rejects OnFailure and throw an error!

The Deployment "my-deployment" is invalid: spec.template.spec.restartPolicy: Unsupported value: "OnFailure": supported values: "Always"

Kubernetes rejects the Deployment, as it does not allow restart policies other than Always.

Key Takeaways

• Deployments only support restartPolicy: Always—other policies are rejected.
• Use Jobs or CronJobs for OnFailure or Never restart behavior.
• Understanding controller-specific behavior prevents misconfigurations.