learninfra · Linux · Networking · Kubernetes · System Design · AI Infrastructure · Exam blueprints · Drills

Pods (CKAD)

The smallest thing Kubernetes runs: what a pod is, how it starts, what restarts it, and what does not bring it back.

An interactive Kubernetes lesson: 20 steps, about 30 minutes, on a live simulation in your browser.

You have a cluster and a container image for the shop's web frontend: nginx:1.27. Your job is to get it running. One command does it: kubectl run web --image=nginx:1.27 --port=80.

Look at what appeared on the stage. Kubernetes did not start a container on its own. It created a pod called web, and the pod holds the container. A pod is the smallest thing Kubernetes knows how to create, place on a node and count.

What you will learn

  1. Your first pod

    • Run one container: Kubernetes never runs a bare container. It runs pods, and a pod wraps one or more containers.
    • What is inside a pod: spec is what you asked for. status is what the cluster observed. Kubernetes works to make the second match the first.
    • Every pod gets an IP: One pod, one IP. The containers inside share it, and any other pod can reach it directly.

  2. From request to Running

    • Take the scheduler away: A pod with no node is only a record in the API. It runs nowhere until the scheduler assigns it.
    • Three programs, one pod: The API server records the pod, the scheduler assigns a node, the kubelet on that node runs it.
    • Phase is not the STATUS column: Phase is one of five words about the whole pod. The STATUS column is kubectl's summary of what its containers are doing.

  3. When the container dies

    • The process inside crashes: A crashed container is restarted in place by the kubelet: same pod, same node, same IP, RESTARTS plus one.
    • restartPolicy: Always, OnFailure, Never: restartPolicy is the kubelet's instruction for an exited container: Always restarts, OnFailure restarts only after an error, Never leaves it.
    • Exit 0 under restartPolicy: Always: CrashLoopBackOff is a waiting state, not a cause: the container keeps exiting and the kubelet is pausing before the next restart.
    • Break it: a real crash loop

  4. Pods are mortal

    • Delete the pod itself: A bare pod has no owner. Delete it, or lose its node, and nothing brings it back.
    • Bring it back, with labels: Labels are tags you choose; selectors are queries over those tags. That pair is how everything in Kubernetes finds its pods.
    • Why controllers exist: A pod is what runs. A controller is the reason it is still running tomorrow.

  5. When it will not start

    • Break it: a typo in the image tag: ImagePullBackOff means the node could not fetch the image: wrong name, wrong tag, or no permission. The pod's events say which.
    • One order for every broken pod: STATUS names the layer, describe shows the events, logs give the process's own words. Read them in that order.

  6. At exam speed

    • Let kubectl write the YAML: Generate, then edit: kubectl writes the manifest, and you type only the part that is special.
    • Drill: generate a manifest
    • Drill: why did it crash?

  7. Recap & playground

    • Cheat sheet
    • Playground