Chapter 18: Kubernetes Workloads

18.3 A Pod, and the phase you can quote

By Ravindra Bagale · Last updated: 25 October 2026

Why: the problem

At Paytm an on-call engineer gets a message: "payments is down". She runs kubectl get pods and sees a word in the STATUS column. That one word tells her where to look. Pending means the Pod has no node yet, so she checks CPU, memory and node selectors. ContainerCreating means it is pulling the image or mounting a disk. Running means the process started. CrashLoopBackOff means the process keeps dying, so she reads logs. ImagePullBackOff means the image name or the registry login is wrong. If you cannot read these words, you guess. Guessing during an outage wastes time. A Pod is the smallest thing Kubernetes runs: one or more containers that share one IP and can share disks. The Pod has a phase (Pending, Running, Succeeded, Failed, Unknown) and each container has its own state and reason. In this lesson you will make each status happen on purpose, so you know it by face.

What we will build

You will write a Pod YAML by hand and watch it go Pending to Running. Then you will break it four ways: an image that does not exist, a command that exits, a CPU request too big for the node, and a command that finishes. You will read the phase, the reason and the Events for each one.

What: every term in plain English

• Pod — The smallest unit Kubernetes runs. One or more containers with one shared IP.
• Phase — The Pod life stage: Pending, Running, Succeeded, Failed or Unknown.
• Container state — Waiting, Running or Terminated, with a reason like ErrImagePull.
• STATUS column — What kubectl get pods shows. It mixes phase and reason, for example CrashLoopBackOff.
• Events — Short messages from the scheduler and kubelet, shown at the end of kubectl describe.
• kubelet — The agent on each node that starts and watches containers.
• restartPolicy — Always, OnFailure or Never. Decides if a dead container is started again.
• Sidecar — A helper container in the same Pod, like a log shipper.

How: step by step

1. Start a lab cluster

   kind create cluster --name lab
   kubectl get nodes
   

   What it does: Creates a one-node Kubernetes cluster inside Docker. On minikube use: minikube start.

   Output

   Creating cluster "lab" ...
    ✓ Ensuring node image (kindest/node:v1.30.0)
    ✓ Preparing nodes
   Set kubectl context to "kind-lab"
   NAME                STATUS   ROLES           AGE   VERSION
   lab-control-plane   Ready    control-plane   40s   v1.30.0
   

2. Make a practice namespace

   kubectl create namespace practice
   kubectl config set-context --current --namespace=practice
   

   What it does: Keeps all lab objects in one folder that is easy to delete.

   Output

   namespace/practice created
   Context "kind-lab" modified.
   

3. Write a Pod

   cat pod.yaml
   

   What it does: A simple Pod with one nginx container.

   Output

   apiVersion: v1
   kind: Pod
   metadata:
     name: hello
     labels: {app: hello}
   spec:
     containers:
     - name: web
       image: nginx:1.27
       ports: [{containerPort: 80}]
   

4. Apply and watch

   kubectl apply -f pod.yaml && kubectl get pod hello -w
   

   What it does: Shows the status change live. Press Ctrl+C to stop.

   Output

   pod/hello created
   NAME    READY   STATUS              RESTARTS   AGE
   hello   0/1     ContainerCreating   0          1s
   hello   1/1     Running             0          4s
   

5. Read the phase only

   kubectl get pod hello -o jsonpath='{.status.phase}{"\n"}'
   

   What it does: Prints the real phase field, not the mixed STATUS word.

   Output

   Running
   

6. Describe the Pod

   kubectl describe pod hello | tail -8
   

   What it does: Events show each step: scheduled, pulled, created, started.

   Output

   Events:
     Type    Reason     Age   From               Message
     Normal  Scheduled  20s   default-scheduler  Successfully assigned practice/hello to lab-control-plane
     Normal  Pulled     18s   kubelet            Container image "nginx:1.27" already present
     Normal  Created    18s   kubelet            Created container web
     Normal  Started    18s   kubelet            Started container web
   

7. Open a shell inside

   kubectl exec -it hello -- sh -c "hostname; curl -s localhost | head -4"
   

   What it does: Runs commands inside the container.

   Output

   hello
   !DOCTYPE html
   html
   head
   titleWelcome to nginx!/title
   

Example 1: ImagePullBackOff: a wrong image tag

Stack. kind, Kubernetes 1.30

Goal. ImagePullBackOff: a wrong image tag.

bad-image.yaml

apiVersion: v1
kind: Pod
metadata:
  name: badimg
spec:
  containers:
  - name: web
    image: nginx:9.99-doesnotexist

terminal

kubectl apply -f bad-image.yaml
sleep 20
kubectl get pod badimg
kubectl describe pod badimg | grep -A3 "Failed"

What each line does

• image: nginx:9.99-doesnotexist — A tag that is not on Docker Hub.
• sleep 20 — Gives the kubelet time to try and fail.
• STATUS ImagePullBackOff — The kubelet gave up for now and will retry with growing waits.
• grep Failed — Shows the exact pull error from the registry.

Green output

pod/badimg created
NAME     READY   STATUS             RESTARTS   AGE
badimg   0/1     ImagePullBackOff   0          21s
  Warning  Failed     15s   kubelet  Failed to pull image "nginx:9.99-doesnotexist": rpc error: code = NotFound desc = failed to resolve reference "docker.io/library/nginx:9.99-doesnotexist": not found
  Warning  Failed     15s   kubelet  Error: ErrImagePull

Expected result. Phase is Pending, because no container ever ran. Fix: correct the tag, or add imagePullSecrets for a private registry.

Example 2: CrashLoopBackOff: a process that keeps exiting

Stack. busybox 1.36

Goal. CrashLoopBackOff: a process that keeps exiting.

crash.yaml

apiVersion: v1
kind: Pod
metadata:
  name: crash
spec:
  containers:
  - name: app
    image: busybox:1.36
    command: ["sh", "-c", "echo starting; echo DB_HOST missing; exit 1"]

terminal

kubectl apply -f crash.yaml
sleep 40
kubectl get pod crash
kubectl logs crash --previous
kubectl get pod crash -o jsonpath='{.status.containerStatuses[0].lastState.terminated.exitCode}{"\n"}'

What each line does

• command ... exit 1 — The app starts and exits with an error, like a missing setting.
• RESTARTS — Counts how many times kubelet restarted it.
• CrashLoopBackOff — kubelet waits 10s, 20s, 40s and up to 5 minutes between restarts.
• logs --previous — Shows the logs of the last dead container, where the reason is.
• exitCode — 1 means app error. 137 means killed, often out of memory.

Green output

pod/crash created
NAME    READY   STATUS             RESTARTS      AGE
crash   0/1     CrashLoopBackOff   3 (15s ago)   41s
starting
DB_HOST missing
1

Expected result. The phase is Running, but the container keeps dying. Always read logs --previous first.

Example 3: Pending and Succeeded: no room, and a job that finishes

Stack. kind, Kubernetes 1.30

Goal. Pending and Succeeded: no room, and a job that finishes.

big.yaml

apiVersion: v1
kind: Pod
metadata:
  name: big
spec:
  containers:
  - name: app
    image: nginx:1.27
    resources:
      requests:
        cpu: "64"
        memory: 512Gi
---
apiVersion: v1
kind: Pod
metadata:
  name: once
spec:
  restartPolicy: Never
  containers:
  - name: task
    image: busybox:1.36
    command: ["sh", "-c", "echo report done"]

terminal

kubectl apply -f big.yaml
sleep 10
kubectl get pods big once
kubectl describe pod big | grep -A2 FailedScheduling
kubectl logs once

What each line does

• requests cpu: "64" — Asks for 64 CPUs. No kind node has that.
• FailedScheduling — The scheduler found no node that fits.
• restartPolicy: Never — Do not restart when the process ends.
• Completed — The STATUS word for phase Succeeded.

Green output

pod/big created
pod/once created
NAME   READY   STATUS      RESTARTS   AGE
big    0/1     Pending     0          10s
once   0/1     Completed   0          10s
  Warning  FailedScheduling  9s  default-scheduler  0/1 nodes are available: 1 Insufficient cpu, 1 Insufficient memory. preemption: 0/1 nodes are available
report done

Expected result. Pending with FailedScheduling means "no node fits". Completed means the process exited 0 and will not restart.

Verify: did it work?

1. List all Pods with phase

   kubectl get pods -o custom-columns=NAME:.metadata.name,PHASE:.status.phase,STATUS:.status.containerStatuses[0].state
   

   What it does: Puts the real phase next to the container state.

   Output

   NAME     PHASE       STATUS
   badimg   Pending     map[waiting:map[reason:ImagePullBackOff]]
   crash    Running     map[waiting:map[reason:CrashLoopBackOff]]
   hello    Running     map[running:map[startedAt:...]]
   once     Succeeded   map[terminated:map[exitCode:0 reason:Completed]]
   

2. Recent warnings only

   kubectl get events --field-selector type=Warning
   

   What it does: A fast list of what is going wrong.

   Output

   LAST SEEN   TYPE      REASON             OBJECT       MESSAGE
   30s         Warning   BackOff            pod/crash    Back-off restarting failed container
   25s         Warning   FailedScheduling   pod/big      0/1 nodes are available: 1 Insufficient cpu
   

When it is red 1

Error from server (BadRequest): previous terminated container "app" in pod "crash" not found

Cause. The container has not crashed yet, so there is no previous log.

Fix, step by step

1. Wait until RESTARTS is 1 or more.
2. Run kubectl logs crash without --previous.

When it is red 2

OOMKilled, exit code 137

Cause. The container used more memory than its limit.

Fix, step by step

1. Run kubectl describe pod and find Last State: Terminated, Reason: OOMKilled.
2. Raise resources.limits.memory or fix the memory leak.
3. Apply again.

When it is red 3

CreateContainerConfigError: secret "db" not found

Cause. The Pod refers to a Secret or ConfigMap that does not exist.

Fix, step by step

1. Run kubectl get secret db.
2. Create it in the same namespace.
3. The Pod starts by itself once it exists.

Cleanup

1. Delete the test Pods

   kubectl delete pod hello badimg crash big once
   

   What it does: Removes all Pods from this lesson.

   Output

   pod "hello" deleted
   pod "badimg" deleted
   pod "crash" deleted
   pod "big" deleted
   pod "once" deleted
   

2. Delete the practice namespace

   kubectl delete namespace practice
   

   What it does: Deletes every object you made in this lesson in one go.

   Output

   namespace "practice" deleted
   

Practice tasks

Homework: try at home

Recap

• A Pod is one or more containers with one shared IP.
• The phase is Pending, Running, Succeeded, Failed or Unknown.
• STATUS words like CrashLoopBackOff are container reasons.
• Read describe Events and logs --previous first.
• Pending with FailedScheduling means no node fits.

FAQ

Why is CrashLoopBackOff phase Running?

The Pod is on a node and at least one container was started. The container keeps dying, which is shown as the reason.

What is the Unknown phase?

The control plane lost contact with the node, so it cannot tell.

Should I run apps as bare Pods?

No. Use a Deployment, StatefulSet or Job so they come back. Lesson 18.4 shows why.

Can I edit a running Pod?

Only a few fields like the image. Most changes need delete and create, which is why controllers exist.

How long is the back-off?

It doubles from 10 seconds up to 5 minutes and resets after 10 minutes of running fine.

Interview questions

1. Pod is in CrashLoopBackOff. What do you do?

   Run kubectl logs --previous, check the exit code in describe, check env, config, secrets and probes, then fix and redeploy.

2. Pod is Pending. What do you check?

   describe Events for FailedScheduling: CPU or memory requests, node selectors, taints, and unbound PVCs.

3. What is the difference between ErrImagePull and ImagePullBackOff?

   ErrImagePull is the first failure. ImagePullBackOff means kubelet is waiting before trying again.

4. What is exit code 137?

   The process got SIGKILL, often because of OOMKilled or a failed liveness probe.

5. Why do containers in a Pod share localhost?

   They share one network namespace, so they have one IP and can reach each other on localhost.

Samajla ka?

Ravi Sir's Cards for this lesson

Revision cardInterview card

मराठी cards