study.

Why a Deleted Pod Comes Back · Kubernetes

Observe how controllers maintain the requested number of Pods.

← Kubernetes

Learn with progress and quizzes

First, understand the concepts.

Ownership

A Deployment manages ReplicaSets, and a ReplicaSet maintains Pods. ownerReferences identifies the object directly managing a Pod.

Continuous reconciliation

If replicas=2 and one Pod disappears, the controller replaces it. The replacement can have a different name and UID.

Change the declaration

To reduce replicas persistently, change the Deployment rather than repeatedly deleting Pods. Reconciliation follows the requested state.

Why it matters

A differently named Pod appearing after deletion may indicate successful reconciliation, not failed deletion. The controller is restoring the declared replica count.

Work through the reasoning

Follow ownership upward

Inspect Pod ownerReferences for its ReplicaSet, then that ReplicaSet’s Deployment owner. Comparing UIDs shows replacement rather than resurrection of the same object.

Change durable intent

Deleting one Pod differs from reducing Deployment replicas. Change the owning declaration to request a new count, and check whether GitOps or an autoscaler also manages it.

Learn by doing.

Use kubectl, a personal practice Kubernetes cluster, and a POSIX shell. Check the context and use the dedicated namespace in each lab. Cleanup deletes all resources in that practice namespace.

Deploy two replicas, delete one Pod from this practice workload, and observe recovery to two. Do not run this experiment in a production namespace.

kubectl config current-context
kubectl create namespace study-k8s-reconcile --dry-run=client -o yaml | kubectl apply -f -
kubectl create deployment web --image=nginx:stable-alpine --replicas=2 -n study-k8s-reconcile --dry-run=client -o yaml | kubectl apply -f -
kubectl rollout status deployment/web -n study-k8s-reconcile --timeout=120s
kubectl get pods -n study-k8s-reconcile
study_pod=$(kubectl get pods -n study-k8s-reconcile -l app=web -o jsonpath='{.items[0].metadata.name}')
kubectl delete pod "$study_pod" -n study-k8s-reconcile
kubectl rollout status deployment/web -n study-k8s-reconcile --timeout=120s
kubectl get pods -n study-k8s-reconcile
kubectl scale deployment web -n study-k8s-reconcile --replicas=1
kubectl rollout status deployment/web -n study-k8s-reconcile --timeout=120s

How to check your result

Compare names before and after deletion and confirm READY returns to 2/2. After scaling, one replica should remain. For Pending Pods, inspect resource, image, and scheduling events.

Clean up practice resources

Run after checking the result. Confirm these resources belong to your practice run and contain no data you need to keep.

kubectl delete namespace study-k8s-reconcile

Apply what you learned

You delete one Pod from a three-replica Deployment. Its replacement is Pending. What stages should you distinguish?

Reveal the reasoning after trying

The controller created a replacement, but placement or preparation is blocked. Confirm ownership, then inspect the replacement’s scheduling, resource, or storage events.

Ready to check your understanding?

How do you keep the replica count at one?

  1. Repeatedly delete a Pod
  2. Set Deployment replicas to 1
  3. Rename the Service
  4. Keep running kubectl get
Reveal the reasoning after trying

Set Deployment replicas to 1

Changing replicas updates the desired state. Deleting a managed Pod alone causes a replacement.

Official documentation · Why a Deleted Pod Comes Back