Chapter 18: Kubernetes Workloads

18.5 ReplicaSet — keep N Pods

By Ravindra Bagale · Last updated: 25 October 2026

Why: the problem

Ola runs its fare service with five copies. One copy dies at peak time. Someone has to notice and start a new one. With a ReplicaSet nobody has to. A ReplicaSet has one job: keep exactly N Pods with a given label alive. It counts the Pods that match its selector. Too few, it creates more from its template. Too many, it deletes some. It does not care why a Pod disappeared. It just fixes the count. You rarely write a ReplicaSet by hand, because a Deployment creates and manages ReplicaSets for you. But every Deployment runs on ReplicaSets, and many interview questions are about how the count and the label selector work. In this lesson you will write a ReplicaSet, delete Pods, add a stray Pod with the same label, and see the ReplicaSet adopt or delete it.

What we will build

You will apply a ReplicaSet with 3 nginx Pods, scale it, delete Pods, add a matching bare Pod, and change a Pod's label to remove it from the set. You will also see why changing the template does not update running Pods.

What: every term in plain English

• ReplicaSet — A controller that keeps a set number of Pods with matching labels running.
• replicas — The number of Pods you want.
• selector — The label rule the ReplicaSet uses to count its Pods.
• template — The Pod recipe used to make new Pods.
• Label — A key and value on an object, like app: fare.
• Adopt — A ReplicaSet takes ownership of a matching Pod that has no owner.
• Orphan — A Pod whose owner was deleted with --cascade=orphan.

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 the ReplicaSet

   cat rs.yaml
   

   What it does: Three Pods with the label app: fare.

   Output

   apiVersion: apps/v1
   kind: ReplicaSet
   metadata:
     name: fare
   spec:
     replicas: 3
     selector:
       matchLabels: {app: fare}
     template:
       metadata:
         labels: {app: fare}
       spec:
         containers:
         - name: web
           image: nginx:1.27
   

4. Apply it

   kubectl apply -f rs.yaml && kubectl get rs fare
   

   What it does: The ReplicaSet makes three Pods.

   Output

   replicaset.apps/fare created
   NAME   DESIRED   CURRENT   READY   AGE
   fare   3         3         3       6s
   

5. Scale it

   kubectl scale rs fare --replicas=5 && kubectl get pods -l app=fare
   

   What it does: Two more Pods appear.

   Output

   replicaset.apps/fare scaled
   NAME         READY   STATUS    
   fare-4xkq2   1/1     Running
   fare-7mzb9   1/1     Running
   fare-9plw3   1/1     Running
   fare-c2vn8   1/1     Running
   fare-tq5rd   1/1     Running
   

6. Delete two Pods

   kubectl delete pod fare-4xkq2 fare-7mzb9 && kubectl get rs fare
   

   What it does: The count drops for a second and is fixed.

   Output

   pod "fare-4xkq2" deleted
   pod "fare-7mzb9" deleted
   NAME   DESIRED   CURRENT   READY
   fare   5         5         4
   

7. Read the events

   kubectl describe rs fare | tail -5
   

   What it does: Every create and delete is logged.

   Output

   Normal  SuccessfulCreate  12s  replicaset-controller  Created pod: fare-h8jx4
   Normal  SuccessfulCreate  12s  replicaset-controller  Created pod: fare-k1rf6
   

Example 1: A stray Pod with the same label is deleted

Stack. kind, Kubernetes 1.30

Goal. A stray Pod with the same label is deleted.

stray.yaml

apiVersion: v1
kind: Pod
metadata:
  name: stray
  labels:
    app: fare
spec:
  containers:
  - name: web
    image: httpd:2.4

terminal

kubectl get rs fare
kubectl apply -f stray.yaml
sleep 3
kubectl get pods -l app=fare
kubectl get rs fare

What each line does

• labels app: fare — Same label as the ReplicaSet selector.
• image httpd — A different image. The ReplicaSet does not check images, only labels.
• Pod count 6 — One too many, so the ReplicaSet deletes one, maybe the stray.

Green output

NAME   DESIRED   CURRENT   READY
fare   5         5         5
pod/stray created
NAME         READY   STATUS        
fare-9plw3   1/1     Running
fare-c2vn8   1/1     Running
fare-h8jx4   1/1     Running
fare-k1rf6   1/1     Running
fare-tq5rd   1/1     Running
stray        0/1     Terminating

Expected result. Labels are the only thing that counts. Never reuse a label that a controller selects.

Example 2: Take a Pod out of the set to debug it

Stack. kubectl label

Goal. Take a Pod out of the set to debug it.

terminal

kubectl label pod fare-9plw3 app=debug --overwrite
kubectl get pods -L app
kubectl exec fare-9plw3 -- nginx -v

What each line does

• label --overwrite — Changes app from fare to debug.
• -L app — Adds a column with the app label.
• New Pod — The ReplicaSet sees only 4 matching Pods and makes a fifth.
• fare-9plw3 — Keeps running, out of traffic and out of the set, ready for you to inspect.

Green output

pod/fare-9plw3 labeled
NAME         READY   STATUS    APP
fare-9plw3   1/1     Running   debug
fare-c2vn8   1/1     Running   fare
fare-h8jx4   1/1     Running   fare
fare-k1rf6   1/1     Running   fare
fare-m3zt7   1/1     Running   fare
fare-tq5rd   1/1     Running   fare
nginx version: nginx/1.27.0

Expected result. A real on-call trick: pull a bad Pod out of the Service by changing its label, keep it alive, and study it while a fresh copy serves traffic.

Example 3: Template changes do not touch running Pods

Stack. kind, Kubernetes 1.30

Goal. Template changes do not touch running Pods.

terminal

kubectl set image rs/fare web=nginx:1.26
kubectl get pods -l app=fare -o custom-columns=NAME:.metadata.name,IMAGE:.spec.containers[0].image
kubectl delete pod -l app=fare --wait=false
sleep 5
kubectl get pods -l app=fare -o custom-columns=NAME:.metadata.name,IMAGE:.spec.containers[0].image

What each line does

• set image rs/fare — Changes the template only.
• Old Pods — Still run nginx:1.27. A ReplicaSet only counts; it never updates.
• delete pod — New Pods are made from the new template with nginx:1.26.
• Lesson — This is why you use a Deployment for updates.

Green output

replicaset.apps/fare image updated
NAME         IMAGE
fare-c2vn8   nginx:1.27
fare-h8jx4   nginx:1.27
NAME         IMAGE
fare-p9d2k   nginx:1.26
fare-w4n8s   nginx:1.26

Expected result. A ReplicaSet does not roll out. The Deployment in lesson 18.7 makes a new ReplicaSet for each version and moves Pods across safely.

Verify: did it work?

1. Check desired equals ready

   kubectl get rs fare -o jsonpath='{.spec.replicas} {.status.readyReplicas}{"\n"}'
   

   What it does: Both numbers must match.

   Output

   5 5
   

2. See the owner of a Pod

   kubectl get pod -l app=fare -o jsonpath='{.items[0].metadata.ownerReferences[0].name}{"\n"}'
   

   What it does: Shows the ReplicaSet name.

   Output

   fare
   

When it is red 1

The ReplicaSet "fare" is invalid: spec.template.metadata.labels: Invalid value: map[string]string{"app":"web"}: `selector` does not match template `labels`

Cause. The template labels do not match the selector.

Fix, step by step

1. Make template.metadata.labels contain every key in selector.matchLabels.
2. Apply again.

When it is red 2

field is immutable (spec.selector)

Cause. You tried to change the selector of an existing ReplicaSet.

Fix, step by step

1. Delete the ReplicaSet and create it again with the new selector.

When it is red 3

Pods keep disappearing right after you create them

Cause. Another controller selects the same labels and deletes extras.

Fix, step by step

1. Run kubectl get rs,deploy -o wide and compare selectors.
2. Use unique labels per app.

Cleanup

1. Delete the ReplicaSet and Pods

   kubectl delete rs fare
   kubectl delete pod fare-9plw3 --ignore-not-found
   

   What it does: Deletes the set and its Pods, plus the debug Pod.

   Output

   replicaset.apps "fare" deleted
   pod "fare-9plw3" 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 ReplicaSet keeps N Pods with matching labels.
• It counts by label only, not image.
• It never updates running Pods.
• Relabel a Pod to pull it out for debugging.
• Deployments create ReplicaSets for you.

FAQ

Should I create ReplicaSets myself?

Almost never. Use a Deployment, which adds rolling updates and history.

What is matchExpressions?

A richer selector, like key in a list of values.

Why does the stray Pod get deleted and not another?

The controller prefers to delete Pods that are not ready, newer, or on busier nodes.

Can two ReplicaSets share labels?

They can, but they will fight. Always use unique labels.

What happens to Pods when I delete a ReplicaSet?

They are deleted too, unless you use --cascade=orphan.

Interview questions

1. What is a ReplicaSet?

   A controller that ensures a specified number of Pod replicas matching a selector are running.

2. How does it decide which Pods are its own?

   By the label selector and ownerReferences.

3. Why use a Deployment instead?

   A Deployment manages ReplicaSets to give rolling updates, rollback and revision history.

4. What happens if you create a Pod with a matching label?

   The ReplicaSet may adopt it and, if over the count, delete a Pod.

5. Can you change a ReplicaSet selector?

   No, it is immutable in apps/v1.

Samajla ka?

Ravi Sir's Cards for this lesson

Revision cardInterview card

मराठी cards