How to Scale a Kubernetes Deployment

Why a Deployment can be "READY 0/0"

A Kubernetes Deployment is the desired-state description of your app. The replicas field tells the scheduler how many pods to keep running. When someone scales a Deployment to zero - to save resources, drain traffic, or pause a rollout - every pod is terminated immediately. The Deployment object itself stays; the app is just gone.

kubectl get pods                  # nothing listed
kubectl get deployment web        # READY 0/0, AVAILABLE 0
kubectl describe deployment web   # replicas: 0 in the spec

Option 1 - scale directly with kubectl

The fastest path is kubectl scale. No file editing needed:

kubectl scale deployment web --replicas=2
kubectl rollout status deployment/web   # watch pods come up

This patches the replica count live. It works immediately, but the change is not reflected in your manifest - the next kubectl apply from the old file will put it back to zero.

Option 2 - edit the manifest and re-apply

The GitOps-safe approach: fix the source of truth (the YAML file), then apply it so the cluster and the file agree.

# deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 2        # was 0 - set to a sensible number
  selector:
    matchLabels:
      app: web
  template:
    metadata:
      labels:
        app: web
    spec:
      containers:
        - name: web
          image: nginx:stable

Apply the fixed manifest:

kubectl apply -f deployment.yaml
kubectl rollout status deployment/web

Pods appear within seconds. kubectl get pods will show them moving through ContainerCreating to Running.

Which approach to use

Use kubectl scale for an immediate incident fix. Then update your manifest (or Helm values) so the file matches the live state - otherwise the next deployment pipeline run silently scales back to zero.

In production, the bigger question is why it was zero in the first place. Check the audit log:

kubectl get events --sort-by=.lastTimestamp

Most shops lock down kubectl scale to specific service accounts and require a pull request for replica count changes - so a scale-to-zero incident is traceable to a commit or a pipeline step, not a mystery.