Kubernetes Autoscaling Explained: HPA, VPA, and Cluster Autoscaler with kubectl

Kubernetes autoscaling has three independent knobs: the HorizontalPodAutoscaler (HPA) changes replica count, the VerticalPodAutoscaler (VPA) changes per-Pod CPU/memory requests, and the Cluster Autoscaler changes the number of Nodes. All three read from the same Metrics Server.

Architecture Overview

Prerequisites

kubectl version --client
kubectl get deployment metrics-server -n kube-system

Step 1: Install the Metrics Server

kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

kubectl get deployment metrics-server -n kube-system
kubectl top nodes
kubectl top pods

Step 2: Deploy a Workload with Resource Requests

apiVersion: apps/v1
kind: Deployment
metadata:
  name: web
spec:
  replicas: 3
  selector:
    matchLabels:
      app: web
  template:
    metadata:
      labels:
        app: web
    spec:
      containers:
      - name: web
        image: registry.example.com/web:2.1.0
        resources:
          requests:
            cpu: "200m"
            memory: "256Mi"
          limits:
            cpu: "500m"
            memory: "512Mi"

kubectl apply -f deployment.yaml
kubectl get deployment web

Step 3: Create a HorizontalPodAutoscaler

kubectl autoscale deployment web \
  --cpu-percent=60 \
  --min=3 \
  --max=10

kubectl get hpa web
kubectl describe hpa web

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: web
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web
  minReplicas: 3
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 60

Step 4: Load Test and Watch It Scale

# In one terminal, watch the HPA react
kubectl get hpa web --watch

# In another terminal, generate load
kubectl run load-generator --image=busybox --restart=Never -- \
  /bin/sh -c "while true; do wget -q -O- http://web; done"

kubectl get pods -l app=web
kubectl delete pod load-generator

Step 5: Right-Size Requests with a VerticalPodAutoscaler

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: web-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: web
  updatePolicy:
    updateMode: "Auto"

kubectl apply -f vpa.yaml
kubectl describe vpa web-vpa
# recommendation section shows suggested CPU/memory requests

Step 6: Scale Nodes with the Cluster Autoscaler

kubectl -n kube-system get deployment cluster-autoscaler

kubectl -n kube-system logs deployment/cluster-autoscaler \
  --tail=50 | grep -i "scale"

# Pods that cannot be scheduled trigger a scale-up
kubectl get pods --field-selector=status.phase=Pending
kubectl describe pod <pending-pod> | grep -A5 Events

Autoscaling Cheat Sheet

kubectl top nodes                          # node CPU/memory usage
kubectl top pods                           # pod CPU/memory usage
kubectl get hpa                            # list HorizontalPodAutoscalers
kubectl describe hpa <name>                # see current/target metrics
kubectl get vpa                            # list VerticalPodAutoscalers
kubectl autoscale deployment <name> \
  --cpu-percent=60 --min=2 --max=10        # quick HPA from the CLI
kubectl get pods --field-selector=status.phase=Pending  # find scale-up triggers

FAQ

Can I run HPA and VPA on the same Deployment? Not safely for CPU/memory at the same time; they can fight over the same signal. A common pattern is HPA on custom or memory metrics while VPA manages CPU/memory requests, or VPA in recommendation-only mode.

Does HPA work without the Metrics Server? No, the default CPU/memory-based HPA needs the metrics.k8s.io API that the Metrics Server provides. Custom or external metrics need a separate adapter.

What triggers the Cluster Autoscaler? Pods stuck in Pending because no Node has enough allocatable CPU or memory. It then adds Nodes from a configured node group, and removes underutilized Nodes later.

Summary

You installed the Metrics Server, created an HPA that scales replicas on CPU utilization, added a VPA to right-size resource requests, and confirmed how the Cluster Autoscaler reacts to Pending Pods by adding Nodes, all through kubectl.