Kubernetes Autoscaling: HPA, VPA and Cluster Autoscaler

Right-sizing by hand wastes money and pages you at 3am. Kubernetes has three autoscalers that handle load and capacity automatically.

Three layers of scaling

HPA scales pods on metrics, VPA right-sizes pod requests, and the Cluster Autoscaler adds or removes nodes as demand changes.

A basic HPA

Scale a deployment between 2 and 10 pods on CPU.

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
spec:
  scaleTargetRef: { kind: Deployment, name: api }
  minReplicas: 2
  maxReplicas: 10
  metrics:
    - type: Resource
      resource: { name: cpu, target: { type: Utilization, averageUtilization: 70 } }

Tune for cost and stability

Set sane min and max replicas, use custom metrics for real load, and let the Cluster Autoscaler keep nodes lean when traffic drops.

Wrapping Up

Layered autoscaling keeps services responsive under spikes and your cloud bill honest when traffic falls.