MidKubernetes

What are the ways to autoscale in Kubernetes (HPA, VPA, Cluster Autoscaler), and how do they fit together?

What they are really testing: Whether you distinguish scaling pods, scaling pod resources, and scaling nodes, and know they work at different layers. Confusing them leads to scaling that does not actually help.

A real interview question

What are the ways to autoscale in Kubernetes (HPA, VPA, Cluster Autoscaler), and how do they fit together?

What most people say

drag me

You use the Horizontal Pod Autoscaler to add more pods when CPU is high.

HPA alone is not the full picture. If the cluster has no spare capacity, more pods just go Pending, so you also need node scaling. It also misses VPA for right-sizing.

The follow-ups they ask next

  • You set up HPA but under load new pods stay Pending. Why?

    No node capacity: HPA created replicas but the scheduler has nowhere to place them. You need the Cluster Autoscaler (or Karpenter) to add nodes. HPA scales pods, not the cluster.

  • Why should you avoid running HPA and VPA on the same CPU metric?

    They conflict: VPA changes requests while HPA scales on utilization against those requests, so they oscillate. Use VPA for right-sizing and HPA for load, on different signals, or VPA in recommendation mode.

What the interviewer is listening for

  • Distinguishes pods (HPA), pod-size (VPA), nodes (Cluster Autoscaler)
  • Knows HPA pods go Pending without node scaling
  • Knows HPA+VPA conflict; mentions Karpenter

What sinks the answer

  • Only knows HPA
  • Thinks HPA adds nodes
  • Runs HPA and VPA on the same metric

If you genuinely do not know

Say this instead of freezing. Reasoning out loud from what you do know beats silence every single time, and a good interviewer is listening for exactly that.

Three layers: [HPA scales pod replicas on CPU/memory/custom metrics], [VPA right-sizes a pod requests/limits], [Cluster Autoscaler (or Karpenter) adds/removes nodes]. They compose: [HPA adds pods, the Cluster Autoscaler adds nodes when there is no room]. Avoid [HPA and VPA on the same metric, they fight].

Keep going with kubernetes

All 336 cloud engineer questions

Knowing the answer is not the same as recalling it under pressure

Sign in to send the questions you fumble to spaced recall, so they come back right before you would forget them, and learn the concepts behind them with hands-on labs.

Start free