World map

 

BM

World K5: Lunch Rush

VPA and node autoscaling

You will learn: when a tray needs a bigger plate, and when the food court needs another stall.

Checked with Kubernetes 1.37 on 31 Oct 2026

The problem

Your database tray keeps running out of memory. You can’t just copy a database. More trays won’t help. What will?

Watch

A VerticalPodAutoscaler (VPA): Gives a tray a bigger or smaller plate: it changes requests from real use. An add-on. (VPA) changes the plate size: the tray’s requests. It is an add-on, not built in. Put its loop in order.

First step at the top. Put these in order. Use the arrows, or drag.

  • Give the tray that size
  • Watch what the tray really uses
  • Suggest a better request size

Try it

Three helpers. Which one fits each problem?

Drag a card into a box. Or tap the card, then tap the box.

        Try it

        Fit the trays onto the stalls. One will not fit anywhere. Trays waiting like that wake a node autoscaler: Opens a new stall when trays have nowhere to sit, and closes empty ones. Cluster Autoscaler and Karpenter are two.. It opens a new stall.

        Tap a tray, then tap a stall. Or drag the tray.

        Waiting trays

        • stall-a
          CPU 0 of 2
          Memory 0Mi of 4Gi
          • stall-b
            CPU 0 of 2
            Memory 0Mi of 4Gi
            • Pending

              Trays that fit on no stall wait here.

              Check

              How does a node autoscaler decide to add a stall?

              Check

              True or not? Sort each sentence.

              Drag a card into a box. Or tap the card, then tap the box.

                  Predict

                  You turn on an HPA and a VPA, both on CPU, for the same Deployment. What happens?

                  You did it!

                  You can now choose: more trays, bigger plates, or more stalls.

                  Go deeper (optional)

                  Since Kubernetes 1.35, you can resize a running pod’s CPU and memory in place. VPA’s InPlaceOrRecreate mode uses this. If a resize can’t happen, VPA replaces the pod. For queues and events, look at KEDA, a CNCF project. Karpenter can pick machine sizes itself. Cluster Autoscaler usually grows machine groups you set up.