World map

 

BM

World K5: Lunch Rush

Cluster upgrades and version skew

You will learn: how far apart a cluster's parts may be, and the safe order to upgrade them.

Checked with Kubernetes 1.37 on 31 Oct 2026

The problem

The Kubernetes project patches only the three newest minor versions: 1.37, 1.36 and 1.35. Your old cluster runs 1.34. What is the risk?

Watch

You are here: cluster kind-dev namespace default

Your kubectl, the API server and each stall’s kubelet all have versions. version skew: How far apart the versions of a cluster's parts may be, and still work together. is how far apart they may be and still work together.

Try it

The API server runs 1.37. Which parts are close enough?

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

      Watch

      So the API server goes first, and stalls follow. Put a minor upgrade in order.

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

      • Uncordon it, then do the next stall
      • Drain one stall
      • Read the release notes for removed APIs
      • Upgrade that stall’s kubelet
      • Upgrade the control plane, API server first

      Predict

      Your control plane runs 1.35. Can you upgrade it straight to 1.37?

      Try it

      Practise the stall routine. Drain dev-worker. Pretend you upgrade its kubelet. Then uncordon it.

      You are here: cluster kind-dev namespace default
      dev-control-planeOpenManager

      No trays here.

      dev-workerOpen
      • shop·b7f55
      • shop·jwjsl
      dev-worker2Open
      • shop·fsjkb
      Time: +0s (sped up)

      Check

      Good habit, or needs work?

      Swipe right for good, left for needs work. Or use the buttons.

      1 / 4

      Check for removed APIs before the upgrade

      You did it!

      You can now upgrade a cluster in a safe order, inside the skew rules.

      Go deeper (optional)

      Each minor gets about a year of patches. Managed clouds can upgrade the control plane for you. You still check your YAML first. Self-built clusters use tools like kubeadm upgrade. In-place minor kubelet upgrades are not supported: drain the stall first.