We earn commissions when you shop through affiliate links on this site, at no extra cost to you. Our guides and technical recommendations are written independently of these relationships: a commission never decides what we recommend, how we diagnose a problem or what we publish, and the content does not depend on it. Affiliate disclosure

News · Published 2026-11-03

AKS Azure Linux 2.0 Deadline Passed: Move to 3.0

AKS removes Azure Linux 2.0 node images from 31 October 2026 and blocks scaling. How to find the pools and move them to 3.0. A heads-up, not tested.

TL;DR

  • Azure Linux 2.0 on AKS is out of support. Support and security updates ended on 30 November 2025, and the node image is frozen at release 202512.06.0.
  • 31 October 2026 was the removal date. Microsoft's documentation says that from that date the node images are removed and you can no longer scale node pools that still use them.
  • Ready nodes prove nothing here. Existing nodes keep running. What breaks is adding, replacing and reimaging nodes, which is exactly what you need during an incident.
  • Two documented ways out: upgrade the node pool to Kubernetes 1.32 or later with the AzureLinux OS SKU (Azure Linux 3.0 is the default there), or set the OS SKU of the existing pool to AzureLinux3.
  • The OS SKU change reimages the pool. Treat it as a node operation: check disruption budgets and capacity, and try it on a non-production pool first.
  • Terraform needs a plan review. The AzureRM provider documentation says changing os_sku between AzureLinux and Ubuntu does not replace the node pool, and that any other change forces a new one. A move from AzureLinux to AzureLinux3 is "any other change".
  • Not verified: that the removal has actually happened. Microsoft's page still words it in the future tense. See the last section.

How this was checked

The facts below were read on 3 November 2026 from these pages: Node images in AKS (last updated by Microsoft on 20 October 2026), Upgrade OS versions in AKS, Use Azure Linux on AKS, the AKS retirement issue and the AzureRM provider documentation for azurerm_kubernetes_cluster_node_pool.

Nothing was run. There was no AKS cluster, so every command here is the documented form, not output we saw. Sentences marked our reasoning are inferences from the documentation, not statements Microsoft makes.

What Microsoft says

DateWhat happens to Azure Linux 2.0 on AKS
30 November 2025AKS stops supporting it and stops providing security updates. The node image is frozen at 202512.06.0.
31 October 2026Node images are removed, and you can no longer scale node pools that use them.

The same page also carries a general warning: older node images "can contain unpatched security vulnerabilities and might not work properly with recently released features", can lead to "issues with scaling, node readiness, and security", and can place the cluster outside the support scope.

Microsoft's stated way out is to "migrate to a supported Azure Linux version by upgrading your node pools to a supported Kubernetes version or migrating to osSku AzureLinux3".

What it means for a running cluster

  • Existing nodes are not switched off. The documentation talks about scaling, not about stopping running nodes. We have not seen a cluster on the removal date.
  • Our reasoning: anything that has to create a node from the image is exposed. That includes the cluster autoscaler adding capacity, a manual scale-out and replacing a node that failed. A pool that looks healthy can still be unable to heal.
  • Our reasoning: a node image upgrade does not help. The image is frozen, so there is no newer Azure Linux 2.0 image to move to. The pool has to change OS version.

Find the pools that are still on Azure Linux 2.0

List every node pool in each cluster you run. This is the documented command with a query added; it only reads.

# Read-only: lists the node pools of one cluster.
az aks nodepool list \
  --resource-group "$RG" \
  --cluster-name "$AKS" \
  --query "[].{name:name,osSku:osSku,k8s:orchestratorVersion,nodeImage:nodeImageVersion,mode:mode,count:count}" \
  --output table

Read the result like this:

  • osSku is AzureLinux3: Azure Linux 3.0, nothing to do.
  • osSku is AzureLinux: the version depends on Kubernetes. Microsoft's table says Azure Linux 2.0 is the default for Kubernetes 1.27 to 1.31 and Azure Linux 3.0 is the default for 1.32 and later. One exception: on 1.31 with the AzureLinuxV3Preview feature flag enabled, AzureLinux already means 3.0.
  • nodeImage: a pool still on Azure Linux 2.0 is on the frozen 202512.06.0 release, so the version string is a second signal. Treat the AKS metadata as the authority and kubectl get nodes -o wide as a cross-check.

Do this for every cluster in every subscription. A cluster you forgot about is the one that will page you.

The two documented ways out

1. Upgrade Kubernetes. With the default AzureLinux OS SKU you get the default Azure Linux version for your Kubernetes version, so upgrading the node pool to 1.32 or later moves it to Azure Linux 3.0. No OS change is needed on your side. The cost is that you are doing a Kubernetes upgrade, with everything that comes with one.

2. Change the OS SKU on the pool. AzureLinux3 is supported on Kubernetes 1.28 to 1.36, and Microsoft says you can use it "to migrate from Azure Linux 2.0 to Azure Linux 3.0" without upgrading Kubernetes. The documented form of the command is:

# Not run here. Reimages the nodes in the pool.
az aks nodepool update \
  --resource-group "$RG" \
  --cluster-name "$AKS" \
  --name "$POOL" \
  --os-sku AzureLinux3

Two cautions from the same page. Updating the OS SKU "might trigger an automatic reimage", and the command "might fail if the target OS doesn't have a supported node image for your Kubernetes version, VM size, or FIPS enablement". Check the combination before you schedule anything.

A versioned SKU like AzureLinux3 also has a cost later: Microsoft says that with a versioned SKU you "need to manually migrate to a new OS version to avoid blocked Kubernetes upgrades". The default AzureLinux SKU follows the Kubernetes version for you. The same documentation describes rolling back to the default SKU with the same az aks nodepool update command, so the choice is reversible.

A third route, our suggestion and not from Microsoft's page: add a new pool with --os-sku AzureLinux3, move the workloads over, and delete the old pool. It costs extra capacity for a while and gives you a clean rollback boundary. For a system pool, keep a valid system pool at every step.

Before you touch a pool

This list is ours. It is the usual checklist for anything that drains nodes.

  • Look at PodDisruptionBudgets (kubectl get pdb -A). A strict budget can stop a drain and leave the operation waiting.
  • Check that the rest of the cluster, or a surge node, can hold the evicted Pods. Check compute quota and, with Azure CNI, free IP addresses in the subnet.
  • Check node selectors, taints and tolerations that tie workloads to this specific pool.
  • Do a non-production pool first, with the same VM size, Kubernetes version and features (FIPS, Arm64, Trusted Launch).
  • Do not confuse AzureLinux3 with Azure Container Linux (AzureContainerLinux). They are different OS SKUs. Azure Container Linux needs Kubernetes 1.34 or later and Trusted Launch.

Terraform

If Terraform owns the node pool, a change made in the portal or the CLI is drift, and the next plan will try to put it back. Worse, the plan may replace the pool. The AzureRM documentation lists AzureLinux3 among the allowed os_sku values and says: "Changing between AzureLinux and Ubuntu does not replace the resource; any other change forces a new resource to be created."

So for a pool defined like this:

# Example, not run: the one line that changes.
resource "azurerm_kubernetes_cluster_node_pool" "user" {
  # ...
  os_sku = "AzureLinux3" # was "AzureLinux"
}

expect terraform plan to show a replacement, not an update. We did not run it, and the provider may behave differently in the version your repository pins, so read the plan. Do not apply a plan that destroys a production pool without a decision. If it plans a replacement, the parallel-pool route above is the safer way to get the same result: add the new pool, move the workloads, then remove the old pool in a separate change.

Verify afterwards

Run the same az aks nodepool list command and check three things: osSku is AzureLinux3 or the cluster is on 1.32 or later with AzureLinux, nodeImage is no longer the frozen release, and every node is Ready. Then check that the workloads are healthy at the application level. provisioningState: Succeeded and a green kubectl get nodes are not proof that the application survived the reimage.

Common traps

  • "The nodes are Ready, so we are fine." Ready says the node works today. It says nothing about whether the pool can add a node tomorrow.
  • "The autoscaler will handle it." Our reasoning: the autoscaler asks the pool for a node. If the pool cannot create one, asking harder does not help.
  • "We will do a node image upgrade." The image is frozen. There is nothing newer to upgrade to.
  • "Terraform will sort it out." It will do what the plan says, which for this change is probably replace the pool.

Also on the AKS calendar

The same Microsoft pages list other dates that are not far away. These are from the page as read on 3 November 2026.

DateWhatWhat Microsoft says to do
10 December 2026Azure Linux with OS Guard (preview): support ends, no new pools. Images are removed on 10 March 2027.Migrate to Azure Container Linux.
17 March 2027Ubuntu 20.04: support ends and node images are deleted, so those pools cannot scale.Upgrade node pools to Kubernetes 1.35 or later.
30 June 2027Ubuntu 22.04: support ends. Images are removed on 30 April 2028.Move to Ubuntu 24.04 or later.

What we have not verified

  • That the removal has happened. The page we read still says images "will be removed", and it was last updated on 20 October 2026. The date has passed, but we did not confirm the result on a cluster. If you have a pool on Azure Linux 2.0, try a small scale-out on a non-production pool and see what AKS says.
  • How exactly a failure shows up: which error, which operation, whether the cluster autoscaler reports it or fails quietly.
  • What the Terraform plan shows for your provider version.
  • Whether a node's osImage string reliably tells Azure Linux 2.0 from 3.0. We did not rely on it.

If you run one of these and see something different, tell us at info@opsbreakfix.com and we will correct the page.

Related guides

Sources

More from the blog

This article is educational information, not incident-response or legal advice. Verify package status with your operating system, hosting, or runtime vendor.