A glowing Kubernetes diagram shows containerized workloads distributed across connected server platforms in a futuristic lab.
Microsoft has listed a "Pod snapshot cluster extension" as a public preview for Azure Kubernetes Service (AKS). The Azure Updates entry itself is thin. The public listing page fetched for this story rendered no item body, so the title and preview status are all it confirms. Azure's status definitions say "In preview" means available to all Azure customers for non-production use and testing. Treat this as something to evaluate in a lab, not something to put under production workloads.

Public engineering artifacts and Microsoft's troubleshooting documentation fill in more than the announcement does. They are summarized below. Where they are silent, I say so.

A glowing Kubernetes diagram shows containerized workloads distributed across connected server platforms in a futuristic lab. What the public evidence shows​

It is an AKS cluster extension with the type name Microsoft.PodSnapshot. Microsoft Learn has a preview troubleshooting article for Pod snapshot cluster extension errors (Microsoft.PodSnapshot) in Azure Kubernetes Service. That article, not the announcement, is the first public Microsoft description of how the extension behaves.

It has a cluster-level API profile. Microsoft's managed-cluster reference lists a Pod Snapshot Restore profile with an enabled flag. It says this controls whether the Pod Snapshot Restore core extension is enabled for the managed cluster. A pull request in the Azure REST API specs repository adds managedCluster.properties.podSnapshotProfile.enabled to the 2026-08-02-preview API version. Its example uses a system-assigned managed identity and Kubernetes 1.35.6. That is an example in an API spec, not a stated support matrix.

There are two management paths in tooling. A pull request to the aks-preview Azure CLI extension describes direct Pod Snapshot Restore management through the existing aks extension wrapper, using k8s-extension rather than the AKS managed profile. The commands it lists include az aks extension create --extension-type Microsoft.PodSnapshot --config-file psr-config.json, plus update, show, list and delete. That PR was open when I looked, so syntax and availability may change before it ships. Check the current az aks extension help output rather than copying commands from a pull request.

How restore appears to work​

The troubleshooting article gives a partial picture of the restore workflow. Everything below comes from that page's error guidance and is subject to change during preview.

  • Namespace: the extension's default namespace is pod-snapshot-system. If you find no extension entry, the guidance is to confirm that the extension was installed on the intended cluster.
  • RuntimeClass: The Pod snapshot cluster extension doesn't create or default a workload's RuntimeClass. You must set that yourself.
  • Restore pods: these are marked with Kubernetes annotations. The documented ones are podsnapshot.aks.azure.com/restore: "true" and podsnapshot.aks.azure.com/restore-from: <snapshot-name>.
  • Admission denials: a restore pod can be rejected at admission. The listed causes are that the caller lacks the custom restore permission, the selected snapshot isn't usable, or the restore pod doesn't satisfy compatibility checks.
  • Permissions: The extension provides the pod-snapshot-restorer ClusterRole but doesn't create a restore binding by default. Expect to create the RBAC binding yourself.
  • Scheduling gate: a restore pod may sit in SchedulingGated. That means it carries the podsnapshot.aks.azure.com/awaiting-restore gate. The extension removes the gate after preparation succeeds.
  • Failure behavior: A failed preparation doesn't cause the Pod snapshot cluster extension to start the workload from scratch. A stuck gated pod is therefore not a silent fresh start.
  • Kata: for Kata, the extension uses a ConfirmRestore node operation to check the destination runtime. This suggests the destination runtime matters for restore, though the page excerpt does not spell out the full compatibility rules.

The troubleshooting page also mentions an Agent Sandbox cluster extension restore workflow as an alternative path. The CLI pull requests show a separate Agent Sandbox extension (Microsoft.AgentSandbox) being added alongside Pod Snapshot. The two appear to be related, but the evidence I have does not describe how.

What this is not​

The name invites assumptions, so here are the boundaries.

  • It is not documented as a replacement for Azure Backup for AKS. Azure Backup for AKS has its own Backup extension. Microsoft says it backs up cluster resources and persistent volumes attached to the cluster.
  • It is not node pool snapshots. That older feature captures node pool configuration. Microsoft describes it as holding the node image version, kubernetes version, OS type, and OS SKU. It has nothing to do with snapshotting running pods.
  • It is not a volume snapshot feature. Azure Container Storage has its own volume snapshot and restore flow for persistent volume claims.

I could not find official text stating exactly what state a pod snapshot captures, such as memory, process state or filesystem. I also found no stated limits on regions, pricing, Kubernetes versions or supported runtimes. Do not assume it provides checkpoint-and-resume of arbitrary workloads until Microsoft documents that.

A cautious evaluation plan​

  1. Use a disposable cluster. Preview means non-production, and the extension adds a namespace, a ClusterRole and admission behavior to the cluster.
  2. Confirm cluster prerequisites. Microsoft's general extension documentation says clusters must use managed identity, because cluster extensions don't work with service principal-based clusters. It also notes that Azure Policy rules or resource locks that block the cluster-extensions resource type can interfere with management.
  3. Check the current CLI help. Run az aks extension help in your own environment, with the aks-preview extension if needed, to see the supported extension types and parameters.
  4. Plan RBAC. Because no restore binding is created by default, decide who is allowed to restore before you test.
  5. Set RuntimeClass deliberately. The extension will not choose one for your workload.
  6. When something fails, inspect the pod-snapshot-system namespace and the restore pod's events and gates before assuming a bug. The documented failure modes are mostly permissions, snapshot usability, compatibility and runtime checks.

Analysis​

Extensions are Microsoft's route for delivering AKS features outside the core resource provider. The docs say an extension is added as part of a separate resource provider in the Azure API, whereas add-ons live inside the AKS provider. Pod snapshot and restore arriving as an extension fits that model. It also appears to be paired with the new Agent Sandbox work, which hints at sandboxed, restorable workloads, though that is my inference from the tooling and not a Microsoft claim.

The caveat is the gap between the headline and the documentation. A public preview with a blank announcement page, API-spec examples and a troubleshooting article is enough to start experimenting. It is not enough to plan an architecture around. Watch for Microsoft publishing a fuller overview that defines what is captured and the supported runtimes and versions.

 

References

  1. (In preview) Public Preview: Pod snapshot cluster extension Azure Updates 2026-11-17T20:00:32Z
  2. (aks-preview) Add direct Pod Snapshot Restore extension commands by alimaazamat · Pull Request #10518 · Azure/azure-cli-extensions · GitHub github.com
  3. Add Pod Snapshot Restore profile to 2026-08 preview by mboersma · Pull Request #47001 · Azure/azure-rest-api-specs · GitHub github.com