Release 1.7.639.0 Release · Azure/AKS-Edge

We are excited to release the July Update of AKS Edge Essentials.

⚠️ WARNING: Note that some of the features in this release are not supported in GA yet.

New Features

Kubernetes versions supported: On K8S and K3S - 1.27.6 and 1.28.3
Update from K3S 1.27.6 (Feb release) to 1.27.6 (July release), and on K8S from 1.27.6 (Feb release) to 1.27.6 (July release).
Upgrade from K3S 1.26.10 (Feb release) to 1.27.6 (July release), and on K8S from 1.26.6 (Feb release) to 1.27.6 (July release).
If you are using a ScalableCluster with the network adapter disabled/disconnected, we have added a precheck to ensure provided IP addresses for the network adapter are not in the reserved (169.254.0.0-169.254.255.255) range. If this precheck fails, you will see the error message "Please ensure the adapter is not disabled/disconnected".

Run Windows containerized workloads.
Create multi-node deployments either on a single machine or across multiple machines.
GPU-PV is in preview. You can find documentation here.

For more information feature set and benefits, see AKS Edge Essentials Overview.

When removing nodes on a multi-machine cluster, manual validation that the removed node no longer served as the leader was required, as documented here. We have added a check to test this automatically.
Get-ClusterId can be unreliable when the service is unreachable. We have added retry logic to the command to increase the success rate of this command when the service is unreachable.

When updating from 1.5.203 to 1.6.384.0 on both Linux and Windows, customers may see one of the following non-fatal errors:
- "Get-Content: Cannot find path 'C:\Windows\System32\akswindows-manifest.json' because it does not exist."
- "Get-Content: Cannot find path 'C:\Program Files\AksEdge\akswindows-manifest.json' because it does not exist."

This is a non-fatal error and can be ignored.

Updates from K8S 1.26.3 to 1.26.6 fails when disk pressure is high. As a workaround, extract the tarball /opt/container-images/container-images-tar-file.tar.gz and use ctr to reimport the images that are missing, using the command ctr -n=k8s.io images import {extracted tar files location}

New deployment fails when the execution policy is set to AllSigned (Set-ExecutionPolicy –ExecutionPolicy AllSigned). Recommend setting the policy to 'ByPass'.
Uninstalling and reinstalling k8s/k3s causes a “deployment already exists” error or the 'Virtual machine creation failed' error, when creating a new deployment, so recommend restarting your machine before deployment .
Event logging fails when uninstalling and reinstalling an MSI without a reboot in between . Particularly true when switching between using K3S and K8S.
When a deployment fails due to lack of memory/disk space, it creates an incomplete VHDX file that causes errors in later deployments .
Deployment with External virtual switch fails when using Windows RS5.

When you use the Remove-AksEdgeNode to remove a node, it might still show up in the kubectl get nodes as the cluster tries to reconcile the state of the present nodes. Please note that the command Remove-AksEdgeNode is a asynchronous call and will return before the cluster state is updated to reflect the new state.

On a K3s Windows node, load balancer klipper-lb pulls images from the internet when connected .
Adding a Windows node takes around ~10 minutes to complete .
No storage provisioner is available for the Windows node .
NFS subdir external provisioner pod fails on Windows nodes.