Updating node group OS

Starting with Kubernetes version 1.30, the Managed Service for Kubernetes cluster node OS changed from Ubuntu 20.04 to Ubuntu 22.04 in all release channels.

When you update node groups within those clusters, new nodes are automatically created from an Ubuntu 22.04 VM image.

User resource updatesUser resource updates

In Ubuntu 22.04, system libraries and Linux kernel headers were updated, so GPU driver compilation may not work for node groups with custom GPU drivers.

How the problem manifests itself:

• You get driver build errors.
• GPU cannot be detected.
• Node group update fails.

To avoid all this, make sure your GPU Operator and driver versions are compatible when preparing for the update:

• Update GPU Operator to version 24.9.x+.
• Update your driver to version 550.144.03 or higher.
• Use precompiled drivers. Do it by setting --driver.usePrecompiled=true when installing GPU Operator.

For more information on using a GPU with a custom driver, see Using node groups with GPUs without pre-installed drivers.

Preparation for migrationPreparation for migration

Before migrating your Kubernetes cluster to a new OS version, test the update on the new cluster:

1. Create a Managed Service for Kubernetes cluster.

2. Create a node group in the cluster.

3. In the new cluster, test your apps, which may prove OS version-dependent.

   Check key load indicators:

   • GPU load.
   • App status monitoring.
   • The functioning of monitoring agents and drivers.

How to check node OS versionHow to check node OS version

All nodes in the group use the same basic OS version image. You can check the OS version using these commands:

kubectl get nodes -o jsonpath='{range .items[*]}{.metadata.name}{"\t"}{.status.nodeInfo.osImage}{"\n"}{end}'

kubectl get node <node-name> -o jsonpath='{.status.nodeInfo.osImage}{"\n"}'