How to handle a cluster with multiple kernel version.

[yujunz]

1. Issue or feature description

gpu-operator/pkg/controller/specialresource/specialresource_controls.go

Line 236 in cb3bbd9

// Assuming all nodes are running the same kernel version,

The code comments assumes all node have the same kernel version which is not true in my case (the GPU nodes have a different version to the CPU nodes).

Is there a way to handle a cluster with multiple kernel version?

2. Steps to reproduce the issue

N/A

3. Information to attach (optional if deemed irrelevant)

N/A

[yujunz]

BTW, it is not rare case to have multiple version of kernel if the cluster is upgraded several times. For example:

[yujunz]

My current workaround is modifying the DaemonSet manually.

But then pod crashed. Seems the same problem in #18 .

[RenaudWasTaken]

Hello!

This is a feature we plan to tackle pretty soon (1-2 months).

It already should work as driver containers will recompile the drivers for the current kernel.
The line you highlighted points to a feature where the driver that lands on a node is already compiled for that specific kernel.