fedora-coreos as Kubernetes Node OS support

[diablinux]

I got a kubernetes cluster for dev purposes. The host OS is CEntOS 10 and I am using Qemu/KVM (libvirt) for Virtualization layer. I got a GPU GeForce RTX 3070 working; I am able to execute ML model training podman containers successfully at host level. Then I decided to enable gpu passthrough support and added the PCI device to one of my Kubernetes worker nodes, the idea? you got it, run ML jobs in k8s. I followed the steps to setup the gpu-operator via helm chart. Sadly, it doesn't work for Fedora CoreOS 43.

Here is the error:

Mar 08 03:54:43 worker02 kubelet[1235]: E0308 03:54:43.280297 1235 pod_workers.go:1324] "Error syncing pod, skipping" err="failed to \"StartContainer\" for \"nvidia-driver-ctr\" with ImagePullBackOff: \"Back-off pulling image \\\"nvcr.io/nvidia/driver:580.105.08-fedora43\\\": ErrImagePull: rpc error: code = NotFound desc = failed to pull and unpack image \\\"nvcr.io/nvidia/driver:580.105.08-fedora43\\\": failed to resolve image: nvcr.io/nvidia/driver:580.105.08-fedora43: not found\"" pod="gpu-operator/nvidia-driver-daemonset-d2hvh" podUID="8551a036-564a-484a-992b-b465e2df527b"

Instead of looking for a workaround, I would like to hear your opinion. As DevOps or SRe's what would be your Dev Stack to go while pursuing stable dev environments (k8s on prem with Virtualization):

• Host OS - ?
• K8s node OS - ?

So I can plan on rebuilding my dev environment.

I see images for Ubuntu 22.04 and 24.04, Rocky Linux, RHEL 9 - 10,

Thoughts?

[diablinux]

Instead of rebuilding the whole cluster, I just added a new worker node (Ubuntu 24.04.4 LTS)
Reinstalled the gpu-operator helm chart. All good.
It would be great to have support for Fedora CoreOS