Nvidia Drivers Installer

Important Note: The repository is currently under development and not yet ready for production use

• nvidia-driver-installer installs a DaemonSet which runs a pod on all the nodes labeled node-role.kubernetes.io/ai-worker=true. The pod on each node runs a privileged container that runs a script to install the drivers on the node.

• The script after installing the drivers writes a file /host/var/run/reboot-needed which kured looks for to trigger the node reboot.

Usage Guide

Prerequisites

Before proceeding, ensure you have the following:

1. A running Kubernetes cluster with nodes based on SLES and/or SLE Micro.
2. Access to the cluster's kubeconfig file.
3. The Helm CLI installed.

---

Installation Steps

1. Label GPU Nodes

Label the nodes with GPUs using the node-role.kubernetes.io/ai-worker=true label. The following command targets all worker nodes that are not control plane nodes:

kubectl get nodes -l node-role.kubernetes.io/worker=true -o json \
  | jq -r '.items[] | select(.metadata.labels["node-role.kubernetes.io/control-plane"] != "true") | .metadata.name' \
  | sed 's/\r//' \
  | xargs -r -I{} kubectl label node {} node-role.kubernetes.io/ai-worker=true --overwrite

Note: Adjust the label logic if your cluster has different roles or labeling conventions.

Note: The changes applied by the daemonset to the host are not reversible when you remove the label. You need to manually perform the cleanup if required.

---

2. Configure values.yaml

Create or update a values.yaml file with your desired configuration. You can define either or both sles and slem sections under os, based on the operating systems present in your cluster.

namespace: kube-system

nvidia:
  image: ghcr.io/suse/nvidia-driver-installer:0.1.0
  nodeSelector:
    node-role.kubernetes.io/ai-worker: "true"

os:
  sles:
    regcode: "VALID-SCC-REG-CODE"
    regemail: "VALID-SCC-EMAIL"
    driverVersion: "570.144"
  slem:
    regcode: "VALID-SCC-REG-CODE"
    regemail: "VALID-SCC-EMAIL"
    driverVersion: "550.54.14"

kured:
  image: ghcr.io/kubereboot/kured:1.17.1
  updatePeriod: "1m"
  rebootSentinel: "/sentinel/reboot-needed"

---

3. Install the Helm Chart

Deploy the chart using the Helm CLI. If your current Kubernetes context is set appropriately, you may omit the --kubeconfig flag.

helm install nvidia-driver-installer oci://ghcr.io/suse/chart/nvidia-driver-installer \
  --version 0.1.0 \
  -f values.yaml \
  --kubeconfig /path/to/your/kubeconfig