I’m trying to provision a new K3s Kubernetes cluster through the Rancher UI and would like to configure it to use the NVIDIA container runtime (nvidia) as the default runtime.
In standalone K3s setups, I’m used to adding --default-runtime=nvidia to the systemd service (/etc/systemd/system/k3s.service). However, when Rancher manages the node, this change is reverted after a reboot or upgrade, making it neither scalable nor persistent.
I’ve attempted to configure the runtime via the Kubelet Arguments in the Rancher UI, but this doesn’t seem to have the intended effect.
Has anyone successfully configured Rancher-managed K3s clusters to default to the NVIDIA runtime? Any guidance or best practices would be greatly appreciated.
Thans in advance!
Thanks,
Sam