This message was deleted.

Don't use the Nvidia operator to set the config template? Rke2 already detects runtimes and adds sections for them automatically.

@creamy-pencil-82913 great thanks for your help. I am confusing about this because rke2 official documentation says people should modify config template rather config.toml to change containerd behavior. would you mind sharing me a documentation about how to set up gpu operator for rke2, because I am following the advice from nvidia official documentation on rke2 and they told me to do in that way. besides, I have found an issue talking about this, https://github.com/NVIDIA/gpu-operator/issues/623 which also gives some advices on how to set up config.toml.tmpl on rke2, and they mentioned an issue of rke2 about the conflict.

What specifically do you need to modify the containerd config for? The only reason you'd need to do that is to change the default runtime class, which isn't necessary. Just install the runtime and set the runtimeClassName on your GPU pods.

@creamy-pencil-82913 the truth is that

gpu-operator

is going to change the default runtime class by modifying the containerd config for me. so I could not control the behavior of

gpu-operator

and he does that thing which would make conflicts.

surely you can modify the operator config so that it just drops the runtime without changing the containerd template