Hi! We've been using the GPU Operator on a self-managed k8s cluster with our own custom AMIs. We are using the Containerd runtime with a custom config that is already embedded in our AMIs. That allows us to avoid having to do the Containerd setup at runtime, which seems to be a bit "brutal" for our use case.
With the migration to nvidia-ctk-installer on v1.18, it seems like it is not possible to separate the toolkit installation and the runtime setup. Would it be possible to add a flag to be able to run only the toolkit installation? We're currently using this but we would love to make it available for all nvidia-container-toolkit users: DataDog@fd7c054
(Note that I'm not opinionated at all on the flag name/implementation specifics)
Hi! We've been using the GPU Operator on a self-managed k8s cluster with our own custom AMIs. We are using the Containerd runtime with a custom config that is already embedded in our AMIs. That allows us to avoid having to do the Containerd setup at runtime, which seems to be a bit "brutal" for our use case.
With the migration to
nvidia-ctk-installeron v1.18, it seems like it is not possible to separate the toolkit installation and the runtime setup. Would it be possible to add a flag to be able to run only the toolkit installation? We're currently using this but we would love to make it available for all nvidia-container-toolkit users: DataDog@fd7c054(Note that I'm not opinionated at all on the flag name/implementation specifics)