How to evaluate if the nvidia driver is available on the node

[mgazz]

With #14 we introduced a direct dependency to the nvidia-gpu-operator

Dependabot flagged the dependency for potential vulnerability:

• https://github.com/CoHDI/composable-resource-operator/security/dependabot/10
• https://github.com/CoHDI/composable-resource-operator/security/dependabot/11

The goal of this issue is to discuss how to proceed.

I looked at the source code and the operator checks for the nvidia gpu operator cluster policy to know if the driver is enabled.

• https://github.com/CoHDI/composable-resource-operator/blob/main/internal/utils/gpus.go#L1228

Here some of the calls:

• https://github.com/CoHDI/composable-resource-operator/blob/main/internal/utils/gpus.go#L91
• https://github.com/CoHDI/composable-resource-operator/blob/main/internal/utils/gpus.go#L192

While checking the cluster policy gives us hints on the gpu-operator configuration, this does not ensure that the driver is actually installed. Errors during the installation might translate in the driver not being present.

Question: Should we verify if the driver is enabled by inspecting the current status of the node?
For instance, files like /proc/driver/nvidia/version or /proc/driver/nvidia/version can give us insight on the current state. As a bonus point we also remove the direct dependency to the nvidia gpu-operator