Mismatch between kubevirt nvidia device plugin and kubevirt

[Y D]

Hello,

It seems that there is a mismatch between:

   1) the environment variables that encode the name of the resource (link), and

   2) the environment variable that kubevirt expects (link).

I got this error when I create a VM:

{"component":"virt-launcher","level":"warning","msg":"PCI_RESOURCE_NVIDIA_COM_GV100GL_TESLA_V100_SXM2_32GB not set for resource nvidia.com/GV100GL_Tesla_V100_SXM2_32GB","pos":"addresspool.go:50","timestamp":"2022-01-19T10:53:40.248215Z"}

The env set by the device plugin is a mix of upper and lower cases:

"2022/01/19 10:38:31 Allocated devices map[PCI_RESOURCE_NVIDIA_COM_GV100GL_Tesla_V100_SXM2_32GB:0000:62:00.0]",

whereas the one expected by kubevirt is upper case only:

"PCI_RESOURCE_NVIDIA_COM_GV100GL_TESLA_V100_SXM2_32GB".

Does this seem to be a bug, either in the device plugin or kubevirt?

Thanks

[Ryan Hallisey US]

Looks like a bug in the device plugin.  What version of the gpu device plugin were you using?

[Y D]

It is v1.1.0.

https://github.com/NVIDIA/kubevirt-gpu-device-plugin/blob/master/manifests/nvidia-kubevirt-gpu-device-plugin.yaml#L27

[Ryan Hallisey US]

Fix is being tracked in this issue: https://github.com/NVIDIA/kubevirt-gpu-device-plugin/issues/31

[Y D]

Thanks!

[Ryan Hallisey US]

Closed issue #31 with https://github.com/NVIDIA/kubevirt-gpu-device-plugin/releases/tag/v1.1.1