This PR has a few different objectives that are primarily driven by the need to convert the taints from using nvidia-gpu-only to nvidia.com/gpu.
It appears that nvidia.com/gpu is the more common and officially recommended taint for GPU nodes. After speaking with Cory it sounds like he original chose nvidia-gpu-only because of an issue with the auto scaling that was caused by the / in the taint which has since been resolved.
Other updates include:
Updating the script to create the nodes using a patch file to set the taints instead of hard coding them into the repo. This should hopefully make it more configurable in the future if we or a customer wants to use a different taint.
Extracting the tolerations out of the base ClusterPolicy into a component to make it more easily configurable as well
Updating the Accelerator Profile to use the new taint
This PR has a few different objectives that are primarily driven by the need to convert the taints from using
nvidia-gpu-only
tonvidia.com/gpu
.It appears that
nvidia.com/gpu
is the more common and officially recommended taint for GPU nodes. After speaking with Cory it sounds like he original chosenvidia-gpu-only
because of an issue with the auto scaling that was caused by the/
in the taint which has since been resolved.Other updates include: