search

intel / helm-charts

Apache License 2.0
12 stars 15 forks source link

make nodeSelector unconfigured in values? #27

Closed budimanjojo closed 1 year ago

budimanjojo commented 1 year ago

Hi, the current values.yaml file in gpu-device-plugin is pre filled with intel.feature.node.kubernetes.io/gpu: 'true' and there's no way to override that because it's a mapping and it will just add the user defined one to the map. maybe make the logic to default to intel.feature.node.kubernetes.io/gpu: 'true' if the value is not defined instead?

mythi commented 1 year ago

@budimanjojo all of our *-device-plugin charts have the same limitation. Currently our thinking is we keep it like this and make NFD as the mandatory dependency (see #26). Do you have case where you'd prefer sunning the plugin without any nodeSelector values so that the plugin runs on all nodes?

budimanjojo commented 1 year ago

@mythi okay fair enough. I'm running NFD too but I didn't apply the NFD config patch so I wanted to just do nodeSelector: feature.node.kubernetes.io/pci-0300_8086.present: "true" because I only have one type of gpu. But I just applied the NFD config patch and now I'm good. I'll close this issue now.

2fst4u commented 1 year ago

@mythi okay fair enough. I'm running NFD too but I didn't apply the NFD config patch so I wanted to just do nodeSelector: feature.node.kubernetes.io/pci-0300_8086.present: "true" because I only have one type of gpu. But I just applied the NFD config patch and now I'm good. I'll close this issue now.

What patch did you need to apply? I can't find any reference to how to make this label automatically apply to nodes.

mythi commented 1 year ago

@2fst4u install GPU plugin helm chart using --set nodeFeatureRule=true