[Issue]: need a way to overwrite node_selector labels

[rptaylor]

Problem Description

There are default node_selector labels: https://github.com/ROCm/k8s-device-plugin/blob/master/helm/amd-gpu/values.yaml#L35

In my environment (maybe my NFD is configured differently, not sure) I need feature.node.kubernetes.io/pci-1002.present instead of feature.node.kubernetes.io/pci-0300_1002.present.

If I set this in my values file, it is the normal behaviour of Helm to combine all the keys together, resulting in both labels being applied so the daemonset doesn't run anywhere.

This line 35 has been there for awhile so it is not related to any recent change. Unless feature.node.kubernetes.io/pci-0300_1002.present is actually incorrect for default NFD installations (?) it should remain unchanged. Instead there will need to be some logic that overwrites node_selector with the user values instead of appending them.

Operating System

N/A

CPU

N/A

GPU

AMD Instinct MI210

ROCm Version

ROCm 6.0.0

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[danieljkemp]

I can confirm that this is still an issue, as I have just encountered it.

I was wondering why the pods weren't running then saw

node_selector:
  feature.node.kubernetes.io/pci-1002.present: "true"
  feature.node.kubernetes.io/pci-0300_1002.present: "true"

in the pod specs.

If you set

node_selector:
  feature.node.kubernetes.io/pci-1002.present: "true"
  feature.node.kubernetes.io/pci-0300_1002.present: null

you'll get the desired result.