helm: schedule CRD hook Jobs on Linux nodes

Add the operator nodeSelector to the upgrade-crd and cleanup-crd hook
Jobs. When it is unset, default to kubernetes.io/os: linux so these
Linux-only Jobs cannot be scheduled onto untainted Windows nodes in
mixed-OS clusters.

Add the default nodeSelector value to values.yaml.

Fixes: #2608
Co-requested-by: Tariq Ibrahim <tariq1890>
Signed-off-by: Billard <82095453+iacker@users.noreply.github.com>
Signed-off-by: Rahul Sharma <rahulsharm@nvidia.com>

Author: iacker

deployments/gpu-operator/templates/cleanup_crd.yaml

@@ -30,6 +30,8 @@ spec:
       tolerations:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      nodeSelector:
+        {{- toYaml .Values.operator.nodeSelector | nindent 8 }}
       containers:
         - name: cleanup-crd
           image: {{ include "gpu-operator.fullimage" . }}

deployments/gpu-operator/templates/upgrade_crd.yaml

@@ -79,6 +79,8 @@ spec:
       tolerations:
         {{- toYaml . | nindent 8 }}
       {{- end }}
+      nodeSelector:
+        {{- toYaml .Values.operator.nodeSelector | nindent 8 }}
       containers:
         - name: upgrade-crd
           image: {{ include "gpu-operator.fullimage" . }}

deployments/gpu-operator/values.yaml

@@ -82,6 +82,8 @@ operator:
   env: []
   priorityClassName: system-node-critical
   runtimeClass: nvidia
+  nodeSelector:
+    kubernetes.io/os: linux
   use_ocp_driver_toolkit: false
   # cleanup CRD on chart un-install
   cleanupCRD: false