Allow either a resourceClaimName or a deviceClassName to be specified

Signed-off-by: Kevin Klues <kklues@nvidia.com>

Author: klueska

api/nvidia.com/resource/gpu/v1alpha1/multinode.go

@@ -43,8 +43,11 @@ type MultiNodeEnvironmentList struct {
 	Items []MultiNodeEnvironment `json:"items"`
 }
 
+// +kubebuilder:validation:XValidation:rule="(has(self.resourceClaimName) ? !has(self.deviceClassName) : has(self.deviceClassName))",message="Exactly one of 'resourceClaimName' or 'deviceClassName' must be set."
+
 // MultiNodeEnvironmentSpec provides the spec for a MultiNodeEnvironment.
 type MultiNodeEnvironmentSpec struct {
 	NumNodes          int    `json:"numNodes"`
-	ResourceClaimName string `json:"resourceClaimName"`
+	ResourceClaimName string `json:"resourceClaimName,omitempty"`
+	DeviceClassName   string `json:"deviceClassName,omitempty"`
 }

cmd/nvidia-dra-controller/mnenv.go

@@ -172,13 +172,15 @@ func (m *MultiNodeEnvironmentManager) onMultiNodeEnvironmentAdd(obj any) error {
 		Controller: ptr.To(true),
 	}
 
-	dc, err := m.createDeviceClass("", ownerReference)
+	dc, err := m.createDeviceClass(mne.Spec.DeviceClassName, ownerReference)
 	if err != nil {
 		return fmt.Errorf("error creating DeviceClass '%s': %w", "<generated-name>", err)
 	}
 
-	if _, err := m.createResourceClaim(mne.Namespace, mne.Spec.ResourceClaimName, dc.Name, ownerReference); err != nil {
-		return fmt.Errorf("error creating ResourceClaim '%s/%s': %w", mne.Namespace, mne.Spec.ResourceClaimName, err)
+	if mne.Spec.ResourceClaimName != "" {
+		if _, err := m.createResourceClaim(mne.Namespace, mne.Spec.ResourceClaimName, dc.Name, ownerReference); err != nil {
+			return fmt.Errorf("error creating ResourceClaim '%s/%s': %w", mne.Namespace, mne.Spec.ResourceClaimName, err)
+		}
 	}
 
 	return nil

deployments/helm/k8s-dra-driver/crds/gpu.nvidia.com_multinodeenvironments.yaml

@@ -40,14 +40,19 @@ spec:
           spec:
             description: MultiNodeEnvironmentSpec provides the spec for a MultiNodeEnvironment.
             properties:
+              deviceClassName:
+                type: string
               numNodes:
                 type: integer
               resourceClaimName:
                 type: string
             required:
             - numNodes
-            - resourceClaimName
             type: object
+            x-kubernetes-validations:
+            - message: Exactly one of 'resourceClaimName' or 'deviceClassName' must
+                be set.
+              rule: '(has(self.resourceClaimName) ? !has(self.deviceClassName) : has(self.deviceClassName))'
         type: object
     served: true
     storage: true