Daemonsets OwnerRef cleanup labels

assets/gpu-feature-discovery/0210_clusterrole.yaml

@@ -11,3 +11,9 @@ rules:
   - get
   - list
   - watch
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/auth_proxy_client_clusterrole.yaml

@@ -5,3 +5,9 @@ metadata:
 rules:
 - nonResourceURLs: ["/metrics"]
   verbs: ["get"]
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/auth_proxy_role.yaml

@@ -11,3 +11,9 @@ rules:
   resources:
   - subjectaccessreviews
   verbs: ["create"]
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/clusterpolicy_editor_role.yaml

@@ -22,3 +22,9 @@ rules:
   - clusterpolicies/status
   verbs:
   - get
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/clusterpolicy_viewer_role.yaml

@@ -18,3 +18,9 @@ rules:
   - clusterpolicies/status
   verbs:
   - get
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/leader_election_role.yaml

@@ -25,3 +25,9 @@ rules:
   verbs:
   - create
   - patch
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/nvidiadriver_editor_role.yaml

@@ -22,3 +22,9 @@ rules:
   - nvidiadrivers/status
   verbs:
   - get
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

config/rbac/nvidiadriver_viewer_role.yaml

@@ -18,3 +18,9 @@ rules:
   - nvidiadrivers/status
   verbs:
   - get
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get

controllers/clusterpolicy_controller.go

@@ -254,6 +254,16 @@ func updateCRState(ctx context.Context, r *ClusterPolicyReconciler, namespacedNa
 	}
 }
 
+func isOwnedByDaemonSet(ownerRefs []metav1.OwnerReference) bool {
+
+	for _, ownerRef := range ownerRefs {
+		if ownerRef.Kind == "DaemonSet" && ownerRef.Controller != nil && *ownerRef.Controller {
+			return true
+		}
+	}
+	return false
+}
+
 func addWatchNewGPUNode(r *ClusterPolicyReconciler, c controller.Controller, mgr ctrl.Manager) error {
 	// Define a mapping from the Node object in the event to one or more
 	// ClusterPolicy objects to Reconcile
@@ -288,6 +298,15 @@ func addWatchNewGPUNode(r *ClusterPolicyReconciler, c controller.Controller, mgr
 			return hasGPULabels(labels)
 		},
 		UpdateFunc: func(e event.TypedUpdateEvent[*corev1.Node]) bool {
+			r.Log.Info("SHIVAAAAAAAA Calling labels: UpdateFunc", "labels", e.ObjectNew.GetLabels())
+
+			ownerRefs := e.ObjectNew.GetOwnerReferences()
+			if isOwnedByDaemonSet(ownerRefs) {
+				r.Log.Info("SHIVAAAAAAAA DaemonSet owner found", "ownerRefs", ownerRefs)
+				// return false
+			}
+			r.Log.Info("SHIVAAAAAAAA Calling labels: UpdateFunc", "ownerRefs", ownerRefs)
+
 			newLabels := e.ObjectNew.GetLabels()
 			oldLabels := e.ObjectOld.GetLabels()
 			nodeName := e.ObjectNew.GetName()
@@ -323,6 +342,7 @@ func addWatchNewGPUNode(r *ClusterPolicyReconciler, c controller.Controller, mgr
 					"osTreeLabelChanged", osTreeLabelChanged,
 				)
 			}
+			r.Log.Info("SHIVAAAAAAAA Calling labels: UpdateFunc", "needsUpdate", needsUpdate)
 			return needsUpdate
 		},
 		DeleteFunc: func(e event.TypedDeleteEvent[*corev1.Node]) bool {

controllers/object_controls.go

@@ -3846,6 +3846,16 @@ func isDaemonSetReady(name string, n ClusterPolicyController) gpuv1.State {
 				}
 			}
 		}
+		// Create owner ref
+		ownerReference := []metav1.OwnerReference{}
+		// Get pod owner reference
+		ownerReference = append(ownerReference, metav1.OwnerReference{
+			APIVersion: corev1.SchemeGroupVersion.String(),
+			Kind:       "Pod",
+			Name:       pod.Name,
+			UID:        pod.UID,
+		})
+		pod.OwnerReferences = ownerReference
 	}
 
 	// All containers are ready

controllers/state_manager.go

@@ -844,6 +844,7 @@ func (n *ClusterPolicyController) init(ctx context.Context, reconciler *ClusterP
 		n.logger.Info("Pod Security Admission labels added to GPU Operator namespace", "namespace", n.operatorNamespace)
 	}
 
+	n.logger.Info("SHIVAAAAAAAA Calling labels: namespace, hasNFDLabels", "namespace", n.operatorNamespace, "hasNFDLabels", n.hasNFDLabels)
 	// fetch all nodes and label gpu nodes
 	hasNFDLabels, gpuNodeCount, err := n.labelGPUNodes()
 	if err != nil {

deployments/gpu-operator/charts/node-feature-discovery/templates/clusterrole.yaml

@@ -137,4 +137,10 @@ rules:
   verbs:
   - delete
   - list
+- apiGroups:
+  - ""
+  resources:
+  - pods
+  verbs:
+  - get
 {{- end }}