Enhance cluster deletion logic and index registration

- Move index registration to setupControllers for early availability
- Add fallback mechanism in cluster deletion when index queries fail
- Improve system robustness and reliability

Author: 止鹜

cmd/controller-manager/app/controllermanager.go

@@ -267,15 +267,6 @@ func startClusterController(ctx controllerscontext.Context) (enabled bool, err e
 	mgr := ctx.Mgr
 	opts := ctx.Opts
 
-	// Indexes are added to help the cluster-controller and TaintManager quickly locate ResourceBinding
-	// and ClusterResourceBinding resources associated with a given cluster when eviction is needed.
-	if err := indexregistry.RegisterResourceBindingIndexByFieldCluster(ctx.Context, mgr); err != nil {
-		return false, err
-	}
-	if err := indexregistry.RegisterClusterResourceBindingIndexByFieldCluster(ctx.Context, mgr); err != nil {
-		return false, err
-	}
-
 	clusterController := &cluster.Controller{
 		Client:                    mgr.GetClient(),
 		EventRecorder:             mgr.GetEventRecorderFor(cluster.ControllerName),
@@ -937,6 +928,17 @@ func setupControllers(ctx context.Context, mgr controllerruntime.Manager, opts *
 		ClusterClientOption:         clusterClientOption,
 	}
 
+	// Register indexes that are used by multiple controllers to ensure they are available
+	// before any controller that depends on them starts.
+	// These indexes help cluster-controller and TaintManager quickly locate ResourceBinding
+	// and ClusterResourceBinding resources associated with a given cluster.
+	if err := indexregistry.RegisterResourceBindingIndexByFieldCluster(ctx, mgr); err != nil {
+		klog.Fatalf("Failed to register ResourceBinding index: %v", err)
+	}
+	if err := indexregistry.RegisterClusterResourceBindingIndexByFieldCluster(ctx, mgr); err != nil {
+		klog.Fatalf("Failed to register ClusterResourceBinding index: %v", err)
+	}
+
 	if err := controllers.StartControllers(controllerContext, controllersDisabledByDefault); err != nil {
 		klog.Fatalf("error starting controllers: %v", err)
 	}

pkg/controllers/cluster/cluster_controller.go

@@ -255,23 +255,63 @@ func (c *Controller) isTargetClusterRemoved(ctx context.Context, cluster *cluste
 	if err := c.List(ctx, rbList, client.MatchingFieldsSelector{
 		Selector: fields.OneTermEqualSelector(indexregistry.ResourceBindingIndexByFieldCluster, cluster.Name),
 	}); err != nil {
-		klog.ErrorS(err, "Failed to list ResourceBindings", "cluster", cluster.Name)
-		return false, err
-	}
-	if len(rbList.Items) != 0 {
-		return false, nil
+		// If index query fails (e.g., index not registered), fallback to listing all ResourceBindings
+		// and filter manually to ensure cluster deletion can proceed even if index is missing.
+		klog.V(1).ErrorS(err, "Failed to list ResourceBindings using index, falling back to full list", "cluster", cluster.Name)
+		if err := c.List(ctx, rbList); err != nil {
+			klog.ErrorS(err, "Failed to list ResourceBindings", "cluster", cluster.Name)
+			return false, err
+		}
+		// Filter ResourceBindings that reference this cluster
+		filtered := make([]workv1alpha2.ResourceBinding, 0)
+		for i := range rbList.Items {
+			for _, targetCluster := range rbList.Items[i].Spec.Clusters {
+				if targetCluster.Name == cluster.Name {
+					filtered = append(filtered, rbList.Items[i])
+					break
+				}
+			}
+		}
+		if len(filtered) != 0 {
+			return false, nil
+		}
+	} else {
+		if len(rbList.Items) != 0 {
+			return false, nil
+		}
 	}
+
 	// List all ClusterResourceBindings which are assigned to this cluster.
 	crbList := &workv1alpha2.ClusterResourceBindingList{}
 	if err := c.List(ctx, crbList, client.MatchingFieldsSelector{
 		Selector: fields.OneTermEqualSelector(indexregistry.ClusterResourceBindingIndexByFieldCluster, cluster.Name),
 	}); err != nil {
-		klog.ErrorS(err, "Failed to list ClusterResourceBindings", "cluster", cluster.Name)
-		return false, err
-	}
-	if len(crbList.Items) != 0 {
-		return false, nil
+		// If index query fails (e.g., index not registered), fallback to listing all ClusterResourceBindings
+		// and filter manually to ensure cluster deletion can proceed even if index is missing.
+		klog.V(1).ErrorS(err, "Failed to list ClusterResourceBindings using index, falling back to full list", "cluster", cluster.Name)
+		if err := c.List(ctx, crbList); err != nil {
+			klog.ErrorS(err, "Failed to list ClusterResourceBindings", "cluster", cluster.Name)
+			return false, err
+		}
+		// Filter ClusterResourceBindings that reference this cluster
+		filtered := make([]workv1alpha2.ClusterResourceBinding, 0)
+		for i := range crbList.Items {
+			for _, targetCluster := range crbList.Items[i].Spec.Clusters {
+				if targetCluster.Name == cluster.Name {
+					filtered = append(filtered, crbList.Items[i])
+					break
+				}
+			}
+		}
+		if len(filtered) != 0 {
+			return false, nil
+		}
+	} else {
+		if len(crbList.Items) != 0 {
+			return false, nil
+		}
 	}
+
 	return true, nil
 }