Add Watch for Zone changes to update FailureDomains accordingly

Signed-off-by: Gong Zhang <[email protected]>

config/manager/manager.yaml

@@ -21,7 +21,7 @@ spec:
         - "--diagnostics-address=${CAPI_DIAGNOSTICS_ADDRESS:=:8443}"
         - "--insecure-diagnostics=${CAPI_INSECURE_DIAGNOSTICS:=false}"
         - --v=4
-        - "--feature-gates=NodeAntiAffinity=${EXP_NODE_ANTI_AFFINITY:=false},NamespaceScopedZone=${EXP_NAMESPACE_SCOPED_ZONE:=false}"
+        - "--feature-gates=NodeAntiAffinity=${EXP_NODE_ANTI_AFFINITY:=false},NamespaceScopedZones=${EXP_NAMESPACE_SCOPED_ZONES:=false}"
         image: controller:latest
         imagePullPolicy: IfNotPresent
         name: manager

controllers/vmware/vspherecluster_reconciler.go

@@ -370,18 +370,52 @@ func (r *ClusterReconciler) VSphereMachineToCluster(ctx context.Context, o clien
 	}}
 }
 
+// ZoneToVSphereClusters adds reconcile requests for VSphereClusters when Zone has an event.
+func (r *ClusterReconciler) ZoneToVSphereClusters(ctx context.Context, o client.Object) []reconcile.Request {
+	log := ctrl.LoggerFrom(ctx)
+
+	zone, ok := o.(*topologyv1.Zone)
+	if !ok {
+		log.Error(nil, fmt.Sprintf("Expected a Zone but got a %T", o))
+		return nil
+	}
+	log = log.WithValues("Zone", klog.KObj(zone))
+	ctx = ctrl.LoggerInto(ctx, log)
+
+	VSphereClusters := &vmwarev1.VSphereClusterList{}
+	err := r.Client.List(ctx, VSphereClusters, &client.ListOptions{Namespace: zone.Namespace})
+	if err != nil {
+		log.V(4).Error(err, "Failed to get VSphereClusters from Zone")
+		return nil
+	}
+
+	log.V(6).Info("Triggering VSphereCluster reconcile from Zone")
+	requests := []reconcile.Request{}
+	for _, c := range VSphereClusters.Items {
+		r := reconcile.Request{
+			NamespacedName: types.NamespacedName{
+				Name:      c.Name,
+				Namespace: c.Namespace,
+			},
+		}
+		requests = append(requests, r)
+	}
+
+	return requests
+}
+
 // Returns the failure domain information discovered on the cluster
 // hosting this controller.
 func (r *ClusterReconciler) getFailureDomains(ctx context.Context, namespace string) (clusterv1.FailureDomains, error) {
 	failureDomains := clusterv1.FailureDomains{}
-	// Determine the source of failure domain based on feature gates NamespaceScopedZone.
-	// If NamespaceScopedZone is enabled, use Zone which is Namespace scoped,otherwise use
+	// Determine the source of failure domain based on feature gates NamespaceScopedZones.
+	// If NamespaceScopedZones is enabled, use Zone which is Namespace scoped,otherwise use
 	// Availability Zone which is Cluster scoped.
-	if feature.Gates.Enabled(feature.NamespaceScopedZone) {
+	if feature.Gates.Enabled(feature.NamespaceScopedZones) {
 		zoneList := &topologyv1.ZoneList{}
 		listOptions := &client.ListOptions{Namespace: namespace}
 		if err := r.Client.List(ctx, zoneList, listOptions); err != nil {
-			return nil, err
+			return nil, errors.Wrapf(err, fmt.Sprintf("Failed to list Zones in namespace %s", namespace))
 		}
 
 		for _, zone := range zoneList.Items {

controllers/vmware/vspherecluster_reconciler_test.go

@@ -237,7 +237,7 @@ func TestClusterReconciler_getFailureDomains(t *testing.T) {
 					WithObjects(append([]client.Object{namespace}, tt.objects...)...).
 					Build(),
 			}
-			defer utilfeature.SetFeatureGateDuringTest(t, feature.Gates, feature.NamespaceScopedZone, tt.featureGate)()
+			defer utilfeature.SetFeatureGateDuringTest(t, feature.Gates, feature.NamespaceScopedZones, tt.featureGate)()
 			got, err := r.getFailureDomains(ctx, namespace.Name)
 			if (err != nil) != tt.wantErr {
 				t.Errorf("ClusterReconciler.getFailureDomains() error = %v, wantErr %v", err, tt.wantErr)

controllers/vspherecluster_controller.go

@@ -38,6 +38,7 @@ import (
 	vmwarev1 "sigs.k8s.io/cluster-api-provider-vsphere/apis/vmware/v1beta1"
 	"sigs.k8s.io/cluster-api-provider-vsphere/controllers/vmware"
 	"sigs.k8s.io/cluster-api-provider-vsphere/feature"
+	topologyv1 "sigs.k8s.io/cluster-api-provider-vsphere/internal/apis/topology/v1alpha1"
 	capvcontext "sigs.k8s.io/cluster-api-provider-vsphere/pkg/context"
 	inframanager "sigs.k8s.io/cluster-api-provider-vsphere/pkg/manager"
 	"sigs.k8s.io/cluster-api-provider-vsphere/pkg/services"
@@ -72,15 +73,24 @@ func AddClusterControllerToManager(ctx context.Context, controllerManagerCtx *ca
 			},
 			NetworkProvider: networkProvider,
 		}
-		return ctrl.NewControllerManagedBy(mgr).
+		builder := ctrl.NewControllerManagedBy(mgr).
 			For(&vmwarev1.VSphereCluster{}).
 			WithOptions(options).
 			Watches(
 				&vmwarev1.VSphereMachine{},
 				handler.EnqueueRequestsFromMapFunc(reconciler.VSphereMachineToCluster),
 			).
-			WithEventFilter(predicates.ResourceNotPausedAndHasFilterLabel(ctrl.LoggerFrom(ctx), controllerManagerCtx.WatchFilterValue)).
-			Complete(reconciler)
+			WithEventFilter(predicates.ResourceNotPausedAndHasFilterLabel(ctrl.LoggerFrom(ctx), controllerManagerCtx.WatchFilterValue))
+
+		// Conditionally add a Watch for topologyv1.Zone when the feature gate is enabled
+		if feature.Gates.Enabled(feature.NamespaceScopedZones) {
+			builder = builder.Watches(
+				&topologyv1.Zone{},
+				handler.EnqueueRequestsFromMapFunc(reconciler.ZoneToVSphereClusters),
+			).WithEventFilter(predicates.ResourceNotPausedAndHasFilterLabel(ctrl.LoggerFrom(ctx), controllerManagerCtx.WatchFilterValue))
+		}
+
+		return builder.Complete(reconciler)
 	}
 
 	reconciler := &clusterReconciler{

feature/feature.go

@@ -34,10 +34,10 @@ const (
 	// alpha: v1.4
 	NodeAntiAffinity featuregate.Feature = "NodeAntiAffinity"
 
-	// NamespaceScopedZone is a feature gate for the NamespaceScopedZone functionality for supervisor.
+	// NamespaceScopedZones is a feature gate for the NamespaceScopedZones functionality for supervisor.
 	//
 	// alpha: v1.11
-	NamespaceScopedZone featuregate.Feature = "NamespaceScopedZone"
+	NamespaceScopedZones featuregate.Feature = "NamespaceScopedZones"
 )
 
 func init() {
@@ -48,6 +48,6 @@ func init() {
 // To add a new feature, define a key for it above and add it here.
 var defaultCAPVFeatureGates = map[featuregate.Feature]featuregate.FeatureSpec{
 	// Every feature should be initiated here:
-	NodeAntiAffinity:    {Default: false, PreRelease: featuregate.Alpha},
-	NamespaceScopedZone: {Default: false, PreRelease: featuregate.Alpha},
+	NodeAntiAffinity:     {Default: false, PreRelease: featuregate.Alpha},
+	NamespaceScopedZones: {Default: false, PreRelease: featuregate.Alpha},
 }

test/e2e/e2e_suite_test.go

@@ -28,7 +28,6 @@ import (
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 	vmoprv1 "github.com/vmware-tanzu/vm-operator/api/v1alpha2"
-	topologyv1 "github.com/vmware-tanzu/vm-operator/external/tanzu-topology/api/v1alpha1"
 	corev1 "k8s.io/api/core/v1"
 	storagev1 "k8s.io/api/storage/v1"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -42,6 +41,7 @@ import (
 
 	infrav1 "sigs.k8s.io/cluster-api-provider-vsphere/apis/v1beta1"
 	vmwarev1 "sigs.k8s.io/cluster-api-provider-vsphere/apis/vmware/v1beta1"
+	topologyv1 "sigs.k8s.io/cluster-api-provider-vsphere/internal/apis/topology/v1alpha1"
 	vsphereframework "sigs.k8s.io/cluster-api-provider-vsphere/test/framework"
 	vsphereip "sigs.k8s.io/cluster-api-provider-vsphere/test/framework/ip"
 	vspherelog "sigs.k8s.io/cluster-api-provider-vsphere/test/framework/log"

test/framework/vmoperator/vmoperator.go

@@ -28,7 +28,6 @@ import (
 	"github.com/pkg/errors"
 	vmoprv1 "github.com/vmware-tanzu/vm-operator/api/v1alpha2"
 	vmoprv1common "github.com/vmware-tanzu/vm-operator/api/v1alpha2/common"
-	topologyv1 "github.com/vmware-tanzu/vm-operator/external/tanzu-topology/api/v1alpha1"
 	"github.com/vmware/govmomi/pbm"
 	"github.com/vmware/govmomi/vapi/library"
 	"github.com/vmware/govmomi/vapi/rest"
@@ -44,6 +43,7 @@ import (
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
+	topologyv1 "sigs.k8s.io/cluster-api-provider-vsphere/internal/apis/topology/v1alpha1"
 	"sigs.k8s.io/cluster-api-provider-vsphere/pkg/session"
 	vcsimv1 "sigs.k8s.io/cluster-api-provider-vsphere/test/infrastructure/vcsim/api/v1alpha1"
 )

test/go.mod

@@ -14,7 +14,6 @@ require (
 	github.com/vmware-tanzu/net-operator-api v0.0.0-20240326163340-1f32d6bf7f9d
 	// The version of vm-operator should be kept in sync with the manifests at: config/deployments/integration-tests
 	github.com/vmware-tanzu/vm-operator/api v1.8.6
-	github.com/vmware-tanzu/vm-operator/external/tanzu-topology v0.0.0-20240404200847-de75746a9505
 	github.com/vmware/govmomi v0.39.0
 )