feat(gateway): Auto-create ServiceNetwork on Gateway creation

- Gateway reconciler auto-creates a VPC Lattice ServiceNetwork using the
  Gateway's .metadata.name when no external SN exists
- CreateOrUpdate is idempotent: reuses existing SN if found
- Sibling Gateway safety: when deleting a Gateway, skip SN deletion if
  another active Gateway with the same name exists in a different namespace
- Service association guard: block SN deletion if it still has active
  service associations, returning a clear error message
- Ownership tracked via ManagedBy tag; externally-created SNs are never deleted
- Documentation updated with auto-creation behavior, default settings,
  and external SN reuse for advanced configuration
- Unit tests for hasSiblingGateway (4 cases) and Delete with active
  service associations
- E2E integration tests for sibling deletion safety and service
  association delete guard

Author: abdallahmahran10

docs/api-types/gateway.md

@@ -11,14 +11,28 @@ Service networks are identified by Gateway name (without namespace) - for exampl
 will point to a VPC Lattice service network `my-gateway`. If multiple Gateways share the same name, all of them
 will point to the same service network.
 
-VPC Lattice service networks must be managed separately, as it is a broader concept that can cover resources
-outside the Kubernetes cluster. To create and manage a service network, you can either:
-
-- Specify `DEFAULT_SERVICE_NETWORK` configuration option on the controller. This will make the controller
-  to create a service network with such name, and associate the cluster VPC to it for you. This is suitable
-  for simple use cases with single service network.
-- Manage service networks outside the cluster, using AWS Console, CDK, CloudFormation, etc. This is recommended
-  for more advanced use cases that cover multiple clusters and VPCs.
+### ServiceNetwork Lifecycle
+
+When a Gateway is created, the controller automatically creates a VPC Lattice ServiceNetwork with the same name
+and associates it with the cluster VPC. When the Gateway is deleted, the controller deletes the ServiceNetwork
+if it was created by the controller (tracked via the `application-networking.k8s.aws/ManagedBy` tag).
+
+Auto-created ServiceNetworks use **default VPC Lattice settings** — no auth policy, no sharing configuration,
+and no custom attributes. If you need to configure auth type, sharing, or other
+[ServiceNetwork attributes](https://docs.aws.amazon.com/vpc-lattice/latest/APIReference/API_CreateServiceNetwork.html),
+create the ServiceNetwork externally (via Console, CLI, CDK, CloudFormation, etc.) before creating the Gateway.
+The controller will detect the existing ServiceNetwork by name and reuse it without modifying or deleting it.
+
+**Deletion behavior:**
+
+- If multiple Gateways share the same ServiceNetwork name, deleting one Gateway will **not** delete the
+  ServiceNetwork as long as another active Gateway with the same name still exists.
+- If the ServiceNetwork has active service associations, the controller will not delete it and will report
+  an error asking you to detach all services first.
+- Externally-created ServiceNetworks are never deleted by the controller.
+
+In addition to auto-creation, you can also use the `DEFAULT_SERVICE_NETWORK` configuration option on the controller
+to create a default ServiceNetwork at startup.
 
 Gateways with `amazon-vpc-lattice` GatewayClass do not create a single entrypoint to bind Listeners and Routes
 under them. Instead, each Route will have its own domain name assigned. To see an example of how domain names

pkg/controllers/gateway_controller.go

@@ -61,6 +61,7 @@ type gatewayReconciler struct {
 	finalizerManager k8s.FinalizerManager
 	eventRecorder    record.EventRecorder
 	cloud            aws.Cloud
+	snManager        deploy.ServiceNetworkManager
 }
 
 func RegisterGatewayController(
@@ -73,18 +74,20 @@ func RegisterGatewayController(
 	scheme := mgr.GetScheme()
 	evtRec := mgr.GetEventRecorderFor("gateway")
 
+	snManager := deploy.NewDefaultServiceNetworkManager(log, cloud)
+
 	r := &gatewayReconciler{
 		log:              log,
 		client:           mgrClient,
 		scheme:           scheme,
 		finalizerManager: finalizerManager,
 		eventRecorder:    evtRec,
 		cloud:            cloud,
+		snManager:        snManager,
 	}
 
 	if config.DefaultServiceNetwork != "" {
 		// Attempt creation of default service network, move gracefully even if it fails.
-		snManager := deploy.NewDefaultServiceNetworkManager(log, cloud)
 		_, err := snManager.CreateOrUpdate(context.Background(), &model.ServiceNetwork{
 			Spec: model.ServiceNetworkSpec{
 				Name: config.DefaultServiceNetwork,
@@ -177,12 +180,40 @@ func (r *gatewayReconciler) reconcileDelete(ctx context.Context, gw *gwv1.Gatewa
 		}
 	}
 
-	err = r.finalizerManager.RemoveFinalizers(ctx, gw, gatewayFinalizer)
+	hasSibling, err := r.hasSiblingGateway(ctx, gw)
 	if err != nil {
 		return err
 	}
+	if hasSibling {
+		r.log.Infof(ctx, "Skipping ServiceNetwork deletion for %s, another Gateway with the same name exists", gw.Name)
+	} else {
+		if err := r.snManager.Delete(ctx, gw.Name); err != nil {
+			return err
+		}
+	}
 
-	return nil
+	return r.finalizerManager.RemoveFinalizers(ctx, gw, gatewayFinalizer)
+}
+
+// hasSiblingGateway checks if another active Lattice-controlled Gateway with the same .Name exists.
+// Gateways that are themselves being deleted (DeletionTimestamp set) are not counted as siblings,
+// so that simultaneous deletion of all Gateways sharing an SN name still cleans up the SN.
+func (r *gatewayReconciler) hasSiblingGateway(ctx context.Context, gw *gwv1.Gateway) (bool, error) {
+	gwList := &gwv1.GatewayList{}
+	if err := r.client.List(ctx, gwList); err != nil {
+		return false, fmt.Errorf("failed to list gateways: %w", err)
+	}
+	for i := range gwList.Items {
+		other := &gwList.Items[i]
+		if other.UID == gw.UID {
+			continue
+		}
+		if other.Name == gw.Name && other.DeletionTimestamp.IsZero() &&
+			k8s.IsControlledByLatticeGatewayController(ctx, r.client, other) {
+			return true, nil
+		}
+	}
+	return false, nil
 }
 
 func (r *gatewayReconciler) reconcileUpsert(ctx context.Context, gw *gwv1.Gateway) error {
@@ -205,14 +236,12 @@ func (r *gatewayReconciler) reconcileUpsert(ctx context.Context, gw *gwv1.Gatewa
 		return err
 	}
 
-	snInfo, err := r.cloud.Lattice().FindServiceNetwork(ctx, gw.Name)
+	snStatus, err := r.snManager.CreateOrUpdate(ctx, &model.ServiceNetwork{
+		Spec: model.ServiceNetworkSpec{
+			Name: gw.Name,
+		},
+	})
 	if err != nil {
-		if services.IsNotFoundError(err) {
-			if err = r.updateGatewayProgrammedStatus(ctx, gw, gwv1.GatewayReasonPending, "VPC Lattice Service Network not found"); err != nil {
-				return lattice_runtime.NewRetryError()
-			}
-			return nil
-		}
 		if errors.Is(err, services.ErrNameConflict) {
 			if err = r.updateGatewayProgrammedStatus(ctx, gw, gwv1.GatewayReasonInvalid, "Found multiple VPC Lattice Service Networks matching Gateway name. Either ensure only one Service Network has a matching name, or use the Service Network's id as the Gateway name."); err != nil {
 				return lattice_runtime.NewRetryError()
@@ -222,7 +251,7 @@ func (r *gatewayReconciler) reconcileUpsert(ctx context.Context, gw *gwv1.Gatewa
 		return err
 	}
 
-	err = r.updateGatewayProgrammedStatus(ctx, gw, gwv1.GatewayReasonProgrammed, fmt.Sprintf("aws-service-network-arn: %s", *snInfo.SvcNetwork.Arn))
+	err = r.updateGatewayProgrammedStatus(ctx, gw, gwv1.GatewayReasonProgrammed, fmt.Sprintf("aws-service-network-arn: %s", snStatus.ServiceNetworkARN))
 	if err != nil {
 		return err
 	}

pkg/controllers/gateway_controller_test.go

@@ -701,3 +701,79 @@ func TestUpdateGWListenerStatus_SupportedKinds(t *testing.T) {
 		})
 	}
 }
+
+func TestHasSiblingGateway(t *testing.T) {
+	scheme := runtime.NewScheme()
+	clientgoscheme.AddToScheme(scheme)
+	gwv1.Install(scheme)
+	gwv1alpha2.Install(scheme)
+	addOptionalCRDs(scheme)
+
+	latticeControllerName := gwv1.GatewayController("application-networking.k8s.aws/gateway-api-controller")
+	gwClass := &gwv1.GatewayClass{
+		ObjectMeta: metav1.ObjectMeta{Name: "amazon-vpc-lattice"},
+		Spec:       gwv1.GatewayClassSpec{ControllerName: latticeControllerName},
+	}
+
+	makeGW := func(name, ns string, uid types.UID, deleting bool) *gwv1.Gateway {
+		gw := &gwv1.Gateway{
+			ObjectMeta: metav1.ObjectMeta{
+				Name: name, Namespace: ns, UID: uid,
+			},
+			Spec: gwv1.GatewaySpec{GatewayClassName: "amazon-vpc-lattice"},
+		}
+		if deleting {
+			now := metav1.Now()
+			gw.DeletionTimestamp = &now
+			gw.Finalizers = []string{"test"}
+		}
+		return gw
+	}
+
+	tests := []struct {
+		name     string
+		gw       *gwv1.Gateway
+		others   []*gwv1.Gateway
+		expected bool
+	}{
+		{
+			name:     "no sibling — only gateway with this name",
+			gw:       makeGW("my-network", "ns1", "uid-1", false),
+			others:   nil,
+			expected: false,
+		},
+		{
+			name:     "has sibling — same name different namespace",
+			gw:       makeGW("my-network", "ns1", "uid-1", false),
+			others:   []*gwv1.Gateway{makeGW("my-network", "ns2", "uid-2", false)},
+			expected: true,
+		},
+		{
+			name:     "sibling being deleted — not counted",
+			gw:       makeGW("my-network", "ns1", "uid-1", false),
+			others:   []*gwv1.Gateway{makeGW("my-network", "ns2", "uid-2", true)},
+			expected: false,
+		},
+		{
+			name:     "different name — not a sibling",
+			gw:       makeGW("my-network", "ns1", "uid-1", false),
+			others:   []*gwv1.Gateway{makeGW("other-network", "ns2", "uid-2", false)},
+			expected: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			objs := []runtime.Object{gwClass, tt.gw}
+			for _, o := range tt.others {
+				objs = append(objs, o)
+			}
+			k8sClient := testclient.NewClientBuilder().WithScheme(scheme).WithRuntimeObjects(objs...).Build()
+
+			r := &gatewayReconciler{client: k8sClient}
+			result, err := r.hasSiblingGateway(context.Background(), tt.gw)
+			assert.NoError(t, err)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}

pkg/deploy/lattice/service_network_manager.go

@@ -27,6 +27,7 @@ type ServiceNetworkManager interface {
 	DeleteVpcAssociation(ctx context.Context, snName string) error
 
 	CreateOrUpdate(ctx context.Context, serviceNetwork *model.ServiceNetwork) (model.ServiceNetworkStatus, error)
+	Delete(ctx context.Context, snName string) error
 }
 
 func NewDefaultServiceNetworkManager(log gwlog.Logger, cloud pkg_aws.Cloud) *defaultServiceNetworkManager {
@@ -172,6 +173,52 @@ func (m *defaultServiceNetworkManager) DeleteVpcAssociation(ctx context.Context,
 	return nil
 }
 
+func (m *defaultServiceNetworkManager) Delete(ctx context.Context, snName string) error {
+	sn, err := m.cloud.Lattice().FindServiceNetwork(ctx, snName)
+	if err != nil {
+		if services.IsNotFoundError(err) {
+			return nil
+		}
+		return err
+	}
+
+	snArn := aws.StringValue(sn.SvcNetwork.Arn)
+	owned, err := m.cloud.IsArnManaged(ctx, snArn)
+	if err != nil {
+		m.log.Warnf(ctx, "cannot check ownership of ServiceNetwork %s: %s, skipping deletion", snName, err)
+		return nil
+	}
+	if !owned {
+		m.log.Infof(ctx, "ServiceNetwork %s not owned by controller, skipping deletion", snName)
+		return nil
+	}
+
+	assocs, err := m.cloud.Lattice().ListServiceNetworkServiceAssociationsAsList(ctx,
+		&vpclattice.ListServiceNetworkServiceAssociationsInput{
+			ServiceNetworkIdentifier: sn.SvcNetwork.Id,
+		})
+	if err != nil {
+		return fmt.Errorf("failed to list service associations for ServiceNetwork %s: %w", snName, err)
+	}
+	if len(assocs) > 0 {
+		return fmt.Errorf("cannot delete ServiceNetwork %s: %d service association(s) still active, "+
+			"detach all services before deleting the Gateway", snName, len(assocs))
+	}
+
+	if err := m.DeleteVpcAssociation(ctx, snName); err != nil {
+		return err
+	}
+
+	_, err = m.cloud.Lattice().DeleteServiceNetworkWithContext(ctx, &vpclattice.DeleteServiceNetworkInput{
+		ServiceNetworkIdentifier: sn.SvcNetwork.Id,
+	})
+	if err != nil {
+		return err
+	}
+	m.log.Infof(ctx, "Deleted ServiceNetwork %s", snName)
+	return nil
+}
+
 func (m *defaultServiceNetworkManager) getActiveVpcAssociation(ctx context.Context, serviceNetworkId string) (*vpclattice.ServiceNetworkVpcAssociationSummary, error) {
 	vpcLatticeSess := m.cloud.Lattice()
 	associationStatusInput := vpclattice.ListServiceNetworkVpcAssociationsInput{