Add support for FailureDomains to AzureMachinePool

cloud/services/scalesets/vmss.go

@@ -49,6 +49,7 @@ type (
 		PublicLoadBalancerName string
 		AdditionalTags         infrav1.Tags
 		AcceleratedNetworking  *bool
+		FailureDomains         []string
 	}
 )
 
@@ -173,6 +174,7 @@ func (s *Service) Reconcile(ctx context.Context, spec interface{}) error {
 				},
 			},
 		},
+		Zones: to.StringSlicePtr(vmssSpec.FailureDomains),
 	}
 
 	err = s.Client.CreateOrUpdate(

cloud/services/scalesets/vmss_test.go

@@ -273,7 +273,8 @@ func TestService_Reconcile(t *testing.T) {
 					Image: &infrav1.Image{
 						ID: to.StringPtr("image"),
 					},
-					CustomData: "customData",
+					CustomData:     "customData",
+					FailureDomains: []string{"1"},
 				}
 			},
 			Setup: func(ctx context.Context, g *gomega.GomegaWithT, svc *Service, scope *scope.ClusterScope, mpScope *scope.MachinePoolScope, spec *Spec) {
@@ -350,6 +351,7 @@ func TestService_Reconcile(t *testing.T) {
 							},
 						},
 					},
+					Zones: to.StringSlicePtr(spec.FailureDomains),
 				}
 
 				skusMock.EXPECT().HasAcceleratedNetworking(gomock.Any(), gomock.Any()).Return(false, nil)
@@ -384,7 +386,8 @@ func TestService_Reconcile(t *testing.T) {
 					Image: &infrav1.Image{
 						ID: to.StringPtr("image"),
 					},
-					CustomData: "customData",
+					CustomData:     "customData",
+					FailureDomains: []string{"1"},
 				}
 			},
 			Setup: func(ctx context.Context, g *gomega.GomegaWithT, svc *Service, scope *scope.ClusterScope, mpScope *scope.MachinePoolScope, spec *Spec) {
@@ -461,6 +464,7 @@ func TestService_Reconcile(t *testing.T) {
 							},
 						},
 					},
+					Zones: to.StringSlicePtr(spec.FailureDomains),
 				}
 
 				skusMock.EXPECT().HasAcceleratedNetworking(gomock.Any(), gomock.Any()).Return(true, nil)

exp/controllers/azuremachinepool_controller.go

@@ -489,6 +489,11 @@ func (s *azureMachinePoolService) CreateOrUpdate() (*infrav1exp.VMSS, error) {
 		return nil, errors.Wrap(err, "failed to retrieve bootstrap data")
 	}
 
+	failureDomains, err := getFailureDomains(s)
+	if err != nil {
+		return nil, errors.Wrap(err, "failed to retrieve failure domains")
+	}
+
 	vmssSpec := &scalesets.Spec{
 		Name:                   s.machinePoolScope.Name(),
 		ResourceGroup:          s.clusterScope.ResourceGroup(),
@@ -505,6 +510,7 @@ func (s *azureMachinePoolService) CreateOrUpdate() (*infrav1exp.VMSS, error) {
 		SubnetID:               s.clusterScope.AzureCluster.Spec.NetworkSpec.Subnets[0].ID,
 		PublicLoadBalancerName: s.clusterScope.Name(),
 		AcceleratedNetworking:  ampSpec.Template.AcceleratedNetworking,
+		FailureDomains:         failureDomains,
 	}
 
 	err = s.virtualMachinesScaleSetSvc.Reconcile(context.TODO(), vmssSpec)
@@ -539,6 +545,30 @@ func (s *azureMachinePoolService) CreateOrUpdate() (*infrav1exp.VMSS, error) {
 	return vmss, nil
 }
 
+func getFailureDomains(s *azureMachinePoolService) ([]string, error) {
+	if s.machinePoolScope.MachinePool.Spec.FailureDomains == nil {
+		return getFailureDomainsIDs(s.clusterScope.AzureCluster.Status.FailureDomains), nil
+	}
+
+	selectedFDsIDs := getFailureDomainsIDs(s.machinePoolScope.MachinePool.Spec.FailureDomains)
+	for _, fdName := range selectedFDsIDs {
+		if _, exists := s.clusterScope.AzureCluster.Status.FailureDomains[fdName]; !exists {
+			return []string{}, errors.Errorf("Selected FailureDomain %#q is not available on this location", fdName)
+		}
+	}
+
+	return selectedFDsIDs, nil
+}
+
+func getFailureDomainsIDs(selectedFailureDomains capiv1.FailureDomains) []string {
+	failureDomains := make([]string, 0, len(selectedFailureDomains))
+	for fdName := range selectedFailureDomains {
+		failureDomains = append(failureDomains, fdName)
+	}
+
+	return failureDomains
+}
+
 // Delete reconciles all the services in pre determined order
 func (s *azureMachinePoolService) Delete() error {
 	vmssSpec := &scalesets.Spec{