try adding vmss update

Author: mweibel

api/v1beta1/types.go

@@ -63,6 +63,8 @@ const (
 	PatchFuture string = "PATCH"
 	// PutFuture is a future that was derived from a PUT request.
 	PutFuture string = "PUT"
+	// PostFuture is a future that was derived from a POST request.
+	PostFuture string = "POST"
 	// DeleteFuture is a future that was derived from a DELETE request.
 	DeleteFuture string = "DELETE"
 )

azure/converters/vmss.go

@@ -154,6 +154,10 @@ func SDKToVMSSVM(sdkInstance armcompute.VirtualMachineScaleSetVM) *azure.VMSSVM
 		instance.AvailabilityZone = *sdkInstance.Zones[0]
 	}
 
+	if sdkInstance.Properties != nil && sdkInstance.Properties.LatestModelApplied != nil {
+		instance.LatestModelApplied = *sdkInstance.Properties.LatestModelApplied
+	}
+
 	return &instance
 }
 

azure/services/scalesets/client.go

@@ -36,6 +36,8 @@ type Client interface {
 
 	CreateOrUpdateAsync(ctx context.Context, spec azure.ResourceSpecGetter, resumeToken string, parameters interface{}) (result interface{}, poller *runtime.Poller[armcompute.VirtualMachineScaleSetsClientCreateOrUpdateResponse], err error)
 	DeleteAsync(ctx context.Context, spec azure.ResourceSpecGetter, resumeToken string) (poller *runtime.Poller[armcompute.VirtualMachineScaleSetsClientDeleteResponse], err error)
+
+	BeginUpdateInstances(ctx context.Context, spec azure.ResourceSpecGetter, vmInstanceIDs armcompute.VirtualMachineScaleSetVMInstanceRequiredIDs, resumeToken string) (*runtime.Poller[armcompute.VirtualMachineScaleSetsClientUpdateInstancesResponse], error)
 }
 
 // AzureClient contains the Azure go-sdk Client.
@@ -206,3 +208,34 @@ func (ac *AzureClient) DeleteAsync(ctx context.Context, spec azure.ResourceSpecG
 	// if the operation completed, return a nil poller.
 	return nil, err
 }
+
+// BeginUpdateInstances - Upgrades one or more virtual machines to the latest SKU set in the VM scale set model.
+// If the operation fails it returns an *azcore.ResponseError type.
+//
+// Parameters
+//   - spec - The ResourceSpecGetter containing used for name and resource group of the virtual machine scale set.
+//   - vmInstanceIDs - A list of virtual machine instance IDs from the VM scale set.
+func (ac *AzureClient) BeginUpdateInstances(ctx context.Context, spec azure.ResourceSpecGetter, vmInstanceIDs armcompute.VirtualMachineScaleSetVMInstanceRequiredIDs, resumeToken string) (poller *runtime.Poller[armcompute.VirtualMachineScaleSetsClientUpdateInstancesResponse], err error) {
+	ctx, _, done := tele.StartSpanWithLogger(ctx, "scalesets.AzureClient.BeginUpdateInstances")
+	defer done()
+
+	opts := &armcompute.VirtualMachineScaleSetsClientBeginUpdateInstancesOptions{ResumeToken: resumeToken}
+	poller, err = ac.scalesets.BeginUpdateInstances(ctx, spec.ResourceGroupName(), spec.ResourceName(), vmInstanceIDs, opts)
+	if err != nil {
+		return nil, err
+	}
+
+	ctx, cancel := context.WithTimeout(ctx, ac.apiCallTimeout)
+	defer cancel()
+
+	pollOpts := &runtime.PollUntilDoneOptions{Frequency: async.DefaultPollerFrequency}
+	_, err = poller.PollUntilDone(ctx, pollOpts)
+	if err != nil {
+		// if an error occurs, return the Poller.
+		// this means the long-running operation didn't finish in the specified timeout.
+		return poller, err
+	}
+
+	// if the operation completed, return a nil poller.
+	return nil, err
+}

azure/services/scalesets/mock_scalesets/client_mock.go

@@ -130,6 +130,44 @@ func (s *Service) Reconcile(ctx context.Context) (retErr error) {
 		}
 		s.Scope.SetProviderID(providerID)
 		s.Scope.SetVMSSState(&fetchedVMSS)
+
+		azLatestModelApplied := true
+		for _, instance := range scaleSetSpec.VMSSInstances {
+			if instance.Properties.LatestModelApplied != nil && !*instance.Properties.LatestModelApplied {
+				azLatestModelApplied = false
+				break
+			}
+		}
+		if !azLatestModelApplied {
+			resourceName := spec.ResourceName()
+			futureType := infrav1.PostFuture
+			// Check for an ongoing long-running operation.
+			resumeToken := ""
+			if future := s.Scope.GetLongRunningOperationState(resourceName, serviceName, futureType); future != nil {
+				t, err := converters.FutureToResumeToken(*future)
+				if err != nil {
+					s.Scope.DeleteLongRunningOperationState(resourceName, serviceName, futureType)
+					return errors.Wrap(err, "could not decode future data, resetting long-running operation state")
+				}
+				resumeToken = t
+			}
+			target := "*"
+			poller, err := s.Client.BeginUpdateInstances(ctx, spec, armcompute.VirtualMachineScaleSetVMInstanceRequiredIDs{
+				InstanceIDs: []*string{&target},
+			}, resumeToken)
+			if poller != nil && azure.IsContextDeadlineExceededOrCanceledError(err) {
+				future, err := converters.PollerToFuture(poller, futureType, serviceName, resourceName, spec.ResourceGroupName())
+				if err != nil {
+					return errors.Wrap(err, "failed to convert poller to future")
+				}
+				s.Scope.SetLongRunningOperationState(future)
+				return azure.WithTransientError(azure.NewOperationNotDoneError(future), s.Scope.DefaultedReconcilerRequeue())
+			}
+
+			// Once the operation is done, delete the long-running operation state. Even if the operation ended with
+			// an error, clear out any lingering state to try the operation again.
+			s.Scope.DeleteLongRunningOperationState(resourceName, serviceName, futureType)
+		}
 	}
 
 	return err

azure/services/scalesets/scalesets.go

@@ -104,6 +104,7 @@ type (
 		State              infrav1.ProvisioningState     `json:"vmState,omitempty"`
 		BootstrappingState infrav1.ProvisioningState     `json:"bootstrappingState,omitempty"`
 		OrchestrationMode  infrav1.OrchestrationModeType `json:"orchestrationMode,omitempty"`
+		LatestModelApplied bool                          `json:"latestModelApplied,omitempty"`
 	}
 
 	// VMSS defines a virtual machine scale set.

azure/types.go

@@ -8,5 +8,5 @@ spec:
     spec:
       containers:
       # Change the value of image field below to your controller image URL
-      - image: gcr.io/k8s-staging-cluster-api-azure/cluster-api-azure-controller:main
+      - image: localhost:5000/cluster-api-azure-controller-amd64:dev
         name: manager

config/capz/manager_image_patch.yaml

@@ -332,6 +332,7 @@ func (ampmr *AzureMachinePoolMachineController) reconcileDelete(ctx context.Cont
 	if !machineScope.AzureMachinePool.ObjectMeta.DeletionTimestamp.IsZero() {
 		log.Info("Skipping VMSS VM deletion as VMSS delete will delete individual instances")
 
+		// FIXME(mw): check if VMSS VM is already deleted to avoid recreating AMPM
 		controllerutil.RemoveFinalizer(machineScope.AzureMachinePoolMachine, infrav1exp.AzureMachinePoolMachineFinalizer)
 		return reconcile.Result{}, nil
 	}