Delete Orphaned Machines (#149)

* delete orphaned machines

* fiix pipeline

* up the tag

Makefile

@@ -54,7 +54,7 @@ ETCD=$(TOOLS_BIN_DIR)/etcd
 # Version
 MAJOR_VER ?= 0
 MINOR_VER ?= 3
-PATCH_VER ?= 9-alpha
+PATCH_VER ?= 10-alpha
 
 # Define Docker related variables. Releases should modify and double check these vars.
 REGISTRY ?= mocimages.azurecr.io

azure-pipelines-release.yml

@@ -38,10 +38,10 @@ steps:
       exit 1
     fi
 
-    make REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller RELEASE_TAG=$(RELEASE_TAG) docker-build docker-push
-    make REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller RELEASE_TAG=$(RELEASE_TAG) release
-    make REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller RELEASE_TAG=$(RELEASE_TAG) generate-flavors
-    make REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller RELEASE_TAG=$(RELEASE_TAG) release-pipelines
+    make PROD_REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller TAG=$(RELEASE_TAG) docker-build docker-push
+    make PROD_REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller TAG=$(RELEASE_TAG) release
+    make PROD_REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller TAG=$(RELEASE_TAG) generate-flavors
+    make PROD_REGISTRY=ecpacr.azurecr.io IMAGE_NAME=caphcontroller TAG=$(RELEASE_TAG) release-pipelines
   workingDirectory: '$(System.DefaultWorkingDirectory)'
   displayName: 'Build CAPH'
 

config/manager/manager_image_patch.yaml

@@ -7,5 +7,5 @@ spec:
   template:
     spec:
       containers:
-      - image: mocimages.azurecr.io/caphcontroller:0.3.9-alpha
+      - image: mocimages.azurecr.io/caphcontroller:0.3.10-alpha
         name: manager

controllers/azurestackhcicluster_controller.go

@@ -192,6 +192,15 @@ func (r *AzureStackHCIClusterReconciler) reconcileDelete(clusterScope *scope.Clu
 	}
 
 	if len(azhciMachines) > 0 {
+
+		err := r.deleteOrphanedMachines(clusterScope, azhciMachines)
+		if err != nil {
+			wrappedErr := errors.Wrapf(err, "failed to delete orphaned AzureStackHCIMachines part of AzureStackHCIClusters %s/%s", clusterScope.AzureStackHCICluster.Namespace, clusterScope.AzureStackHCICluster.Name)
+			r.Recorder.Eventf(azureStackHCICluster, corev1.EventTypeWarning, "FailureListMachinesInCluster", wrappedErr.Error())
+			conditions.MarkFalse(azureStackHCICluster, infrav1.NetworkInfrastructureReadyCondition, clusterv1.DeletionFailedReason, clusterv1.ConditionSeverityWarning, err.Error())
+			return reconcile.Result{}, wrappedErr
+		}
+
 		clusterScope.Info("Waiting for AzureStackHCIMachines to be deleted", "count", len(azhciMachines))
 		conditions.MarkFalse(azureStackHCICluster, infrav1.NetworkInfrastructureReadyCondition, infrav1.AzureStackHCIMachinesDeletingReason, clusterv1.ConditionSeverityWarning, "")
 		return reconcile.Result{RequeueAfter: 20 * time.Second}, nil
@@ -230,6 +239,26 @@ func (r *AzureStackHCIClusterReconciler) reconcileDelete(clusterScope *scope.Clu
 	return reconcile.Result{}, nil
 }
 
+func (r *AzureStackHCIClusterReconciler) deleteOrphanedMachines(clusterScope *scope.ClusterScope, azhciMachines []*infrav1.AzureStackHCIMachine) error {
+
+	for _, azhciMachine := range azhciMachines {
+		machine, err := util.GetOwnerMachine(clusterScope.Context, clusterScope.Client, azhciMachine.ObjectMeta)
+		if err != nil {
+			return err
+		}
+		if machine == nil {
+			clusterScope.Info("Deleting Orphaned Machine", "Name", azhciMachine.Name, "AzureStackHCICluster", clusterScope.AzureStackHCICluster.Name)
+			if err := r.Client.Delete(clusterScope.Context, azhciMachine); err != nil {
+				if !apierrors.IsNotFound(err) {
+					return errors.Wrapf(err, "Failed to delete AzureStackHCIMachine %s", azhciMachine)
+				}
+			}
+		}
+	}
+
+	return nil
+}
+
 func (r *AzureStackHCIClusterReconciler) reconcileAzureStackHCILoadBalancer(clusterScope *scope.ClusterScope) (bool, error) {
 	if clusterScope.AzureStackHCILoadBalancer() == nil {
 		clusterScope.Info("Skipping load balancer reconciliation since AzureStackHCICluster.Spec.AzureStackHCILoadBalancer is nil")