Allow ICI status to be restored

Adds the restore status annotation to ICI CRs and also waits for the
status to be added to resources that are created by a restore operation.

Because the spec and status need to be set by two different API calls to
the API server the controller needs to know when to wait for the status
to be set before initializing the conditions.

Initializing the conditions in a restore scenario can lead to a conflict
where the restore fails and the old status doesn't get set.

https://issues.redhat.com/browse/MGMT-22276

Author: carbonin

controllers/imageclusterinstall_controller.go

@@ -91,28 +91,31 @@ type imagePullSecret struct {
 }
 
 const (
-	detachedAnnotation          = "baremetalhost.metal3.io/detached"
-	detachedAnnotationValue     = "imageclusterinstall-controller"
-	inspectAnnotation           = "inspect.metal3.io"
-	rebootAnnotation            = "reboot.metal3.io"
-	rebootAnnotationValue       = ""
-	ibioManagedBMH              = "image-based-install-managed"
-	ClusterConfigDir            = "cluster-configuration"
-	extraManifestsDir           = "extra-manifests"
-	nmstateSecretKey            = "nmstate"
-	clusterInstallFinalizerName = "imageclusterinstall." + v1alpha1.Group + "/deprovision"
-	caBundleFileName            = "tls-ca-bundle.pem"
-	imageBasedInstallInvoker    = "image-based-install"
-	invokerCMFileName           = "invoker-cm.yaml"
-	installTimeoutAnnotation    = "imageclusterinstall." + v1alpha1.Group + "/install-timeout"
-	backupLabel                 = "cluster.open-cluster-management.io/backup"
-	backupLabelValue            = "true"
-	imageBasedConfigFilename    = "image-based-config.yaml"
-	installConfigFilename       = "install-config.yaml"
-	authDir                     = "auth"
-	kubeAdminFile               = "kubeadmin-password"
-	FilesDir                    = "files"
-	IsoName                     = "imagebasedconfig.iso"
+	detachedAnnotation           = "baremetalhost.metal3.io/detached"
+	detachedAnnotationValue      = "imageclusterinstall-controller"
+	inspectAnnotation            = "inspect.metal3.io"
+	rebootAnnotation             = "reboot.metal3.io"
+	rebootAnnotationValue        = ""
+	ibioManagedBMH               = "image-based-install-managed"
+	ClusterConfigDir             = "cluster-configuration"
+	extraManifestsDir            = "extra-manifests"
+	nmstateSecretKey             = "nmstate"
+	clusterInstallFinalizerName  = "imageclusterinstall." + v1alpha1.Group + "/deprovision"
+	caBundleFileName             = "tls-ca-bundle.pem"
+	imageBasedInstallInvoker     = "image-based-install"
+	invokerCMFileName            = "invoker-cm.yaml"
+	installTimeoutAnnotation     = "imageclusterinstall." + v1alpha1.Group + "/install-timeout"
+	backupLabel                  = "cluster.open-cluster-management.io/backup"
+	backupLabelValue             = "true"
+	imageBasedConfigFilename     = "image-based-config.yaml"
+	installConfigFilename        = "install-config.yaml"
+	authDir                      = "auth"
+	kubeAdminFile                = "kubeadmin-password"
+	FilesDir                     = "files"
+	IsoName                      = "imagebasedconfig.iso"
+	restoreStatusAnnotation      = "velero.io/restore-status"
+	restoreStatusAnnotationValue = "true"
+	restoreSourceLabel           = "velero.io/restore-name"
 )
 
 //+kubebuilder:rbac:groups="",resources=secrets,verbs=get;list;watch;create;update;patch;delete
@@ -164,6 +167,16 @@ func (r *ImageClusterInstallReconciler) Reconcile(ctx context.Context, req ctrl.
 		log.Info("Running reconcile for ici with bootTime set")
 	}
 
+	if iciIsBeingRestored(ici) {
+		log.Info("Waiting for status restore")
+		return ctrl.Result{}, nil
+	}
+
+	if err := r.ensureRestoreStatusAnnotation(ctx, ici); err != nil {
+		log.Errorf("Failed to ensure restore status annotation: %s", err)
+		return ctrl.Result{}, err
+	}
+
 	if err := r.initializeConditions(ctx, ici); err != nil {
 		log.Errorf("Failed to initialize conditions: %s", err)
 		return ctrl.Result{}, err
@@ -1457,3 +1470,21 @@ func verifyIsoAndAuthExists(clusterConfigPath string) bool {
 
 	return true
 }
+
+// iciIsBeingRestored returns true if the status is completely empty and the velero restore source label is present
+func iciIsBeingRestored(ici *v1alpha1.ImageClusterInstall) bool {
+	_, hasLabel := ici.ObjectMeta.Labels[restoreSourceLabel]
+	return hasLabel &&
+		len(ici.Status.Conditions) == 0 &&
+		ici.Status.InstallRestarts == 0 &&
+		ici.Status.BareMetalHostRef == nil &&
+		ici.Status.BootTime.IsZero()
+}
+
+func (r *ImageClusterInstallReconciler) ensureRestoreStatusAnnotation(ctx context.Context, ici *v1alpha1.ImageClusterInstall) error {
+	patch := client.MergeFrom(ici.DeepCopy())
+	if !setAnnotationIfNotExists(&ici.ObjectMeta, restoreStatusAnnotation, restoreStatusAnnotationValue) {
+		return nil
+	}
+	return r.Patch(ctx, ici, patch)
+}

controllers/imageclusterinstall_controller_test.go

@@ -566,6 +566,51 @@ var _ = Describe("Reconcile", func() {
 		Expect(res).To(Equal(ctrl.Result{}))
 	})
 
+	It("sets restore status annotation on image cluster install", func() {
+		Expect(c.Create(ctx, clusterInstall)).To(Succeed())
+		Expect(c.Create(ctx, clusterDeployment)).To(Succeed())
+
+		key := types.NamespacedName{
+			Namespace: clusterInstallNamespace,
+			Name:      clusterInstallName,
+		}
+		installerSuccess()
+		res, err := r.Reconcile(ctx, ctrl.Request{NamespacedName: key})
+		Expect(err).NotTo(HaveOccurred())
+		Expect(res).To(Equal(ctrl.Result{}))
+
+		// Verify the restore status annotation was set
+		updatedICI := &v1alpha1.ImageClusterInstall{}
+		Expect(c.Get(ctx, key, updatedICI)).To(Succeed())
+		Expect(updatedICI.Annotations).To(HaveKey(restoreStatusAnnotation))
+		Expect(updatedICI.Annotations[restoreStatusAnnotation]).To(Equal(restoreStatusAnnotationValue))
+	})
+
+	It("stops reconcile early when status is empty and restore name label is set", func() {
+		clusterInstall.ObjectMeta.Labels = map[string]string{
+			restoreSourceLabel: "test-restore",
+		}
+		Expect(c.Create(ctx, clusterInstall)).To(Succeed())
+
+		key := types.NamespacedName{
+			Namespace: clusterInstallNamespace,
+			Name:      clusterInstallName,
+		}
+
+		// Reconcile should stop early - no installer methods should be called
+		res, err := r.Reconcile(ctx, ctrl.Request{NamespacedName: key})
+		Expect(err).NotTo(HaveOccurred())
+		Expect(res).To(Equal(ctrl.Result{}))
+
+		// Verify the status remains empty (reconcile stopped early)
+		updatedICI := &v1alpha1.ImageClusterInstall{}
+		Expect(c.Get(ctx, key, updatedICI)).To(Succeed())
+		Expect(len(updatedICI.Status.Conditions)).To(Equal(0))
+		Expect(updatedICI.Status.InstallRestarts).To(Equal(0))
+		Expect(updatedICI.Status.BareMetalHostRef).To(BeNil())
+		Expect(updatedICI.Status.BootTime.IsZero()).To(BeTrue())
+	})
+
 	It("creates the ca bundle", func() {
 		caData := map[string]string{caBundleFileName: "mycabundle"}
 		cm := &corev1.ConfigMap{