fix(vd): prevent VirtualDisk from stuck in WaitForFirstConsumer phase when VM is attached (#1516)

Fix VirtualDisk remaining in WaitForFirstConsumer phase even after VM attachment and provisioning has started.

When using WFFC storage class with volume populators:
  1. VD transitions to `WaitForFirstConsumer` waiting for VM
  2. VM is created and attached to VD
  3. Volume provisioning starts (importer pod running)
  4. Issue: VD controller continues setting phase to `WaitForFirstConsumer` because DataVolume is in `PendingPopulation` state, even though the "first consumer" (VM) already exists

Signed-off-by: Daniil Loktev <[email protected]>

Author: loktev-d

images/virtualization-artifact/pkg/controller/vd/internal/source/object_ref_cvi_test.go

@@ -184,6 +184,13 @@ var _ = Describe("ObjectRef ClusterVirtualImage", func() {
 
 		It("waits for the first consumer", func() {
 			dv.Status.Phase = cdiv1.PendingPopulation
+			dv.Status.Conditions = []cdiv1.DataVolumeCondition{
+				{
+					Type:   cdiv1.DataVolumeRunning,
+					Status: corev1.ConditionFalse,
+					Reason: "",
+				},
+			}
 			sc.VolumeBindingMode = ptr.To(storagev1.VolumeBindingWaitForFirstConsumer)
 			client := fake.NewClientBuilder().WithScheme(scheme).WithObjects(pvc, dv, sc).Build()
 

images/virtualization-artifact/pkg/controller/vd/internal/source/object_ref_vi_test.go

@@ -184,6 +184,13 @@ var _ = Describe("ObjectRef VirtualImage", func() {
 
 		It("waits for the first consumer", func() {
 			dv.Status.Phase = cdiv1.PendingPopulation
+			dv.Status.Conditions = []cdiv1.DataVolumeCondition{
+				{
+					Type:   cdiv1.DataVolumeRunning,
+					Status: corev1.ConditionFalse,
+					Reason: "",
+				},
+			}
 			sc.VolumeBindingMode = ptr.To(storagev1.VolumeBindingWaitForFirstConsumer)
 			client := fake.NewClientBuilder().WithScheme(scheme).WithObjects(pvc, dv, sc).Build()
 

images/virtualization-artifact/pkg/controller/vd/internal/source/sources.go

@@ -182,7 +182,9 @@ func setPhaseConditionForPVCProvisioningDisk(
 				Message("Waiting for the pvc importer to be created")
 			return nil
 		}
-		if isStorageClassWFFC(sc) && (dv.Status.Phase == cdiv1.PendingPopulation || dv.Status.Phase == cdiv1.WaitForFirstConsumer) {
+
+		dvRunningCond, _ := conditions.GetDataVolumeCondition(conditions.DVRunningConditionType, dv.Status.Conditions)
+		if isStorageClassWFFC(sc) && (dv.Status.Phase == cdiv1.PendingPopulation || dv.Status.Phase == cdiv1.WaitForFirstConsumer) && dvRunningCond.Reason == "" {
 			vd.Status.Phase = v1alpha2.DiskWaitForFirstConsumer
 			cb.
 				Status(metav1.ConditionFalse).

images/virtualization-artifact/pkg/controller/vd/internal/source/step/wait_for_dv_step.go

@@ -126,13 +126,18 @@ func (s WaitForDVStep) setForProvisioning(vd *v1alpha2.VirtualDisk) (set bool) {
 }
 
 func (s WaitForDVStep) setForFirstConsumerIsAwaited(ctx context.Context, vd *v1alpha2.VirtualDisk) (set bool, err error) {
+	if vd.Status.StorageClassName == "" {
+		return false, fmt.Errorf("StorageClassName is empty, please report a bug")
+	}
+
 	sc, err := object.FetchObject(ctx, types.NamespacedName{Name: vd.Status.StorageClassName}, s.client, &storagev1.StorageClass{})
 	if err != nil {
 		return false, fmt.Errorf("get sc: %w", err)
 	}
 
+	dvRunningCond, _ := conditions.GetDataVolumeCondition(conditions.DVRunningConditionType, s.dv.Status.Conditions)
 	isWFFC := sc != nil && sc.VolumeBindingMode != nil && *sc.VolumeBindingMode == storagev1.VolumeBindingWaitForFirstConsumer
-	if isWFFC && (s.dv.Status.Phase == cdiv1.PendingPopulation || s.dv.Status.Phase == cdiv1.WaitForFirstConsumer) {
+	if isWFFC && (s.dv.Status.Phase == cdiv1.PendingPopulation || s.dv.Status.Phase == cdiv1.WaitForFirstConsumer) && dvRunningCond.Reason == "" {
 		vd.Status.Phase = v1alpha2.DiskWaitForFirstConsumer
 		s.cb.
 			Status(metav1.ConditionFalse).

images/virtualization-artifact/pkg/controller/vd/internal/watcher/datavolume_watcher.go

@@ -28,7 +28,6 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/source"
 
 	"github.com/deckhouse/virtualization-controller/pkg/controller/conditions"
-	"github.com/deckhouse/virtualization-controller/pkg/controller/service"
 	"github.com/deckhouse/virtualization/api/core/v1alpha2"
 )
 
@@ -73,8 +72,14 @@ func (w *DataVolumeWatcher) Watch(mgr manager.Manager, ctr controller.Controller
 						return true
 					}
 
-					dvRunning := service.GetDataVolumeCondition(cdiv1.DataVolumeRunning, e.ObjectNew.Status.Conditions)
-					return dvRunning != nil && (dvRunning.Reason == "Error" || dvRunning.Reason == "ImagePullFailed")
+					oldDVRunning, _ := conditions.GetDataVolumeCondition(conditions.DVRunningConditionType, e.ObjectOld.Status.Conditions)
+					newDVRunning, _ := conditions.GetDataVolumeCondition(conditions.DVRunningConditionType, e.ObjectNew.Status.Conditions)
+
+					if oldDVRunning.Reason != newDVRunning.Reason {
+						return true
+					}
+
+					return newDVRunning.Reason == "Error" || newDVRunning.Reason == "ImagePullFailed"
 				},
 			},
 		),