@@ -0,0 +1,19 @@
+Additional retry for the target Pod creation race, separate from Kube-OVN #6834.
+Keep until an equivalent guard is verified in the selected upstream version.
+
+diff --git a/pkg/controller/kubevirt.go b/pkg/controller/kubevirt.go
+--- a/pkg/controller/kubevirt.go
++++ b/pkg/controller/kubevirt.go
+@@ -210,9 +210,10 @@
+ 				klog.Infof("successfully set migrate options for lsp %s from %s to %s", portName, sourceNode, targetPod.Spec.NodeName)
+ 			}
+ 		} else {
+-			klog.Warningf("target pod not yet created for migration job UID %s in phase %s, waiting for pod creation",
++			// Pod creation does not change the migration phase, so retry even when
++			// the target pod has not reached the informer cache yet.
++			return fmt.Errorf("target pod not yet created for migration job UID %s in phase %s, waiting for pod creation",
+ 				vmiMigration.UID, vmiMigration.Status.Phase)
+-			return nil
+ 		}
+ 	case kubevirtv1.MigrationSucceeded:
+ 		for _, portName := range portNames {
@@ -20,17 +20,19 @@ import (
 
 func TestVMIMigrationSchedulingRetry(t *testing.T) {
 	for _, tc := range []struct {
-		name       string
-		sourceNode string
-		targetNode string
-		wantRetry  bool
-		wantSetup  bool
+		name             string
+		sourceNode       string
+		targetNode       string
+		wantRetry        bool
+		wantSetup        bool
+		missingTargetPod bool
 	}{
-		{"target not scheduled", "source", "", true, true},
-		{"source not known", "", "target", true, true},
-		{"both nodes unknown", "", "", true, true},
-		{"already scheduled", "source", "target", false, true},
-		{"same node", "source", "source", false, false},
+		{"target pod not created", "source", "target", true, true, true},
+		{"target not scheduled", "source", "", true, true, false},
+		{"source not known", "", "target", true, true, false},
+		{"both nodes unknown", "", "", true, true, false},
+		{"already scheduled", "source", "target", false, true, false},
+		{"same node", "source", "source", false, false, false},
 	} {
 		t.Run(tc.name, func(t *testing.T) {
 			mockCtrl := gomock.NewController(t)
@@ -57,7 +59,9 @@ func TestVMIMigrationSchedulingRetry(t *testing.T) {
 				Spec: corev1.PodSpec{NodeName: tc.targetNode},
 			}
 			indexer := cache.NewIndexer(cache.MetaNamespaceKeyFunc, cache.Indexers{cache.NamespaceIndex: cache.MetaNamespaceIndexFunc})
-			require.NoError(t, indexer.Add(pod))
+			if !tc.missingTargetPod {
+				require.NoError(t, indexer.Add(pod))
+			}
 			virtClient.EXPECT().VirtualMachineInstanceMigration("default").Return(migrationClient).AnyTimes()
 			virtClient.EXPECT().VirtualMachineInstance("default").Return(vmiClient).AnyTimes()
 			migrationClient.EXPECT().Get(gomock.Any(), "migration", metav1.GetOptions{}).Return(migration, nil).AnyTimes()
@@ -76,11 +80,15 @@ func TestVMIMigrationSchedulingRetry(t *testing.T) {
 			queue.Add(key)
 			require.True(t, processNextWorkItem("migration", queue, controller.handleAddOrUpdateVMIMigration, getWorkItemKey))
 			if tc.wantRetry {
-				require.Equal(t, 1, queue.NumRequeues(key), "an unscheduled target must not drop the migration event")
+				require.Equal(t, 1, queue.NumRequeues(key), "a missing or unscheduled target must not drop the migration event")
 				vmi.Status.NodeName = "source"
 				scheduledPod := pod.DeepCopy()
 				scheduledPod.Spec.NodeName = "target"
-				require.NoError(t, indexer.Update(scheduledPod))
+				if tc.missingTargetPod {
+					require.NoError(t, indexer.Add(scheduledPod))
+				} else {
+					require.NoError(t, indexer.Update(scheduledPod))
+				}
 				// The VMIM stays in Scheduling: only the rate-limited retry can finish setup.
 				require.Eventually(t, func() bool { return queue.Len() == 1 }, time.Second, time.Millisecond)
 				require.True(t, processNextWorkItem("migration", queue, controller.handleAddOrUpdateVMIMigration, getWorkItemKey))