diff --git a/test/e2e/e2e_evictioninbackground_test.go b/test/e2e/e2e_evictioninbackground_test.go
index e469db5a7..7e024e22a 100644
--- a/test/e2e/e2e_evictioninbackground_test.go
+++ b/test/e2e/e2e_evictioninbackground_test.go
@@ -110,6 +110,71 @@ ethernets:
 	}
 }
 
+// ensureVMIsLiveMigratable waits until every VMI reports the LiveMigratable
+// condition with status True. If a VMI fails to become migratable within the
+// per-attempt timeout, it is deleted and recreated. This works around an
+// upstream KubeVirt race where virt-handler computes the containerdisk
+// checksum before the disk socket is ready, fails, and never retries; the
+// recreated VMI lands on a node that already has the containerdisk image
+// cached, so the socket comes up before virt-handler's first attempt.
+// See https://github.com/kubernetes-sigs/descheduler/pull/1874 for context.
+func ensureVMIsLiveMigratable(t *testing.T, ctx context.Context, kvClient kubevirtclient.Interface) {
+	t.Helper()
+	const (
+		maxAttempts    = 3
+		perAttemptWait = 120 * time.Second
+		deleteWait     = 60 * time.Second
+	)
+
+	isLiveMigratable := func(vmi *kvcorev1.VirtualMachineInstance) bool {
+		for _, c := range vmi.Status.Conditions {
+			if c.Type == kvcorev1.VirtualMachineInstanceIsMigratable && c.Status == corev1.ConditionTrue {
+				return true
+			}
+		}
+		return false
+	}
+
+	for i := 1; i <= vmiCount; i++ {
+		name := fmt.Sprintf("kubevirtvmi-%v", i)
+		var lastVMI *kvcorev1.VirtualMachineInstance
+		for attempt := 1; attempt <= maxAttempts; attempt++ {
+			err := wait.PollUntilContextTimeout(ctx, 5*time.Second, perAttemptWait, true, func(ctx context.Context) (bool, error) {
+				vmi, err := kvClient.KubevirtV1().VirtualMachineInstances("default").Get(ctx, name, metav1.GetOptions{})
+				if err != nil {
+					klog.Infof("Unable to get vmi %v: %v", name, err)
+					return false, nil
+				}
+				lastVMI = vmi
+				return isLiveMigratable(vmi), nil
+			})
+			if err == nil {
+				klog.Infof("vmi %v is LiveMigratable (attempt %d/%d)", name, attempt, maxAttempts)
+				break
+			}
+			if attempt == maxAttempts {
+				if lastVMI != nil {
+					klog.Infof("Final vmi %v status: phase=%v, conditions=%#v", name, lastVMI.Status.Phase, lastVMI.Status.Conditions)
+				}
+				t.Fatalf("vmi %v never became LiveMigratable after %d attempts", name, maxAttempts)
+			}
+			klog.Warningf("vmi %v not LiveMigratable after %v, recreating (attempt %d/%d) to work around virt-handler containerdisk-socket race", name, perAttemptWait, attempt, maxAttempts)
+			if err := kvClient.KubevirtV1().VirtualMachineInstances("default").Delete(ctx, name, metav1.DeleteOptions{}); err != nil && !apierrors.IsNotFound(err) {
+				t.Fatalf("Unable to delete vmi %v for retry: %v", name, err)
+			}
+			if err := wait.PollUntilContextTimeout(ctx, 2*time.Second, deleteWait, true, func(ctx context.Context) (bool, error) {
+				_, err := kvClient.KubevirtV1().VirtualMachineInstances("default").Get(ctx, name, metav1.GetOptions{})
+				return apierrors.IsNotFound(err), nil
+			}); err != nil {
+				t.Fatalf("Timed out waiting for vmi %v to be deleted: %v", name, err)
+			}
+			if _, err := kvClient.KubevirtV1().VirtualMachineInstances("default").Create(ctx, virtualMachineInstance(i), metav1.CreateOptions{}); err != nil {
+				t.Fatalf("Unable to recreate vmi %v: %v", name, err)
+			}
+		}
+	}
+}
+
 func waitForKubevirtReady(t *testing.T, ctx context.Context, kvClient kubevirtclient.Interface) {
 	obj, err := kvClient.KubevirtV1().KubeVirts("kubevirt").Get(ctx, "kubevirt", metav1.GetOptions{})
 	if err != nil {
@@ -441,6 +506,18 @@ func TestLiveMigrationInBackground(t *testing.T) {
 		t.Fatalf("Error waiting for all vmi active pods to be running: %v", err)
 	}
 
+	// Ensure VMIs are actually live-migratable before starting the descheduler;
+	// recreates any VMI that lost the virt-handler containerdisk-socket race.
+	ensureVMIsLiveMigratable(t, ctx, kvClient)
+
+	// VMIs may have been recreated above, so re-check that pods are Running
+	// before snapshotting the active virt-launcher pod names below.
+	if err := wait.PollUntilContextTimeout(ctx, 5*time.Second, 300*time.Second, true, func(ctx context.Context) (bool, error) {
+		return allVMIsHaveRunningPods(t, ctx, kubeClient, kvClient)
+	}); err != nil {
+		t.Fatalf("Error waiting for all vmi active pods to be running after recreate: %v", err)
+	}
+
 	usedRunningPodNames := make(map[string]struct{})
 	// vmiCount number of names is expected
 	names := kVirtRunningPodNames(t, ctx, kubeClient)