diff --git a/ray-operator/controllers/ray/raycluster_controller.go b/ray-operator/controllers/ray/raycluster_controller.go
index 18b57f19922..7de38e35271 100644
--- a/ray-operator/controllers/ray/raycluster_controller.go
+++ b/ray-operator/controllers/ray/raycluster_controller.go
@@ -621,6 +621,17 @@ func (r *RayClusterReconciler) reconcilePods(ctx context.Context, instance *rayv
 	} else if len(headPods.Items) == 0 {
 		originatedFrom := utils.GetCRDType(instance.Labels[utils.RayOriginatedFromCRDLabelKey])
 		if originatedFrom == utils.RayJobCRD {
+			// Recreating the head Pod if the RayCluster created by RayJob is provisioned doesn't help RayJob.
+			//
+			// Case 1: GCS fault tolerance is disabled
+			//
+			// In this case, the worker Pods will be killed by the new head Pod when it is created, so the new Ray job will not be running in
+			// a "provisioned" cluster.
+			//
+			// Case 2: GCS fault tolerance is enabled
+			//
+			// In this case, the worker Pods will not be killed by the new head Pod when it is created, but the submission ID has already been
+			// used by the old Ray job, so the new Ray job will fail.
 			if meta.IsStatusConditionTrue(instance.Status.Conditions, string(rayv1.RayClusterProvisioned)) {
 				logger.Info(
 					"reconcilePods: Found 0 head Pods for a RayJob-managed RayCluster; skipping head creation to let RayJob controller handle the failure",