diff --git a/src/include/daos_srv/rebuild.h b/src/include/daos_srv/rebuild.h
index 55c111b518f..90596e64d4f 100644
--- a/src/include/daos_srv/rebuild.h
+++ b/src/include/daos_srv/rebuild.h
@@ -71,6 +71,7 @@ int ds_rebuild_schedule(struct ds_pool *pool, uint32_t map_ver,
 			daos_epoch_t stable_eph, uint32_t layout_version,
 			struct pool_target_id_list *tgts,
 			daos_rebuild_opc_t rebuild_op, uint64_t delay_sec);
+void ds_rebuild_restart_if_rank_wip(uuid_t pool_uuid, d_rank_t rank);
 int ds_rebuild_query(uuid_t pool_uuid,
 		     struct daos_rebuild_status *status);
 void ds_rebuild_running_query(uuid_t pool_uuid, uint32_t opc, uint32_t *rebuild_ver,
diff --git a/src/pool/srv_pool.c b/src/pool/srv_pool.c
index 6e3a01379fa..aba200343ea 100644
--- a/src/pool/srv_pool.c
+++ b/src/pool/srv_pool.c
@@ -1458,6 +1458,47 @@ resume_event_handling(struct pool_svc *svc)
 	ABT_mutex_unlock(events->pse_mutex);
 }
 
+/*
+ * Restart rebuild if the rank is UPIN in pool map and is in rebuilding.
+ *
+ * This function only used when PS leader gets CRT_EVT_ALIVE event of engine \a rank,
+ * if that rank is UPIN in pool map and with unfinished rebuilding should be massive
+ * failure case -
+ * 1. some engines down and triggered rebuild.
+ * 2. the engine \a rank participated the rebuild, not finished yet, it became down again,
+ *    the #failures exceeds pool RF and will not change pool map.
+ * 3. That engine restarted by administrator.
+ *
+ * In that case should recover the rebuild task on engine \a rank, to simplify it now just
+ * abort and retry the global rebuild task.
+ */
+static void
+pool_restart_rebuild_if_rank_wip(struct ds_pool *pool, d_rank_t rank)
+{
+	struct pool_domain	*dom;
+
+	ABT_rwlock_rdlock(pool->sp_lock);
+	dom = pool_map_find_dom_by_rank(pool->sp_map, rank);
+	if (dom == NULL) {
+		ABT_rwlock_unlock(pool->sp_lock);
+		D_INFO(DF_UUID": rank %d non-exist on pool map.\n",
+		       DP_UUID(pool->sp_uuid), rank);
+		return;
+	}
+
+	if (dom->do_comp.co_status != PO_COMP_ST_UPIN) {
+		ABT_rwlock_unlock(pool->sp_lock);
+		D_INFO(DF_UUID": rank %d status %d in pool map, got CRT_EVT_ALIVE.\n",
+			DP_UUID(pool->sp_uuid), rank, dom->do_comp.co_status);
+		return;
+	}
+	ABT_rwlock_unlock(pool->sp_lock);
+
+	ds_rebuild_restart_if_rank_wip(pool->sp_uuid, rank);
+
+	return;
+}
+
 static int pool_svc_exclude_ranks(struct pool_svc *svc, struct pool_svc_event_set *event_set);
 
 static int
@@ -1489,6 +1530,9 @@ handle_event(struct pool_svc *svc, struct pool_svc_event_set *event_set)
 
 		if (event->psv_src != CRT_EVS_SWIM || event->psv_type != CRT_EVT_ALIVE)
 			continue;
+
+		pool_restart_rebuild_if_rank_wip(svc->ps_pool, event->psv_rank);
+
 		if (ds_pool_map_rank_up(svc->ps_pool->sp_map, event->psv_rank)) {
 			/*
 			 * The rank is up in the pool map. Request a pool map
diff --git a/src/rebuild/srv.c b/src/rebuild/srv.c
index 845f8b21d87..e051d606794 100644
--- a/src/rebuild/srv.c
+++ b/src/rebuild/srv.c
@@ -486,6 +486,45 @@ ds_rebuild_running_query(uuid_t pool_uuid, uint32_t opc, uint32_t *upper_ver,
 		rpt_put(rpt);
 }
 
+/*
+ * Restart rebuild if \a rank's rebuild not finished.
+ * Only used for massive failure recovery case, see pool_restart_rebuild_if_rank_wip().
+ */
+void
+ds_rebuild_restart_if_rank_wip(uuid_t pool_uuid, d_rank_t rank)
+{
+	struct rebuild_global_pool_tracker	*rgt;
+	int					 i;
+
+	rgt = rebuild_global_pool_tracker_lookup(pool_uuid, -1, -1);
+	if (rgt == NULL)
+		return;
+
+	if (rgt->rgt_status.rs_state != DRS_IN_PROGRESS) {
+		rgt_put(rgt);
+		return;
+	}
+
+	for (i = 0; i < rgt->rgt_servers_number; i++) {
+		if (rgt->rgt_servers[i].rank == rank) {
+			if (!rgt->rgt_servers[i].pull_done) {
+				rgt->rgt_status.rs_errno = -DER_STALE;
+				rgt->rgt_abort = 1;
+				rgt->rgt_status.rs_fail_rank = rank;
+				D_INFO(DF_RB ": abort rebuild because rank %d WIP\n",
+				       DP_RB_RGT(rgt), rank);
+			}
+			rgt_put(rgt);
+			return;
+		}
+	}
+
+	D_INFO(DF_RB ": rank %d not in rgt_servers,  rgt_servers_number %d\n",
+	       DP_RB_RGT(rgt), rank, rgt->rgt_servers_number);
+	rgt_put(rgt);
+	return;
+}
+
 /* TODO: Add something about what the current operation is for output status */
 int
 ds_rebuild_query(uuid_t pool_uuid, struct daos_rebuild_status *status)