/* md_cluster_info flags */
 #define                MD_CLUSTER_WAITING_FOR_NEWDISK          1
 #define                MD_CLUSTER_SUSPEND_READ_BALANCING       2
+#define                MD_CLUSTER_BEGIN_JOIN_CLUSTER           3
 
 
 struct md_cluster_info {
        struct md_cluster_info *cinfo = mddev->cluster_info;
 
        cinfo->slot_number = our_slot;
-       complete(&cinfo->completion);
+       /* completion is only need to be complete when node join cluster,
+        * it doesn't need to run during another node's failure */
+       if (test_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state)) {
+               complete(&cinfo->completion);
+               clear_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
+       }
        clear_bit(MD_CLUSTER_SUSPEND_READ_BALANCING, &cinfo->state);
 }
 
+/* the ops is called when node join the cluster, and do lock recovery
+ * if node failure occurs */
 static const struct dlm_lockspace_ops md_ls_ops = {
        .recover_prep = recover_prep,
        .recover_slot = recover_slot,
        INIT_LIST_HEAD(&cinfo->suspend_list);
        spin_lock_init(&cinfo->suspend_lock);
        init_completion(&cinfo->completion);
+       set_bit(MD_CLUSTER_BEGIN_JOIN_CLUSTER, &cinfo->state);
 
        mutex_init(&cinfo->sb_mutex);
        mddev->cluster_info = cinfo;