fix check_snapshot_gc_scn in case of cluster restarted

src/rootserver/freeze/ob_freeze_info_manager.cpp

@@ -31,6 +31,7 @@
 #include "lib/utility/ob_tracepoint.h"
 #include "storage/tx/ob_ts_mgr.h"
 #include "storage/tx/wrs/ob_weak_read_util.h"
+#include "share/ob_server_table_operator.h"
 
 namespace oceanbase
 {
@@ -620,6 +621,8 @@ int ObFreezeInfoManager::check_snapshot_gc_scn()
   SCN cur_gts_scn;
   SCN snapshot_gc_scn;
   int64_t delay = 0;
+  int64_t start_service_time = -1;
+  int64_t total_service_time = -1;
 
   ObRecursiveMutexGuard guard(lock_);
   if (OB_FAIL(check_inner_stat())) {
@@ -638,8 +641,23 @@ int ObFreezeInfoManager::check_snapshot_gc_scn()
 
       if (TC_REACH_TIME_INTERVAL(60 * 1000 * 1000)) {
         if (delay > SNAPSHOT_GC_TS_ERROR) {
-          LOG_ERROR("rs_monitor_check : snapshot_gc_ts delay for a long time",
-                    K(snapshot_gc_time_us), K(delay), K_(tenant_id));
+          // In order to avoid LOG_ERROR when the tenant reloads old snapshot_gc_scn due to the
+          // cluster restarted. LOG_ERROR should satisfy two additional conditions:
+          // 1. start_service_time > 0. start_service_time is initialized to 0 when observer starts.
+          // Then it will be updated to the time when observer starts through heartbeat, which is
+          // scheduled every 2 seconds.
+          // 2. total_service_time > SNAPSHOT_GC_TS_ERROR.
+          ObServerTableOperator st_operator;
+          if (OB_FAIL(st_operator.init(sql_proxy_))) {
+            LOG_WARN("fail to init server table operator", K(ret), K_(tenant_id));
+          } else if (OB_FAIL(st_operator.get_start_service_time(GCONF.self_addr_, start_service_time))) {
+            LOG_WARN("fail to get start service time", KR(ret), K_(tenant_id));
+          } else if (FALSE_IT(total_service_time = ObTimeUtility::current_time() - start_service_time)) {
+          } else if ((start_service_time > 0) && (total_service_time > SNAPSHOT_GC_TS_ERROR)) {
+            LOG_ERROR("rs_monitor_check : snapshot_gc_ts delay for a long time",
+                      K(snapshot_gc_time_us), K(delay), K_(tenant_id), K(start_service_time),
+                      K(total_service_time));
+          }
         } else if (delay > SNAPSHOT_GC_TS_WARN) {
           LOG_WARN("rs_monitor_check : snapshot_gc_ts delay for a long time",
                   K(snapshot_gc_time_us), K(delay), K_(tenant_id));

src/rootserver/freeze/ob_freeze_info_manager.h

@@ -44,7 +44,7 @@ public:
   share::SCN latest_snapshot_gc_scn_;
 
   ObFreezeInfo()
-    : frozen_statuses_(), latest_snapshot_gc_scn_()
+    : frozen_statuses_(), latest_snapshot_gc_scn_(share::SCN::min_scn())
   {}
   ~ObFreezeInfo() {}
 

src/rootserver/freeze/ob_major_merge_scheduler.cpp

@@ -786,7 +786,7 @@ void ObMajorMergeScheduler::check_merge_interval_time(const bool is_merging)
   int64_t global_merge_start_time = -1;
   int64_t max_merge_time = -1;
   int64_t start_service_time = -1;
-  int64_t all_service_time = -1;
+  int64_t total_service_time = -1;
   if (OB_ISNULL(zone_merge_mgr_)) {
     ret = OB_ERR_UNEXPECTED;
     LOG_WARN("zone_merge_mgr_ is unexpected nullptr", KR(ret), K_(tenant_id));
@@ -815,18 +815,23 @@ void ObMajorMergeScheduler::check_merge_interval_time(const bool is_merging)
       } else if (OB_FAIL(st_operator.get_start_service_time(GCONF.self_addr_, start_service_time))) {
         LOG_WARN("fail to get start service time", KR(ret), K_(tenant_id));
       } else {
-        all_service_time = now - start_service_time;
+        total_service_time = now - start_service_time;
       }
     }
-    // LOG_ERROR should satisfy one additional condition: all_service_time > MAX_NO_MERGE_INTERVAL.
-    // So as to avoid LOG_ERROR when the tenant miss daily merge due to the cluster restarted.
-    if (OB_SUCC(ret) && !is_paused() && (all_service_time > MAX_NO_MERGE_INTERVAL)) {
+    // In order to avoid LOG_ERROR when the tenant miss daily merge due to the cluster restarted.
+    // LOG_ERROR should satisfy two additional condition:
+    // 1. start_service_time > 0. start_service_time is initialized to 0 when observer starts.
+    // Then it will be updated to the time when observer starts through heartbeat, which is
+    // scheduled every 2 seconds.
+    // 2. total_service_time > MAX_NO_MERGE_INTERVAL.
+    if (OB_SUCC(ret) && !is_paused() && (start_service_time > 0)
+        && (total_service_time > MAX_NO_MERGE_INTERVAL)) {
       if (is_merging) {
         if ((now - max_merge_time) > MAX_NO_MERGE_INTERVAL) {
           if (TC_REACH_TIME_INTERVAL(30 * 60 * 1000 * 1000)) {
             LOG_ERROR("long time major freeze not finish, please check it", KR(ret),
               K(global_last_merged_time), K(global_merge_start_time), K(max_merge_time),
-              K(now), K_(tenant_id), K(is_merging), K(start_service_time), K(all_service_time));
+              K(now), K_(tenant_id), K(is_merging), K(start_service_time), K(total_service_time));
           }
         }
       } else {
@@ -840,7 +845,7 @@ void ObMajorMergeScheduler::check_merge_interval_time(const bool is_merging)
           if (TC_REACH_TIME_INTERVAL(30 * 60 * 1000 * 1000)) {
             LOG_ERROR("long time no major freeze, please check it", KR(ret),
               K(global_last_merged_time), K(global_merge_start_time), K(max_merge_time),
-              K(now), K_(tenant_id), K(is_merging), K(start_service_time), K(all_service_time));
+              K(now), K_(tenant_id), K(is_merging), K(start_service_time), K(total_service_time));
           }
         }
       }