[4.2][Standby] Optimize error refresh in standby

src/logservice/restoreservice/ob_log_restore_handler.cpp

@@ -513,8 +513,12 @@ void ObLogRestoreHandler::mark_error(share::ObTaskId &trace_id,
     context_.error_context_.ret_code_ = ret_code;
     context_.error_context_.trace_id_.set(trace_id);
     context_.error_context_.err_lsn_ = lsn;
+    if (OB_TIMEOUT == ret_code && ObLogRestoreErrorContext::ErrorType::FETCH_LOG == error_type) {
+      CLOG_LOG(WARN, "fetch log timeout in restore", KPC(parent_), KPC(this));
+    } else if(OB_SUCCESS != ret_code) {
       CLOG_LOG(ERROR, "fatal error occur in restore", KPC(parent_), KPC(this));
     }
+  }
 }
 
 int ObLogRestoreHandler::get_restore_error(share::ObTaskId &trace_id, int &ret_code, bool &error_exist)

src/logservice/restoreservice/ob_log_restore_net_driver.cpp

@@ -201,10 +201,8 @@ int ObLogRestoreNetDriver::scan_ls(const share::ObLogRestoreSourceType &type)
           }
         } else if (OB_FAIL(check_ls_stale_(id, proposal_id, is_stale))) {
           LOG_WARN("check_ls_stale_ failed", K(id));
-        } else if (! is_stale) {
+        }  else if (!is_stale) {
-          if (OB_FAIL(refresh_error_context_(id))) {
+          // do nothing
-            LOG_WARN("refresh error context failed", K(id));
-          }
         } else if (OB_FAIL(fetcher_->remove_ls(id)) && OB_ENTRY_NOT_EXIST != ret) {
           LOG_WARN("remove ls failed", K(id), K(is_stale));
         } else {
@@ -573,18 +571,6 @@ int ObLogRestoreNetDriver::get_ls_count_in_fetcher_(int64_t &count)
   return ret;
 }
 
-int ObLogRestoreNetDriver::refresh_error_context_(const share::ObLSID &ls_id)
-{
-  int ret = OB_SUCCESS;
-  ObRemoteFetchContext context;
-  GET_RESTORE_HANDLER_CTX(ls_id) {
-    if (OB_FAIL(restore_handler->refresh_error_context())) {
-      LOG_WARN("refresh error failed");
-    }
-  }
-  return ret;
-}
-
 int ObLogRestoreNetDriver::set_restore_log_upper_limit()
 {
   RLockGuard guard(lock_);

src/logservice/restoreservice/ob_log_restore_net_driver.h

@@ -116,7 +116,6 @@ private:
 
   int check_ls_stale_(const share::ObLSID &id, const int64_t proposal_id, bool &is_stale);
   int get_ls_count_in_fetcher_(int64_t &count);
-  int refresh_error_context_(const share::ObLSID &ls_id);
 
 private:
   class LogErrHandler : public logfetcher::IObLogErrHandler

src/logservice/restoreservice/ob_log_restore_service.cpp

@@ -201,7 +201,7 @@ void ObLogRestoreService::do_thread_task_()
       last_normal_work_ts_ = common::ObTimeUtility::fast_current_time();
     }
     update_restore_upper_limit_();
-
+    refresh_error_context_();
     set_compressor_type_();
   }
 }
@@ -266,6 +266,40 @@ void ObLogRestoreService::set_compressor_type_()
   }
 }
 
+void ObLogRestoreService::refresh_error_context_()
+{
+  int ret = OB_SUCCESS;
+  ObLS *ls = NULL;
+  ObLSIterator *iter = NULL;
+  common::ObSharedGuard<ObLSIterator> guard;
+  ObLogRestoreHandler *restore_handler = NULL;
+  if (OB_FAIL(ls_svr_->get_ls_iter(guard, ObLSGetMod::LOG_MOD))) {
+    LOG_WARN("get ls iter failed", K(ret));
+  } else if (OB_ISNULL(iter = guard.get_ptr())) {
+    ret = OB_ERR_UNEXPECTED;
+    LOG_ERROR("iter is NULL", K(ret), K(iter));
+  } else {
+    while (OB_SUCC(ret)) {
+      ls = NULL;
+      if (OB_FAIL(iter->get_next(ls))) {
+        if (OB_ITER_END != ret) {
+          LOG_WARN("iter ls get next failed", K(ret));
+        } else {
+          LOG_TRACE("iter to end", K(ret));
+        }
+      } else if (OB_ISNULL(ls)) {
+        ret = OB_ERR_UNEXPECTED;
+        LOG_ERROR("ls is NULL", K(ret), K(ls));
+      } else if (OB_ISNULL(restore_handler = ls->get_log_restore_handler())) {
+        ret = OB_ERR_UNEXPECTED;
+        LOG_INFO("restore_handler is NULL", K(ret), K(ls->get_ls_id()));
+      } else if (OB_FAIL(restore_handler->refresh_error_context())) {
+        LOG_WARN("refresh error failed", K(ls->get_ls_id()));
+      }
+    }
+  }
+}
+
 bool ObLogRestoreService::need_schedule_() const
 {
   return common::ObTimeUtility::fast_current_time() - last_normal_work_ts_ > SCHEDULE_INTERVAL;

src/logservice/restoreservice/ob_log_restore_service.h

@@ -81,6 +81,7 @@ private:
   void report_error_();
   void update_restore_upper_limit_();
   void set_compressor_type_();
+  void refresh_error_context_();
   bool need_schedule_() const;
 
 private:

src/rootserver/ob_recovery_ls_service.cpp

@@ -66,6 +66,10 @@ ERRSIM_POINT_DEF(ERRSIM_END_TRANS_ERROR);
       ret_code = OB_PASSWORD_WRONG;                                            \
     case -ER_CONNECT_FAILED:                                                   \
       ret_code = OB_CONNECT_ERROR;                                             \
+    case OB_IN_STOP_STATE:                                                     \
+      ret_code = OB_IN_STOP_STATE;                                             \
+    default:                                                                   \
+      ret_code = ret;                                                          \
   }                                                                            \
   ROOTSERVICE_EVENT_ADD("root_service", "update_primary_ip_list",              \
     "tenant_id", tenant_id_, K(ret),                                           \