Merge branch 'feature_patch_432_core' into '4_3_2_release'

oceanbase · Oct 21, 2024 · bc486e6 · bc486e6
1 parent 2f32144
commit bc486e6
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 0 deletions.
diff --git a/src/obproxy/ob_proxy_main.cpp b/src/obproxy/ob_proxy_main.cpp
@@ -898,6 +898,7 @@ void ObProxyMain::sig_direct_handler(const int sig)
           if (info.need_conn_accept_) {
             info.disable_net_accept();  // disable accecpt new connection
           }
+          MEM_BARRIER();
           info.graceful_exit_start_time_ = get_hrtime_internal();
           info.graceful_exit_end_time_ = HRTIME_USECONDS(get_global_proxy_config().hot_upgrade_exit_timeout)
                                          + info.graceful_exit_start_time_;
@@ -925,10 +926,17 @@ void ObProxyMain::sig_direct_handler(const int sig)
         LOG_INFO("gcov flush now");
         __gcov_flush();
       #endif
+      LOG_INFO("recv signal, will graceful exit", K(sig), K(info));
       info.received_sig_ = sig;
       if (info.need_conn_accept_) {
         info.disable_net_accept();  // disable accecpt new connection
       }
+      // 1. disable_net_accept will call pthread_kill() to accept thread
+      // 2. when graceful_exit_start_time_ is changed, ObInactivityCop::check_inactivity
+      //    call pthread_cancel() and pthread_join() to accept thread
+      // - to avoid complex parallel multi-thread status and a lower-probability core problem (2024062100102854439),
+      //   need use MEM_BARRIER to restrict pthread_kill() happens before pthread_cancel()
+      MEM_BARRIER();
       info.graceful_exit_start_time_ = get_hrtime_internal();
       info.graceful_exit_end_time_ = HRTIME_USECONDS(get_global_proxy_config().delay_exit_time)
                                      + info.graceful_exit_start_time_;

diff --git a/src/obproxy/obutils/ob_hot_upgrade_processor.cpp b/src/obproxy/obutils/ob_hot_upgrade_processor.cpp
@@ -826,6 +826,7 @@ int ObHotUpgradeProcessor::state_wait_hu_cmd(const ObProxyServerInfo &proxy_info
       }
       if (OB_SUCC(ret) || info_.is_local_exit()) {
         info_.disable_net_accept();
+        MEM_BARRIER();
         info_.is_parent_ = true;// move to parent site, no mater on which site before
         info_.update_parent_status(HU_STATUS_DO_QUICK_EXIT);
         info_.graceful_exit_start_time_ = get_hrtime_internal();
@@ -942,6 +943,7 @@ int ObHotUpgradeProcessor::state_parent_wait_cr_cmd(const ObProxyServerInfo &pro
         //4. cancel timeout_rollback
 
         info_.disable_net_accept();
+        MEM_BARRIER();
         info_.graceful_exit_start_time_ = get_hrtime_internal();
         info_.graceful_exit_end_time_ = HRTIME_USECONDS(get_global_proxy_config().hot_upgrade_exit_timeout)
                                         + info_.graceful_exit_start_time_;// nanosecond
@@ -1090,6 +1092,7 @@ int ObHotUpgradeProcessor::state_sub_wait_cr_cmd()
         //4. cancel timeout_rollback job
 
         info_.disable_net_accept();
+        MEM_BARRIER();
         info_.graceful_exit_start_time_ = get_hrtime_internal();
         info_.graceful_exit_end_time_ = HRTIME_USECONDS(get_global_proxy_config().hot_upgrade_exit_timeout)
                                         + info_.graceful_exit_start_time_; // nanosecond
@@ -1746,6 +1749,7 @@ int ObHotUpgradeProcessor::do_hot_upgrade_work()
         }
         case HUC_LOCAL_EXIT: {
           info_.disable_net_accept();
+          MEM_BARRIER();
           info_.graceful_exit_start_time_ = get_hrtime_internal();
           info_.graceful_exit_end_time_ = HRTIME_USECONDS(get_global_proxy_config().delay_exit_time)
             + info_.graceful_exit_start_time_; // nanosecond