From b472eb0cd37721e7855e56986d45fa80545a447b Mon Sep 17 00:00:00 2001
From: bowenliu <liubowen18@huawei.com>
Date: Tue, 19 Sep 2023 16:41:24 +0800
Subject: [PATCH] fix reform timeout due to twophasecleaner and backend term
 deadlock

---
 src/gausskernel/ddes/adapter/ss_dms_callback.cpp       | 10 +++-------
 src/gausskernel/process/postmaster/twophasecleaner.cpp |  7 +++++++
 2 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/src/gausskernel/ddes/adapter/ss_dms_callback.cpp b/src/gausskernel/ddes/adapter/ss_dms_callback.cpp
index 66dbc94fbc..5929d493b5 100644
--- a/src/gausskernel/ddes/adapter/ss_dms_callback.cpp
+++ b/src/gausskernel/ddes/adapter/ss_dms_callback.cpp
@@ -389,9 +389,6 @@ static int CBSwitchoverPromote(void *db_handle, unsigned char origPrimaryId)
         if (g_instance.dms_cxt.SSClusterState == NODESTATE_STANDBY_PROMOTED) {
             /* flush control file primary id in advance to save new standby's waiting time */
             SSSavePrimaryInstId(SS_MY_INST_ID);
-
-            SSReadControlFile(REFORM_CTRL_PAGE);
-            Assert(SSGetPrimaryInstId() == SS_MY_INST_ID);
             ereport(LOG, (errmodule(MOD_DMS),
                 errmsg("[SS switchover] Standby promote: success, set new primary:%d.", SS_MY_INST_ID)));
             return DMS_SUCCESS;
@@ -429,16 +426,15 @@ static void CBSwitchoverResult(void *db_handle, int result)
 
 static int SetPrimaryIdOnStandby(int primary_id)
 {
-    g_instance.dms_cxt.SSReformerControl.primaryInstId = primary_id;
-
     for (int ntries = 0;; ntries++) {
         SSReadControlFile(REFORM_CTRL_PAGE); /* need to double check */
         if (g_instance.dms_cxt.SSReformerControl.primaryInstId == primary_id) {
             ereport(LOG, (errmodule(MOD_DMS),
-                errmsg("[SS %s] Reform success, this is a standby:%d confirming new primary:%d.",
-                    SS_PERFORMING_SWITCHOVER ? "switchover" : "reform", SS_MY_INST_ID, primary_id)));
+                errmsg("[SS %s] Reform success, this is a standby:%d confirming new primary:%d, confirm ntries=%d.",
+                    SS_PERFORMING_SWITCHOVER ? "switchover" : "reform", SS_MY_INST_ID, primary_id, ntries)));
             return DMS_SUCCESS;
         } else {
+            SSSavePrimaryInstId(primary_id);
             if (ntries >= WAIT_REFORM_CTRL_REFRESH_TRIES) {
                 ereport(ERROR,
                     (errmodule(MOD_DMS), errmsg("[SS %s] Failed to confirm new primary: %d,"
diff --git a/src/gausskernel/process/postmaster/twophasecleaner.cpp b/src/gausskernel/process/postmaster/twophasecleaner.cpp
index 52698e049e..5051e3710e 100644
--- a/src/gausskernel/process/postmaster/twophasecleaner.cpp
+++ b/src/gausskernel/process/postmaster/twophasecleaner.cpp
@@ -205,6 +205,12 @@ NON_EXEC_STATIC void TwoPhaseCleanerMain()
 
         if (t_thrd.tpcleaner_cxt.shutdown_requested) {
             /* Normal exit from the twophasecleaner is here */
+            ereport(LOG, (errmsg("TwoPhaseCleaner exits via SIGTERM")));
+            proc_exit(0);
+        }
+
+        if (SS_PRIMARY_DEMOTING) {
+            ereport(LOG, (errmsg("TwoPhaseCleaner exits via SS global var")));
             proc_exit(0);
         }
 
@@ -326,6 +332,7 @@ static void TwoPCShutdownHandler(SIGNAL_ARGS)
 {
     int save_errno = errno;
 
+    ereport(LOG, (errmsg("TwoPhaseCleaner received SIGTERM")));
     t_thrd.tpcleaner_cxt.shutdown_requested = true;
 
     if (t_thrd.proc)
-- 
Gitee