【bugfix】 修复postmaster线程卡在PM_WAIT_BACKEND

src/gausskernel/process/postmaster/postmaster.cpp

@@ -486,6 +486,7 @@ static bool CheckSignalByFile(const char *filename, void *infoPtr, size_t infoSi
 
 int GaussDbThreadMain(knl_thread_arg* arg);
 const char* GetThreadName(knl_thread_role role);
+void SSOndemandProcExitIfStayWaitBackends();
 
 #ifdef EXEC_BACKEND
 
@@ -3976,6 +3977,10 @@ static int ServerLoop(void)
                             result = BackendStartup(port, isConnectHaPort);
                         }
 
+                        if (SS_CLUSTER_ONDEMAND_RECOVERY && SS_IN_REFORM &&
+                            result != STATUS_OK && pmState == PM_WAIT_BACKENDS) {
+                                SSOndemandProcExitIfStayWaitBackends();
+                        }
                         if (result != STATUS_OK) {
                             if (port->is_logic_conn) {
                                 gs_close_gsocket(&port->gs_sock);
@@ -15086,3 +15091,20 @@ void SSRestartFailoverPromote()
     pmState = PM_WAIT_BACKENDS;
     SShandle_promote_signal();
 }
+
+void SSOndemandProcExitIfStayWaitBackends() 
+{
+    int failTimes = 0;
+    while (failTimes < WAIT_PMSTATE_UPDATE_TRIES && pmState == PM_WAIT_BACKENDS) {
+        PostmasterStateMachine();
+        pg_usleep(REFORM_WAIT_LONG);
+        failTimes++;
+    }
+    if (pmState == PM_WAIT_BACKENDS) {
+        ereport(PANIC, (errmsg("Proc exit because pmState stay %s for %d times, "
+            "when reform failed and in ondemand recovery, "
+            "to avoid pmState being stuck in PM_WAIT_BACKENDS.", 
+            GetPMState(pmState), WAIT_PMSTATE_UPDATE_TRIES)));
+        proc_exit(1);
+    }
+}

src/include/ddes/dms/ss_reform_common.h

@@ -31,6 +31,8 @@
 #define REFORM_WAIT_LONG 100000 /* 0.1 sec */
 #define WAIT_REFORM_CTRL_REFRESH_TRIES 1000
 
+#define WAIT_PMSTATE_UPDATE_TRIES 100
+
 #define REFORM_CTRL_VERSION 1
 
 typedef struct SSBroadcastCancelTrx {