!1277 【bugfix】解决备机故障时主机对应WalSender无法超时退出的问题

Merge pull request !1277 from chenxiaobin/walsndtimeout

src/common/backend/utils/error/elog.cpp

@@ -3639,7 +3639,15 @@ static void send_message_to_frontend(ErrorData* edata)
             !t_thrd.log_cxt.flush_message_immediately)
             return;
 
-        pq_flush();
+        /*
+         * If it is WalSender's timeout message, choose pq_flush_if_writable()
+         * to avoid blokcing in send() if the send buffer of the socket is full.
+         */
+        if (AM_WAL_NORMAL_SENDER && t_thrd.walsender_cxt.isWalSndSendTimeoutMessage) {
+            pq_flush_if_writable();
+        } else {
+            pq_flush();
+        }
 
         if (edata->elevel == FATAL)
             t_thrd.log_cxt.flush_message_immediately = true;

src/gausskernel/process/threadpool/knl_thread.cpp

@@ -1364,6 +1364,7 @@ static void knl_t_walsender_init(knl_t_walsender_context* walsender_cxt)
     walsender_cxt->remotePort = 0;
     walsender_cxt->walSndCaughtUp = false;
     walsender_cxt->last_check_timeout_timestamp = 0;
+    walsender_cxt->isWalSndSendTimeoutMessage = false;
     walsender_cxt->advancePrimaryConn = NULL;
 	
     walsender_cxt->ep_fd = -1;

src/gausskernel/storage/replication/walsender.cpp

@@ -3588,6 +3588,8 @@ static void WalSndCheckTimeOut(TimestampTz now)
          * standby.
          */
         if (log_min_messages <= ERROR || client_min_messages <= ERROR) {
+            t_thrd.walsender_cxt.isWalSndSendTimeoutMessage = true;
+
             WalReplicationTimestampInfo timeStampInfo;
             WalReplicationTimestampToString(&timeStampInfo, now, timeout, *last_reply_time, heartbeat);
             ereport(ERROR, (errmsg("terminating Walsender process due to replication timeout."),

src/include/knl/knl_thread.h

@@ -2283,7 +2283,8 @@ typedef struct knl_t_walsender_context {
     struct pg_conn* advancePrimaryConn;
     /* Timestamp of the last check-timeout time in WalSndCheckTimeOut. */
     TimestampTz last_check_timeout_timestamp;
-
+    /* flag set in WalSndCheckTimeout */
+    bool isWalSndSendTimeoutMessage;
 
     int datafd;
     int ep_fd;