From bde8c7e29cd3b655a6171c5277d1eef48df42ff1 Mon Sep 17 00:00:00 2001
From: Ian Barwick <ian@2ndquadrant.com>
Date: Wed, 6 Mar 2019 14:54:05 +0900
Subject: [PATCH] repmgrd: handle reconnect to restarted server when using
 "connection" checks

---
 configfile.c       |  2 +-
 repmgrd-physical.c |  8 ++++----
 repmgrd.c          | 26 ++++++++++++++++----------
 repmgrd.h          |  2 +-
 4 files changed, 22 insertions(+), 16 deletions(-)

diff --git a/configfile.c b/configfile.c
index 1b6dd1ea..677e1920 100644
--- a/configfile.c
+++ b/configfile.c
@@ -632,7 +632,7 @@ _parse_config(t_configuration_options *options, ItemList *error_list, ItemList *
 			else
 			{
 				item_list_append(error_list,
-								 _("value for \"connection_check_type\" must be \"ping\" or \"connect\"\n"));
+								 _("value for \"connection_check_type\" must be \"ping\" or \"connection\"\n"));
 			}
 		}
 
diff --git a/repmgrd-physical.c b/repmgrd-physical.c
index 0280217b..2ffdaa92 100644
--- a/repmgrd-physical.c
+++ b/repmgrd-physical.c
@@ -831,7 +831,7 @@ monitor_streaming_standby(void)
 	while (true)
 	{
 		log_verbose(LOG_DEBUG, "checking %s", upstream_node_info.conninfo);
-		if (check_upstream_connection(upstream_conn, upstream_node_info.conninfo) == true)
+		if (check_upstream_connection(&upstream_conn, upstream_node_info.conninfo) == true)
 		{
 			set_upstream_last_seen(local_conn);
 		}
@@ -1030,7 +1030,7 @@ monitor_streaming_standby(void)
 					  upstream_node_info.node_id,
 					  degraded_monitoring_elapsed);
 
-			if (check_upstream_connection(upstream_conn, upstream_node_info.conninfo) == true)
+			if (check_upstream_connection(&upstream_conn, upstream_node_info.conninfo) == true)
 			{
 				if (config_file_options.connection_check_type == CHECK_PING)
 					upstream_conn = establish_db_connection(upstream_node_info.conninfo, false);
@@ -1605,7 +1605,7 @@ monitor_streaming_witness(void)
 
 	while (true)
 	{
-		if (check_upstream_connection(upstream_conn, upstream_node_info.conninfo) == false)
+		if (check_upstream_connection(&upstream_conn, upstream_node_info.conninfo) == false)
 		{
 			if (upstream_node_info.node_status == NODE_STATUS_UP)
 			{
@@ -1694,7 +1694,7 @@ monitor_streaming_witness(void)
 					  upstream_node_info.node_id,
 					  degraded_monitoring_elapsed);
 
-			if (check_upstream_connection(primary_conn, upstream_node_info.conninfo) == true)
+			if (check_upstream_connection(&primary_conn, upstream_node_info.conninfo) == true)
 			{
 				if (config_file_options.connection_check_type == CHECK_PING)
 					primary_conn = establish_db_connection(upstream_node_info.conninfo, false);
diff --git a/repmgrd.c b/repmgrd.c
index d62d3fa7..faa47fef 100644
--- a/repmgrd.c
+++ b/repmgrd.c
@@ -819,7 +819,7 @@ show_help(void)
 
 
 bool
-check_upstream_connection(PGconn *conn, const char *conninfo)
+check_upstream_connection(PGconn **conn, const char *conninfo)
 {
 	/* Check the connection status twice in case it changes after reset */
 	bool		twice = false;
@@ -829,30 +829,33 @@ check_upstream_connection(PGconn *conn, const char *conninfo)
 
 	for (;;)
 	{
-		if (PQstatus(conn) != CONNECTION_OK)
+		if (PQstatus(*conn) != CONNECTION_OK)
 		{
+			log_debug("connection not OK");
 			if (twice)
 				return false;
-			PQreset(conn);		/* reconnect */
+			/* reconnect */
+			PQfinish(*conn);
+			*conn = PQconnectdb(conninfo);
 			twice = true;
 		}
 		else
 		{
-			if (!cancel_query(conn, config_file_options.async_query_timeout))
+			if (!cancel_query(*conn, config_file_options.async_query_timeout))
 				goto failed;
 
-			if (wait_connection_availability(conn, config_file_options.async_query_timeout) != 1)
+			if (wait_connection_availability(*conn, config_file_options.async_query_timeout) != 1)
 				goto failed;
 
 			/* execute a simple query to verify connection availability */
-			if (PQsendQuery(conn, "SELECT 1") == 0)
+			if (PQsendQuery(*conn, "SELECT 1") == 0)
 			{
 				log_warning(_("unable to send query to upstream"));
-				log_detail("%s", PQerrorMessage(conn));
+				log_detail("%s", PQerrorMessage(*conn));
 				goto failed;
 			}
 
-			if (wait_connection_availability(conn, config_file_options.async_query_timeout) != 1)
+			if (wait_connection_availability(*conn, config_file_options.async_query_timeout) != 1)
 				goto failed;
 
 			break;
@@ -861,7 +864,10 @@ check_upstream_connection(PGconn *conn, const char *conninfo)
 			/* retry once */
 			if (twice)
 				return false;
-			PQreset(conn);		/* reconnect */
+
+			/* reconnect */
+			PQfinish(*conn);
+			*conn = PQconnectdb(conninfo);
 			twice = true;
 		}
 	}
@@ -926,7 +932,7 @@ try_reconnect(PGconn **conn, t_node_info *node_info)
 
 					if (ping_result != PGRES_TUPLES_OK)
 					{
-						log_info("original connnection no longer available, using new connection");
+						log_info("original connection no longer available, using new connection");
 						close_connection(conn);
 						*conn = our_conn;
 					}
diff --git a/repmgrd.h b/repmgrd.h
index 56c1df70..20fb0118 100644
--- a/repmgrd.h
+++ b/repmgrd.h
@@ -23,7 +23,7 @@ extern PGconn *local_conn;
 extern bool startup_event_logged;
 extern char pid_file[MAXPGPATH];
 
-bool		check_upstream_connection(PGconn *conn, const char *conninfo);
+bool		check_upstream_connection(PGconn **conn, const char *conninfo);
 void		try_reconnect(PGconn **conn, t_node_info *node_info);
 
 int			calculate_elapsed(instr_time start_time);