repmgrd: further BDR sanity checks

dbutils.c

@@ -2756,8 +2756,8 @@ reset_voting_status(PGconn *conn)
 	// COMMAND_OK?
 	if (!res || PQresultStatus(res) != PGRES_TUPLES_OK)
 	{
-		log_error(_("unable to execute repmgr..reset_voting_status():\n  %s"),
+		log_error(_("unable to execute repmgr.reset_voting_status():\n  %s"),
-				PQerrorMessage(conn));
+				  PQerrorMessage(conn));
 	}
 
 	PQclear(res);
@@ -2775,6 +2775,7 @@ get_last_wal_receive_location(PGconn *conn)
 	PGresult		   *res;
 	XLogRecPtr		    ptr = InvalidXLogRecPtr;
 
+	// pre-10 !!!
 	res = PQexec(conn, "SELECT pg_catalog.pg_last_wal_receive_lsn()");
 
 	if (PQresultStatus(res) == PGRES_TUPLES_OK)
@@ -2822,6 +2823,41 @@ is_bdr_db(PGconn *conn)
 }
 
 
+bool
+is_active_bdr_node(PGconn *conn, const char *node_name)
+{
+	PQExpBufferData	  query;
+	PGresult		 *res;
+	bool			  is_active_bdr_node;
+
+	initPQExpBuffer(&query);
+	appendPQExpBuffer(
+		&query,
+		"    SELECT COALESCE(s.active, TRUE) AS active"
+		"      FROM bdr.bdr_nodes n "
+        " LEFT JOIN pg_replication_slots s "
+		"        ON slot_name=bdr.bdr_format_slot_name(n.node_sysid, n.node_timeline, n.node_dboid, (SELECT oid FROM pg_database WHERE datname = current_database())) "
+		"     WHERE node_name='%s' ",
+		node_name);
+
+	res = PQexec(conn, query.data);
+	termPQExpBuffer(&query);
+
+	if (PQresultStatus(res) != PGRES_TUPLES_OK || PQntuples(res) == 0)
+	{
+		is_active_bdr_node = false;
+	}
+	else
+	{
+		is_active_bdr_node = atoi(PQgetvalue(res, 0, 0)) == 1 ? true : false;
+	}
+
+	PQclear(res);
+
+	return is_active_bdr_node;
+}
+
+
 bool
 is_bdr_repmgr(PGconn *conn)
 {
@@ -3034,6 +3070,57 @@ get_all_bdr_node_records(PGconn *conn, BdrNodeInfoList *node_list)
 	return;
 }
 
+RecordStatus
+get_bdr_node_record_by_name(PGconn *conn, const char *node_name, t_bdr_node_info *node_info)
+{
+	PQExpBufferData	  query;
+	PGresult		 *res;
+
+	initPQExpBuffer(&query);
+
+	appendPQExpBuffer(
+		&query,
+		"  SELECT node_sysid, "
+		"         node_timeline, "
+		"         node_dboid, "
+		"         node_status, "
+		"         node_name, "
+		"         node_local_dsn, "
+		"         node_init_from_dsn, "
+		"         node_read_only, "
+		"         node_seq_id "
+		"    FROM bdr.bdr_nodes "
+		"   WHERE node_name = '%s'",
+		node_name);
+
+	log_verbose(LOG_DEBUG, "get_bdr_node_record_by_name():\n%s", query.data);
+
+	res = PQexec(conn, query.data);
+	termPQExpBuffer(&query);
+
+	if (!res || PQresultStatus(res) != PGRES_TUPLES_OK)
+	{
+		log_error(_("unable to retrieve BDR node record for \"%s\":\n  %s"),
+				  node_name,
+				  PQerrorMessage(conn));
+
+		PQclear(res);
+		return RECORD_ERROR;
+	}
+
+	if (PQntuples(res) == 0)
+	{
+		PQclear(res);
+		return RECORD_NOT_FOUND;
+	}
+
+	_populate_bdr_node_record(res, node_info, 0);
+
+	PQclear(res);
+
+	return RECORD_FOUND;
+}
+
 
 static
 void _populate_bdr_node_records(PGresult *res, BdrNodeInfoList *node_list)

dbutils.h

@@ -310,7 +310,9 @@ XLogRecPtr get_last_wal_receive_location(PGconn *conn);
 
 /* BDR functions */
 void		 get_all_bdr_node_records(PGconn *conn, BdrNodeInfoList *node_list);
+RecordStatus get_bdr_node_record_by_name(PGconn *conn, const char *node_name, t_bdr_node_info *node_info);
 bool		 is_bdr_db(PGconn *conn);
+bool		 is_active_bdr_node(PGconn *conn, const char *node_name);
 bool		 is_bdr_repmgr(PGconn *conn);
 bool		 is_table_in_bdr_replication_set(PGconn *conn, const char *tablename, const char *set);
 bool		 add_table_to_bdr_replication_set(PGconn *conn, const char *tablename, const char *set);

repmgrd-bdr.c

@@ -24,11 +24,12 @@ do_bdr_node_check(void)
 void
 monitor_bdr(void)
 {
-	NodeInfoList  nodes = T_NODE_INFO_LIST_INITIALIZER;
+	NodeInfoList    nodes = T_NODE_INFO_LIST_INITIALIZER;
-	PGconn		 *monitoring_conn = NULL;
+	PGconn		   *monitoring_conn = NULL;
-	t_node_info	 *monitored_node = NULL;
+	t_node_info	   *monitored_node = NULL;
-	RecordStatus  record_status;
+	t_bdr_node_info bdr_node_info = T_BDR_NODE_INFO_INITIALIZER;
 
+	RecordStatus  record_status;
 	bool failover_done = false;
 
 	/* sanity check local database */
@@ -72,6 +73,16 @@ monitor_bdr(void)
 		exit(ERR_BAD_CONFIG);
 	}
 
+	record_status = get_bdr_node_record_by_name(local_conn, local_node_info.node_name, &bdr_node_info);
+
+	if (record_status != RECORD_FOUND)
+	{
+		log_error(_("unable to retrieve BDR record for node %s, terminating"),
+				  local_node_info.node_name);
+		PQfinish(local_conn);
+		exit(ERR_BAD_CONFIG);
+	}
+
 	/* Retrieve record for this node from the local database */
 	record_status = get_node_record(local_conn, config_file_options.node_id, &local_node_info);
 
@@ -88,6 +99,7 @@ monitor_bdr(void)
 		exit(ERR_BAD_CONFIG);
 	}
 
+
 	if (local_node_info.active == false)
 	{
 		log_error(_("local node (ID: %i) is marked as inactive in repmgr"),
@@ -97,6 +109,14 @@ monitor_bdr(void)
 		exit(ERR_BAD_CONFIG);
 	}
 
+	if (is_active_bdr_node(local_conn, local_node_info.node_name))
+	{
+		log_error(_("BDR node %s is not active, terminating"),
+				  local_node_info.node_name);
+		PQfinish(local_conn);
+		exit(ERR_BAD_CONFIG);
+	}
+
 	/* Log startup event */
 
 	create_event_record(local_conn,