repmgr: improve standby clone when synchronous replication in use

Fixes GitHub #277

HISTORY

@@ -6,6 +6,7 @@
         repmgr: if replication slots in use, where possible delete slot on old
           upstream node after following new upstream (Ian)
         repmgr: improve logging of rsync actions (Ian)
+        repmgr: improve `standby clone` when synchronous replication in use (Ian)
 
 3.3     2016-12-27
         repmgr: always log to STDERR even if log facility defined (Ian)

dbutils.c

@@ -33,6 +33,7 @@ char repmgr_schema[MAXLEN] = "";
 char repmgr_schema_quoted[MAXLEN] = "";
 
 static int _get_node_record(PGconn *conn, char *cluster, char *sqlquery, t_node_info *node_info);
+static bool _set_config(PGconn *conn, const char *config_param, const char *sqlquery);
 
 PGconn *
 _establish_db_connection(const char *conninfo, const bool exit_on_error, const bool log_notice, const bool verbose_only)
@@ -1150,19 +1151,12 @@ stop_backup(PGconn *conn, char *last_wal_segment)
 }
 
 
+
 bool
-set_config_bool(PGconn *conn, const char *config_param, bool state)
+_set_config(PGconn *conn, const char *config_param, const char *sqlquery)
 {
-	char		sqlquery[QUERY_STR_LEN];
 	PGresult   *res;
 
-	sqlquery_snprintf(sqlquery,
-					  "SET %s TO %s",
-					  config_param,
-					  state ? "TRUE" : "FALSE");
-
-	log_verbose(LOG_DEBUG, "set_config_bool():\n%s\n", sqlquery);
-
 	res = PQexec(conn, sqlquery);
 
 	if (PQresultStatus(res) != PGRES_COMMAND_OK)
@@ -1177,6 +1171,36 @@ set_config_bool(PGconn *conn, const char *config_param, bool state)
 	return true;
 }
 
+bool
+set_config(PGconn *conn, const char *config_param,  const char *config_value)
+{
+	char		sqlquery[QUERY_STR_LEN];
+
+	sqlquery_snprintf(sqlquery,
+					  "SET %s TO '%s'",
+					  config_param,
+					  config_value);
+
+	log_verbose(LOG_DEBUG, "set_config():\n%s\n", sqlquery);
+
+	return _set_config(conn, config_param, sqlquery);
+}
+
+bool
+set_config_bool(PGconn *conn, const char *config_param, bool state)
+{
+	char		sqlquery[QUERY_STR_LEN];
+
+	sqlquery_snprintf(sqlquery,
+					  "SET %s TO %s",
+					  config_param,
+					  state ? "TRUE" : "FALSE");
+
+	log_verbose(LOG_DEBUG, "set_config_bool():\n%s\n", sqlquery);
+
+	return _set_config(conn, config_param, sqlquery);
+}
+
 
 /*
  * witness_copy_node_records()

dbutils.h

@@ -124,6 +124,7 @@ int			get_slot_record(PGconn *conn, char *slot_name, t_replication_slot *record)
 bool		drop_replication_slot(PGconn *conn, char *slot_name);
 bool		start_backup(PGconn *conn, char *first_wal_segment, bool fast_checkpoint);
 bool		stop_backup(PGconn *conn, char *last_wal_segment);
+bool		set_config(PGconn *conn, const char *config_param,  const char *config_value);
 bool		set_config_bool(PGconn *conn, const char *config_param, bool state);
 bool		witness_copy_node_records(PGconn *masterconn, PGconn *witnessconn, char *cluster_name);
 bool		create_node_record(PGconn *conn, char *action, int node, char *type, int upstream_node, char *cluster_name, char *node_name, char *conninfo, int priority, char *slot_name, bool active);

repmgr.c

@@ -758,7 +758,7 @@ main(int argc, char **argv)
 	{
 		if (optind < argc)
 		{
-			if (runtime_options.host_param_provided == true)
+			if (runtime_options.host[0])
 			{
 				PQExpBufferData additional_host_arg;
 				initPQExpBuffer(&additional_host_arg);
@@ -2954,6 +2954,8 @@ do_standby_clone(void)
 				primary_conn = source_conn;
 			}
 
+
+
 			/*
 			 * Sanity-check that the master node has a repmgr schema - if not
 			 * present, fail with an error (unless -F/--force is used)
@@ -2973,6 +2975,7 @@ do_standby_clone(void)
 				log_warning(_("expected repmgr schema '%s' not found on master server\n"), get_repmgr_schema());
 			}
 
+
 			/* Fetch the source's data directory */
 			if (get_pg_setting(source_conn, "data_directory", master_data_directory) == false)
 			{
@@ -2994,6 +2997,8 @@ do_standby_clone(void)
 				log_hint(_("use -D/--data-dir to explicitly specify a data directory\n"));
 			}
 
+
+
 			/*
 			 * Copy the source connection so that we have some default values,
 			 * particularly stuff like passwords extracted from PGPASSFILE;
@@ -3017,6 +3022,18 @@ do_standby_clone(void)
 				strncpy(recovery_conninfo_str, upstream_node_record.conninfo_str, MAXLEN);
 			}
 		}
+
+		/* Finally, set `synchronous_commit` to `local` to avoid problems
+		 * if synchronous commit is in use.
+		 */
+
+		if (primary_conn != NULL && PQstatus(primary_conn) == CONNECTION_OK)
+		{
+			if (set_config(primary_conn, "synchronous_commit", "local") == false)
+			{
+				exit(ERR_DB_QUERY);
+			}
+		}
 	}
 
 	if (mode == barman && PQstatus(source_conn) != CONNECTION_OK)
@@ -3578,9 +3595,6 @@ do_standby_clone(void)
 			/*
 			 * We must create some PGDATA subdirectories because they are
 			 * not included in the Barman backup.
-			 *
-			 * See class RsyncBackupExecutor in the Barman source (barman/backup_executor.py)
-			 * for a definitive list of excluded directories.
 			 */
 			{
 				const char* const dirs[] = {
@@ -3591,14 +3605,14 @@ do_standby_clone(void)
 					/* Only from 9.4 */
 					"pg_dynshmem", "pg_logical", "pg_logical/snapshots", "pg_logical/mappings", "pg_replslot",
 					/* Already in 9.3 */
-					"pg_notify", "pg_serial", "pg_snapshots", "pg_stat", "pg_stat_tmp", "pg_tblspc",
+					"pg_serial", "pg_snapshots", "pg_stat", "pg_stat_tmp", "pg_tblspc",
 					"pg_twophase", "pg_xlog", 0
 				};
 				const int vers[] = {
 					100000,
 					90500,
 					90400, 90400, 90400, 90400, 90400,
-					0, 0, 0, 0, 0, 0,
+					0, 0, 0, 0, 0,
 					0, -100000, 0
 				};
 				for (i = 0; dirs[i]; i++)
@@ -3628,17 +3642,6 @@ do_standby_clone(void)
 				initPQExpBuffer(&tablespace_map);
 			}
 
-			/*
-			 * From 9.1 default is to wait for a sync standby to ack, avoid that by
-			 * turning off sync rep for this session
-			 */
-			if (set_config_bool(source_conn, "synchronous_commit", false) == false)
-			{
-				r = ERR_BAD_CONFIG;
-				retval = ERR_BAD_CONFIG;
-				goto stop_backup;
-			}
-
 			if (start_backup(source_conn, first_wal_segment, runtime_options.fast_checkpoint) == false)
 			{
 				r = ERR_BAD_BASEBACKUP;