Add "--checksum" in rsync when using "--force"

If the user don't put that option in rsync_options using of "--force"
could be unsafe.
While the probability of failures because of this are low they aren't
zero.

HISTORY

@@ -31,3 +31,10 @@
 1.1.0   2011-03-09
         Make options -U, -R and -p not mandatory (Jaime)
 
+1.1.1	2012-04-18
+        Add --ignore-rsync-warning (Cédric)
+        Add strnlen for compatibility with OS X (Greg)
+        Improve performance of repl_status view (Jaime) 
+        Remove last argument from log_err (Jaime, Reported by Jeroen Dekkers)
+        Complete documentation about possible error conditions (Jaime)
+        Document how to clean history (Jaime)

README.rst

@@ -814,6 +814,23 @@ and on "prime."
 
 The servers are now again acting as primary on "prime" and standby on "standby".
 
+Maintainance of monitor history
+-------------------------------
+
+Once you have changed roles (with a failover or to restore original roles)
+you would end up with records saying that node1 is primary and other records
+saying that node2 is the primary. Which could be confusing.
+Also, if you don't do anything about it the monitor history will keep growing.
+For both of those reasons you sometime want to make some maintainance of the 
+``repl_monitor`` table.
+
+If you want to clean the history after a few days you can execute a 
+truncate/delete (wheter you want to completely clean history or want to keep
+a few days of history) in a cron. For example to keep just one day of history
+you can put this in your crontab::
+
+0 1 * * *   psql -c "DELETE FROM repmgr_schema.repl_monitor where now() - last_monitor_time >= '1 day'::interval;" postgres
+
 Configuration and command reference
 ===================================
 
@@ -863,6 +880,7 @@ The output from this program looks like this::
     -R, --remote-user=USERNAME database server username for rsync
     -w, --wal-keep-segments=VALUE  minimum value for the GUC wal_keep_segments (default: 5000)
     -F, --force                force potentially dangerous operations to happen
+    -I, --ignore-rsync-warning Ignore partial transfert warning
 
   repmgr performs some tasks like clone a node, promote it or making follow another node and then exits.
   COMMANDS:
@@ -1023,6 +1041,7 @@ following
 * ERR_DB_QUERY 7:  Error executing a database query.
 * ERR_PROMOTED 8:  Exiting program because the node has been promoted to master.
 * ERR_BAD_PASSWORD 9:  Password used to connect to a database was rejected.
+* ERR_STR_OVERFLOW 10: A string was larger than expected.
 
 License and Contributions
 =========================

dbutils.h

@@ -20,6 +20,8 @@
 #ifndef _REPMGR_DBUTILS_H_
 #define _REPMGR_DBUTILS_H_
 
+#include "strutil.h"
+
 PGconn *establishDBConnection(const char *conninfo, const bool exit_on_error);
 PGconn *establishDBConnectionByParams(const char *keywords[],
                                       const char *values[],

repmgr.c

@@ -28,6 +28,7 @@
 
 #include <stdio.h>
 #include <stdlib.h>
+#include <sys/wait.h>
 #include <time.h>
 #include <unistd.h>
 
@@ -71,7 +72,7 @@ bool need_a_node = true;
 bool require_password = false;
 
 /* Initialization of runtime options */
-t_runtime_options runtime_options = { "", "", "", "", "", "", DEFAULT_WAL_KEEP_SEGMENTS, false, false, "" };
+t_runtime_options runtime_options = { "", "", "", "", "", "", DEFAULT_WAL_KEEP_SEGMENTS, false, false, false, "" };
 t_configuration_options options = { "", -1, "", "", "" };
 
 static char		*server_mode = NULL;
@@ -91,6 +92,7 @@ main(int argc, char **argv)
 		{"remote-user", required_argument, NULL, 'R'},
 		{"wal-keep-segments", required_argument, NULL, 'w'},
 		{"force", no_argument, NULL, 'F'},
+		{"ignore-rsync-warning", no_argument, NULL, 'I'},
 		{"verbose", no_argument, NULL, 'v'},
 		{NULL, 0, NULL, 0}
 	};
@@ -116,7 +118,7 @@ main(int argc, char **argv)
 	}
 
 
-	while ((c = getopt_long(argc, argv, "d:h:p:U:D:f:R:w:F:v", long_options,
+	while ((c = getopt_long(argc, argv, "d:h:p:U:D:f:R:w:FIv", long_options,
 	                        &optindex)) != -1)
 	{
 		switch (c)
@@ -150,6 +152,9 @@ main(int argc, char **argv)
 		case 'F':
 			runtime_options.force = true;
 			break;
+		case 'I':
+			runtime_options.ignore_rsync_warn = true;
+			break;
 		case 'v':
 			runtime_options.verbose = true;
 			break;
@@ -228,7 +233,7 @@ main(int argc, char **argv)
 		break;
 	default:
 		log_err(_("%s: too many command-line arguments (first extra is \"%s\")\n"),
-		        progname, argv[optind + 1]);
+		        progname, argv[optind]);
 		usage();
 		exit(ERR_BAD_CONFIG);
 	}
@@ -346,7 +351,7 @@ do_master_register(void)
 	log_info(_("%s connected to master, checking its state\n"), progname);
 	if (is_standby(conn))
 	{
-		log_err(_("%s needs master to be PostgreSQL 9.0 or better\n"), progname);
+		log_err(_("Trying to register a standby node as a master\n"));
 		PQfinish(conn);
 		exit(ERR_BAD_CONFIG);
 	}
@@ -438,14 +443,13 @@ do_master_register(void)
 
 		/* and the view */
 		sqlquery_snprintf(sqlquery, "CREATE VIEW %s.repl_status AS "
-		                  "  WITH monitor_info AS (SELECT *, ROW_NUMBER() OVER (PARTITION BY primary_node, standby_node "
-		                  " ORDER BY last_monitor_time desc) "
-		                  "  FROM %s.repl_monitor) "
 		                  "  SELECT primary_node, standby_node, last_monitor_time, last_wal_primary_location, "
 		                  "         last_wal_standby_location, pg_size_pretty(replication_lag) replication_lag, "
 		                  "         pg_size_pretty(apply_lag) apply_lag, age(now(), last_monitor_time) AS time_lag "
-		                  "    FROM monitor_info a "
-		                  "   WHERE row_number = 1", repmgr_schema, repmgr_schema);
+		                  "    FROM %s.repl_monitor "
+		                  "  WHERE (standby_node, last_monitor_time) IN (SELECT standby_node, MAX(last_monitor_time) "
+		                  "                                                FROM %s.repl_monitor GROUP BY 1)",
+		                  repmgr_schema, repmgr_schema, repmgr_schema);
 		log_debug("master register: %s\n", sqlquery);
 		if (!PQexec(conn, sqlquery))
 		{
@@ -454,6 +458,19 @@ do_master_register(void)
 			PQfinish(conn);
 			exit(ERR_BAD_CONFIG);
 		}
+
+		/* an index to improve performance of the view */
+		sqlquery_snprintf(sqlquery, "CREATE INDEX idx_repl_status_sort "
+		                  "    ON %s.repl_monitor (last_monitor_time, standby_node) ",
+		                  repmgr_schema);
+		log_debug(_("master register: %s\n"), sqlquery);
+		if (!PQexec(conn, sqlquery))
+		{
+			log_err(_("Cannot indexing table %s.repl_monitor: %s\n"),
+			        repmgr_schema, PQerrorMessage(conn));
+			PQfinish(conn);
+			exit(ERR_BAD_CONFIG);
+		}
 	}
 	else
 	{
@@ -845,6 +862,7 @@ do_standby_clone(void)
 				PQfinish(conn);
 				exit(ERR_BAD_CONFIG);
 			}
+			break;
 		default:
 			/* Trouble accessing directory */
 			log_err(_("%s: could not access directory \"%s\": %s\n"),
@@ -871,6 +889,16 @@ do_standby_clone(void)
 		PQfinish(conn);
 		exit(ERR_BAD_CONFIG);
 	}
+	
+	/* We need all 4 parameters, and they can be retrieved only by superusers */
+	if (PQntuples(res) != 4)
+	{
+		log_err("%s: STANDBY CLONE should be run by a SUPERUSER\n", progname);
+		PQclear(res);
+		PQfinish(conn);
+		exit(ERR_BAD_CONFIG);
+	}
+
 	for (i = 0; i < PQntuples(res); i++)
 	{
 		if (strcmp(PQgetvalue(res, i, 0), "data_directory") == 0)
@@ -886,6 +914,20 @@ do_standby_clone(void)
 	}
 	PQclear(res);
 
+	/*
+	 * in pg 9.1 default is to wait for a sync standby to ack,
+	 * avoid that by turning off sync rep for this session
+	 */
+	sqlquery_snprintf(sqlquery, "SET synchronous_commit TO OFF");
+	res = PQexec(conn, sqlquery);
+	if (PQresultStatus(res) != PGRES_COMMAND_OK)
+	{
+		log_err("Can't set synchronous_commit: %s\n", PQerrorMessage(conn));
+		PQclear(res);
+		PQfinish(conn);
+		exit(ERR_BAD_CONFIG);
+	}
+
 	/*
 	 * inform the master we will start a backup and get the first XLog filename
 	 * so we can say to the user we need those files
@@ -1022,9 +1064,6 @@ stop_backup:
 	 * Don't have this one exit if it fails, so that a more informative
 	 * error message will also appear about the backup not being stopped.
 	 */
-	log_info(_("%s connecting to master database to stop backup\n"), progname);
-	conn=establishDBConnectionByParams(keywords,values,false);
-
 	log_notice("Finishing backup...\n");
 	sqlquery_snprintf(sqlquery, "SELECT pg_xlogfile_name(pg_stop_backup())");
 	log_debug("standby clone: %s\n", sqlquery);
@@ -1039,8 +1078,10 @@ stop_backup:
 	}
 	last_wal_segment = PQgetvalue(res, 0, 0);
 
-	log_info(_("%s requires primary to keep WAL files %s until at least %s\n"),
-	         progname, first_wal_segment, last_wal_segment);
+	/* don't show this message if rsync failed */
+	if (r == 0)
+		log_info(_("%s requires primary to keep WAL files %s until at least %s\n"),
+		         progname, first_wal_segment, last_wal_segment);
 
 	/* Finished with the database connection now */
 	PQclear(res);
@@ -1337,6 +1378,7 @@ void help(const char *progname)
 	printf(_("	-R, --remote-user=USERNAME database server username for rsync\n"));
 	printf(_("	-w, --wal-keep-segments=VALUE  minimum value for the GUC wal_keep_segments (default: 5000)\n"));
 	printf(_("	-F, --force				   force potentially dangerous operations to happen\n"));
+	printf(_("	-I, --ignore-rsync-warning		   Ignore partial transfert warning\n"));
 
 	printf(_("\n%s performs some tasks like clone a node, promote it "), progname);
 	printf(_("or making follow another node and then exits.\n"));
@@ -1446,7 +1488,7 @@ copy_remote_files(char *host, char *remote_user, char *remote_path,
 		maxlen_snprintf(rsync_flags, "%s", options.rsync_options);
 
 	if (runtime_options.force)
-		strcat(rsync_flags, " --delete");
+		strcat(rsync_flags, " --delete --checksum");
 
 	if (!remote_user[0])
 	{
@@ -1473,6 +1515,29 @@ copy_remote_files(char *host, char *remote_user, char *remote_path,
 
 	r = system(script);
 
+	/*
+	 * If we are transfering a directory (ie: data directory, tablespace directories)
+	 * then we can ignore some rsync warning, so if we get some of those errors we
+	 * treat them as 0 if we have --ignore-rsync-warning commandline option set
+	 * List of ignorable rsync errors:
+	 * 24     Partial transfer due to vanished source files
+	 */
+	if ((WEXITSTATUS(r) == 24) && is_directory)
+	{
+		if (!runtime_options.ignore_rsync_warn)
+		{
+			log_warning( _("\nrsync completed with return code 24 "
+			               "\"Partial transfer due to vanished source files\".\n"
+			               "This can happen because of normal operation "
+			               "on the master server, but it may indicate an "
+			               "issue during cloning.  If you are certain no "
+			               "changes were made to the master, try cloning "
+			               "again using \"repmgr --force --ignore-rsync-warning\"."));
+			exit(ERR_BAD_RSYNC);
+		}
+		else
+			r = 0;
+	}
 	if (r != 0)
 		log_err(_("Can't rsync from remote file or directory (%s:%s)\n"),
 		        host_string, remote_path);

repmgr.h

@@ -55,6 +55,7 @@ typedef struct
 	char wal_keep_segments[MAXLEN];
 	bool verbose;
 	bool force;
+	bool ignore_rsync_warn;
 
 	char masterport[MAXLEN];
 

strutil.c

@@ -27,6 +27,15 @@
 
 static int xvsnprintf(char *str, size_t size, const char *format, va_list ap);
 
+/* Add strnlen on platforms that don't have it, like OS X */
+#ifndef strnlen
+size_t
+strnlen(const char *s, size_t n)
+{
+	const char *end = (const char *) memchr(s, '\0', n);
+	return(end ? end - s : n);
+}
+#endif
 
 static int
 xvsnprintf(char *str, size_t size, const char *format, va_list ap)

strutil.h

@@ -35,4 +35,9 @@ extern int xsnprintf(char *str, size_t size, const char *format, ...);
 extern int sqlquery_snprintf(char *str, const char *format, ...);
 extern int maxlen_snprintf(char *str, const char *format, ...);
 
+/* Add strnlen on platforms that don't have it, like OS X */
+#ifndef strnlen
+extern size_t strnlen(const char *s, size_t n);
+#endif
+
 #endif	/* _STRUTIL_H_ */