From 863a3fab530daf21ebb1d4e7789a983ec854d120 Mon Sep 17 00:00:00 2001
From: Jaime Casanova <jaime@2ndQuadrant.com>
Date: Fri, 1 Oct 2010 02:26:39 -0500
Subject: [PATCH] A few changes from repmgrd and improve the SQL of the
 repl_status view in order to actually show something useful

---
 repmgr.sql | 22 +++++++++++++++++-----
 repmgrd.c  | 34 +++++++++++++++++++---------------
 2 files changed, 36 insertions(+), 20 deletions(-)

diff --git a/repmgr.sql b/repmgr.sql
index e7d48741..d625f441 100644
--- a/repmgr.sql
+++ b/repmgr.sql
@@ -26,11 +26,23 @@ CREATE TABLE repl_monitor (
 
 
 /*
- * A useful view 
+ * This view shows the latest monitor info about every node.
+ * Interesting thing to see:
+ * replication_lag: in bytes (this is how far the latest xlog record 
+ *                            we have received is from master)
+ * apply_lag: in bytes (this is how far the latest xlog record
+ *                      we have applied is from the latest record we 
+ *                      have received)
+ * time-lag: how many seconds are we from being up-to-date with master
  */
 drop view if exists repl_status;
 CREATE VIEW repl_status AS
-SELECT *, now() - (select max(last_monitor_time) from repl_monitor b
-                    where b.primary_node = a.primary_node
-                      and b.standby_node = a.standby_node)
-  FROM repl_monitor a;
+WITH monitor_info AS (SELECT *, ROW_NUMBER() OVER (PARTITION BY primary_node, standby_node
+                                                       ORDER BY last_monitor_time desc)
+                        FROM 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;
diff --git a/repmgrd.c b/repmgrd.c
index f76a4c09..ad89f28d 100644
--- a/repmgrd.c
+++ b/repmgrd.c
@@ -33,14 +33,14 @@ bool	verbose = false;
 
 
 static void help(const char *progname);
-void checkClusterConfiguration(void);
-void checkNodeConfiguration(char *conninfo);
-void getPrimaryConnection(void);
+static void checkClusterConfiguration(void);
+static void checkNodeConfiguration(char *conninfo);
+static void getPrimaryConnection(void);
 
-void MonitorCheck(void);
-void MonitorExecute(void);
+static void MonitorCheck(void);
+static void MonitorExecute(void);
 
-unsigned long long int walLocationToBytes(char *wal_location);
+static unsigned long long int walLocationToBytes(char *wal_location);
 
 
 int
@@ -111,6 +111,13 @@ main(int argc, char **argv)
 	 * and start monitor
      */
 	myLocalMode = is_standby(myLocalConn) ? STANDBY_MODE : PRIMARY_MODE;
+	if (myLocalMode == PRIMARY_MODE)
+	{
+		primaryId = myLocalId;
+		strcpy(primaryConninfo, conninfo);
+		primaryConn = myLocalConn;
+	}
+
 	checkClusterConfiguration();
 	checkNodeConfiguration(conninfo);
 	if (myLocalMode == STANDBY_MODE)
@@ -132,9 +139,7 @@ main(int argc, char **argv)
  * This function ask if we are in recovery, if false we are the primary else 
  * we are a standby
  */
-
-
-void
+static void
 getPrimaryConnection(void)
 {
     PGresult *res1;
@@ -199,7 +204,7 @@ getPrimaryConnection(void)
 }
 
 
-void
+static void
 MonitorCheck(void) { 
 	/*
 	 * Every 3 seconds, insert monitor info
@@ -215,8 +220,7 @@ MonitorCheck(void) {
 /*
  * Check if its time for next monitor call and if so, do it.
  */
-
-void
+static void
 MonitorExecute(void)
 {
     PGresult *res;
@@ -291,7 +295,7 @@ MonitorExecute(void)
 }
 
 
-void
+static void
 checkClusterConfiguration(void)
 {
     PGresult   *res;
@@ -324,7 +328,7 @@ checkClusterConfiguration(void)
 }
 
 
-void
+static void
 checkNodeConfiguration(char *conninfo)
 {
     PGresult   *res;
@@ -372,7 +376,7 @@ checkNodeConfiguration(char *conninfo)
 }
 
 
-unsigned long long int 
+static unsigned long long int 
 walLocationToBytes(char *wal_location)
 {
     unsigned int xlogid;