Slight redesign so that secondary box only runs script manually when someone wants it to take over

failover/Makefile

@@ -1,10 +1,17 @@
+FAILOVER=/home/nagios/bin/failover.sh
+
 default:	test
 
 test:
-	./failover.sh -P primary.localdomain -S secondary.localdomain
+	${FAILOVER} -P primary.localdomain -S secondary.localdomain
 
 go:
-	./failover.sh -P primary.localdomain -S secondary.localdomain -E
+	${FAILOVER} -P primary.localdomain -S secondary.localdomain -E
+force:
+	${FAILOVER} -P primary.localdomain -S secondary.localdomain -E --force
 
+copy-test:
+	rsync -avun /home/nagios/princeton/failover/ /home/nagios/bin/
 copy:
 	rsync -avu /home/nagios/princeton/failover/ /home/nagios/bin/
+

failover/START

@@ -0,0 +1,3 @@
+#!/bin/sh
+
+/home/nagios/bin/failover.sh -P primary.localdomain -S secondary.localdomain -E --force

failover/failover.sh

@@ -44,14 +44,10 @@ ping_host() {
 }
 
 stop_nagios() {
-    verbose "Ensuring Nagios services are stopped because we are the secondary"
-    sleep 2
     /home/nagios/bin/nagios_startstop.sh stop
 }
 
 start_nagios() {
-    verbose "Ensuring Nagios services are stopped because we are the secondary"
-    sleep 2
     /home/nagios/bin/nagios_startstop.sh start
 }
 
@@ -82,20 +78,27 @@ do_primary() {
 do_secondary() {
   verbose "We are the secondary.  Checking to see if primary is on the network..."
   ping_host ${primary}
-  if [ $? -eq 0 -a -z "$force"  ]; then
-    warning "The primary (${primary}) is on the network.  This process will be aborted.  Use --force to force it to continue."
-    exit 1
+  if [ $? -eq 0 ]; then
+    if [ -z $force ]; then
+      warning "The primary (${primary}) is on the network.  This process will be aborted.  Use --force to force it to continue."
+      exit 1
+    else
+      warning "The primary (${primary}) is on the network but --force specified.  Proceeding anyway."
+    fi
   fi
   verbose "Expanding package from the primary"
   syncfile=$(basename `ls -tr1 ${backupDir}/${syncname}* | tail -1`)
   [ -z "$syncfile" ] && error "No sync file ($syncfile) found." && exit 2
   /home/nagios/bin/rsync_xi.sh --primary "${primary}" --secondary "${secondary}" --file "/store/backups/nagiosxi/$syncfile"
   if [ $? -eq 0 ]; then
-    # stop_nagios
+    sleep 2
     if [ -r "/usr/local/nagiosxi/html/includes/components/custom-includes/css/header-gradient.css.secondary" -a -w "/usr/local/nagiosxi/html/includes/components/custom-includes/css/header-gradient.css" ]; then
       cp /usr/local/nagiosxi/html/includes/components/custom-includes/css/header-gradient.css.secondary /usr/local/nagiosxi/html/includes/components/custom-includes/css/header-gradient.css
     fi
+    verbose "This is intended to be manually executed when needed, so we assume you want to start Nagios..." -n
+    start_nagios
     rm -f ${syncfile}
+    verbose "Done."
   fi
 }
 

failover/rsync_xi.sh

@@ -8,7 +8,6 @@ PATH=/home/nagios/bin:/usr/bin:/bin
 . /home/nagios/bin/colors.sh
 
 BASEDIR="/usr/local/nagiosxi/scripts"
-SBLOG="/usr/local/nagiosxi/var/components/scheduledbackups.log"
 
 timeStamp=`date +%s`
 hostname="`/bin/hostname`"
@@ -83,8 +82,8 @@ while [ -n "$1" ]; do
 done
 
 do_primary() {
-ping_host $sName
-[ $? -ne 0 ] && echo "Cannot ping $sName.  Use --noping to continue wihout ping." && exit
+#ping_host $sName
+#[ $? -ne 0 ] && echo "Cannot ping $sName.  Use --noping to continue wihout ping." && exit
 echo "Starting sync from $pName to $sName"
 
 # Restart nagios to forcibly update retention.dat
@@ -231,15 +230,15 @@ fi
 ##############################
 # COMPRESS BACKUP
 ##############################
-echo "Compressing backup..."
+verbose "Compressing database file..." -n
 tar czfp $name.tar.gz $name
 rm -rf $name
+verbose "Done."
 
 # Change ownership
 chown $nagiosuser:$nagiosgroup $name.tar.gz
 
 if [ -s $name.tar.gz ];then
-
     echo " "
     echo "==============="
     echo "BACKUP COMPLETE"
@@ -255,6 +254,7 @@ else
     exit 1;
 fi
 do_rsync $rootdir/$name.tar.gz
+rm -Rf $rootdir/$name.tar.gz
 } # End of do_backup_sql
 
 do_backup_files