rewrite nagios check scripts in perl

3 files changed, 103 insertions, 75 deletions

diff --git a/files/checkbackups.pl b/files/checkbackups.pl
new file mode 100755
index 0000000..24632d1
--- /dev/null
+++ b/files/checkbackups.pl
@@ -0,0 +1,101 @@
+#!/usr/bin/perl -w
+
+# This script is designed to check a backup directory populated with
+# subdirectories named after hosts, within which there are backups of various
+# types.
+#
+# Example:
+# /home/backup:
+# foo.example.com
+# 
+# foo.example.com:
+# rdiff-backup .ssh
+#
+# rdiff-backup:
+# root home rdiff-backup-data usr var
+#
+# There are heuristics to determine the backup type. Currently, the following
+# types are supported:
+#
+# rdiff-backup: assumes there is a rdiff-backup/rdiff-backup-data/backup.log file
+# duplicity: assumes there is a dup subdirectory, checks the latest file
+# dump files: assumes there is a dump subdirectory, checks the latest file
+#
+# This script returns output suitable for send_nsca to send the results to
+# nagios and should therefore be used like this:
+#
+# checkbackups.sh | send_nsca -H nagios.example.com
+
+use Getopt::Std;
+
+# XXX: taken from utils.sh from nagios-plugins-basic
+my $STATE_OK=0;
+my $STATE_WARNING=1;
+my $STATE_CRITICAL=2;
+my $STATE_UNKNOWN=3;
+my $STATE_DEPENDENT=4;
+
+our $opt_d = "/backup";
+our $opt_c = 48 * 60 * 60;
+our $opt_w = 24 * 60 * 60;
+
+if (!getopts('d:c:w:')) {
+	print <<EOF
+Usage: $0 [ -d <backupdir> ] [ -c <threshold> ] [ -w <threshold> ]
+EOF
+	;
+	exit();
+}
+
+my $backupdir= $opt_d;
+my $crit = $opt_c;
+my $warn = $opt_w;
+
+# XXX: this should be a complete backup registry instead
+my @hosts=qx{ls $backupdir};
+
+chdir($backupdir);
+foreach my $host (@hosts) {
+	chomp($host);
+	my $flag="";
+	my $type="unknown";
+	if (-d $host) {
+		# guess the backup type and find a proper stamp file to compare
+		# XXX: this doesn't check if the backup was actually successful
+		# XXX: the backup type should be part of the machine registry
+		if (-d "$host/rdiff-backup") {
+			$flag="$host/rdiff-backup/rdiff-backup-data/backup.log";
+			$type="rdiff";
+		} elsif (-d "$host/dump") {
+			$flag="$host/dump/" . `ls -tr $host/dump | tail -1`;
+			chomp($flag);
+			$type="dump";
+		} elsif (-d "$host/dup") {
+			$flag="$host/dup";
+			$type="duplicity";
+		} else {
+			printf "$host\tbackups\t$STATE_UNKNOWN\tunknown system\n";
+			next;
+		}
+		my @stats = stat($flag);
+		if (not @stats) {
+			printf "$host\tbackups\t$STATE_UNKNOWN\tcannot stat flag $flag\n";
+			next;
+		}
+		my $t = time();
+		my $delta = $t - $stats[9];
+		my $state = $STATE_UNKNOWN;
+		if ($delta > $crit) {
+			$state = $STATE_CRITICAL;
+		} elsif ($delta > $warn) {
+			$state = $STATE_WARNING;
+		} elsif ($delta >= 0) {
+			$state = $STATE_OK;
+		}
+		print "$host\t";
+		print "backups\t$state";
+		print "\t$delta seconds old\n";
+	} else {
+		printf "$host\tbackups\t$STATE_UNKNOWN\tno directory\n";
+	}
+}
diff --git a/manifests/server.pp b/manifests/server.pp
index 4284ffa..9b08b22 100644
--- a/manifests/server.pp
+++ b/manifests/server.pp
@@ -25,12 +25,12 @@ class backupninja::server {
   
   file { "/usr/local/bin/checkbackups":
     ensure => "present",
-    content => template("backupninja/checkbackups.sh"),
+    source => "puppet://$servername/backupninja/checkbackups.pl",
     mode => 0755, owner => root, group => root,
   }
 
   cron { checkbackups:
-    command => "/usr/local/bin/checkbackups.sh | /usr/sbin/send_nsca -H nagios.koumbit.net -c /etc/send_nsca.cfg",
+    command => "/usr/local/bin/checkbackups -d $real_backupdir | /usr/sbin/send_nsca -H nagios.koumbit.net -c /etc/send_nsca.cfg | grep -v 'sent to host successfully'",
     user => "root",
     hour => "8-23",
     minute => 59,
diff --git a/templates/checkbackups.sh b/templates/checkbackups.sh
deleted file mode 100755
index 8d143a3..0000000
--- a/templates/checkbackups.sh
+++ /dev/null
@@ -1,73 +0,0 @@
-#!/bin/sh
-
-# This script is designed to check a backup directory populated with
-# subdirectories named after hosts, within which there are backups of various
-# types.
-#
-# Example:
-# /home/backup:
-# foo.example.com
-# 
-# foo.example.com:
-# rdiff-backup .ssh
-#
-# rdiff-backup:
-# root home rdiff-backup-data usr var
-#
-# There are heuristics to determine the backup type. Currently, the following
-# types are supported:
-#
-# rdiff-backup: assumes there is a rdiff-backup/rdiff-backup-data/backup.log file
-# duplicity: assumes there is a dup subdirectory, checks the latest file
-# dump files: assumes there is a dump subdirectory, checks the latest file
-#
-# This script returns output suitable for send_nsca to send the results to
-# nagios and should therefore be used like this:
-#
-# checkbackups.sh | send_nsca -H nagios.example.com
-
-# XXX: taken from utils.sh from nagios-plugins-basic
-STATE_OK=0
-STATE_WARNING=1
-STATE_CRITICAL=2
-STATE_UNKNOWN=3
-STATE_DEPENDENT=4
-
-backupdir="<%= real_backupdir -%>"
-
-# XXX: this should be a complete backup registry instead
-hosts=`ls $backupdir`
-stampfile=$backupdir/.stamp
-
-cd $backupdir
-for host in $hosts; do
-	flag=""
-	type="unknown"
-	if [ -d $host ]; then
-		# guess the backup type and find a proper stamp file to compare
-		# XXX: this doesn't check if the backup was actually successful
-		# XXX: the backup type should be part of the machine registry
-		if [ -d $host/rdiff-backup ]; then
-			flag=$host/rdiff-backup/rdiff-backup-data/backup.log
-			type="rdiff"
-		elif [ -d $host/dump ]; then
-			flag="$host/dump/`ls -tr $host/dump | tail -1`"
-			type="dump"
-		elif [ -d $host/dup ]; then
-			flag="$host/dup/`ls -tr $host/dup | tail -1`"
-			type="duplicity"
-		else
-			printf "$host\tbackups\t$STATE_UNKNOWN\tUNKNOWN unknown system\n"
-			continue
-		fi
-		touch -t `date +%Y%m%d0000.00` $stampfile
-		date=`ls -l $flag | awk '{print $6 " " $7}'`
-		if [ $flag -nt $stampfile ]; then
-			printf "$host\tbackups\t$STATE_OK\tOK timestamp $date\n"
-		else
-			printf "$host\tbackups\t$STATE_CRITICAL\tCRITICAL timestamp $date\n"
-		fi
-	else
-		printf "$host\tbackups\t$STATE_UNKNOWN\tUNKNOWN timestamp no directory\n"
-	fi
-done