#!/usr/bin/perl -w

# cluster_percent               Timothy Denike 8/14/2004 Friendster.com
#
# This script acts as a wrapper to the check_cluster binary included in the
# Nagios distribution.  The problems with check_cluster are that it requires
# you to pipe to stdin a list of hostnames and services, as well as specify
# fixed values for the warning/critical thresholds.
#
# This script addresses those issues by taking a nagios hostgroup name and
# a nagios service name as parameters.  The hostgroup members are looked up 
# from the nagios config file and fed into check_cluster.  Optionally,
# you can specify a percentage for the trigger thresholds, rather than a
# fixed value.

# Edit the three variables below as needed for your system

use strict;
use IPC::Open2;

        my $config      = "/etc/nagios/hostgroups.cfg";
        my $statusfile  = "/var/log/nagios/status.log";
        my $cluster     = "/usr/lib/nagios/plugins/check_cluster";
        my $DEBUG       = 0;


sub alog {
    my $log  = "/var/log/nagios/cluster_percent.log";
    my $text = shift;
    my $time = localtime();
    open LOG, ">> $log" || die "Unable to open $log\n";
    print LOG "[$$ $time] $text\n";
    close LOG;
}

sub barf ($$) {
    # barf is used to output a consistent format and exit status.
    my %ERRORS = ('-1',"UNKNOWN",
              '0' , 'OK',
              '1', 'WARNING',
              '2', 'CRITICAL');
    my $errorcode = shift;
    my $description = shift;

    #print "$ERRORS{$errorcode} : $description\n";
    print "$description\n";
    if ($DEBUG) { alog ("barf - $errorcode: $description"); }
    exit $errorcode;
}

sub roundup {
    my $n = shift;
    return(($n == int($n)) ? $n : int($n + 1))
}

sub syntax {
        print "$0 hostgroup service [warn] [crit]\n";
        print "\thostgroup = Hostgroup name as it appears in
hostgroups.cfg\n";
        print "\tservice   = Service name as is appears in Nagios\n";
        print "\twarn      = % down to trigger Warning - def 25\n";
        print "\tcrit      = % down to trigger Critical - def 33\n";
        if ($DEBUG) { alog ("Called with no parms - syntax output"); }
        exit -1; # Just in case we're called improperly, exit Unknown state
}

sub get_members {
        # Get the members of the hostgroup from the hostgroups.cfg file
        my $hostgroup = shift;
        my $config      = shift;
        my $ingroup = 0;
        my $members = "";
        my @hosts;

        open CONFIG, $config || barf (-1,"Unable to open $config");
        while (<CONFIG>)
        {
                if ( /^\s*hostgroup_name\s*$hostgroup\s*/ )
                {
                        $ingroup=1;
                        # Set this variable if we're inside the hostgroup
                        #  stanza we're looking for.
                }
                if ( $ingroup && /^\s*members\s*(\S*)/ )
                {
                        $members=$1;
                        @hosts = split(/,/,$members);
                        if ($DEBUG) { alog ("Returning member list"); }
                        return @hosts;
                        # Okay - now we have a hostlist, lets return it.
                }
        }
        return 0;
}

MAIN:
{
        my @members;
        my $hostgroup   = shift || syntax();
        my $service     = shift || syntax();
        my $warn        = shift || "25";  # default to 25% for warnings
        my $crit        = shift || "33";  # 33% for critical

        my $hostcount   = 0;
        my $output      = "";
        my $status      = -1;
        my $RCLU;
        my $WCLU;

        if ($DEBUG) { alog ("Starting up.."); }
        # Get the members of the specified hostgroup
        @members = get_members($hostgroup,$config);

        $hostcount      = $#members +1; # Count is short by one

        if ($hostcount == 0) { barf (-1,"No hosts found in group
$hostgroup");}

        # Set the number of failed hosts to trigger a warning/critical
        my $critcount = roundup($hostcount*$crit/100);
        my $warncount = roundup($hostcount*$warn/100);

        # Open the cluster_check binary for input/output
        # We'll pipe in a list of hosts, and parse the output.
        if ($DEBUG) { alog("Dumping $hostcount to check_cluster"); }
        my $pid = open2 ($RCLU,$WCLU,"$cluster --service $statusfile
$warncount $critcount");

        # check_cluster wants hostname;servicename\N as input
        foreach (@members) {
                print $WCLU "$_;$service\n";
        }
        close $WCLU;
        if ($DEBUG) { alog("Finished writing to stdin");}
        # now fetch the output.  It should only be one line.
        while (<$RCLU>)
        { 
          $output=$_;
        }
        if ($DEBUG) { alog ("Got output - waiting for pid to exit");}
        # stick around for the child process to exit.
        waitpid ($pid,0);
        if ($DEBUG) { alog ("Child pid $pid done, going on..");}
        # I'm still not sure why, but the childs exit status is a factor of
256.
        $status = $? / 256;

        # Get rid of any CR on the end out the output
        chomp $output;

        # I don't want the "service cluster ok: " part, just the
        #  good stuff at the end of the line.
        ($output) = $output =~ /.*: (.*)$/;
        # I'd rather have a total count first for reference.
        $output = "$hostcount total - $output";

        # Lets just make sure any weird codes exit with -1 (Unknown)
        if ($status < 0 || $status > 2) { $status = -1; }

        # Call barf to exit with a description and proper exit status.
        barf ($status,$output);

}