#!/usr/bin/perl
#
# Audit the list of supposed cfengine hosts to see which are actually
# responding.  Do this by using cfrun to try to run a sample job on them
# and interpreting the output.
#
# David J. Bianco <bianco@jlab.org>
#

use Getopt::Long;
use Mail::Mailer;

$CFRUN="/var/cfengine/sbin/cfrun";
$CFHOSTS="/local/cfengine/etc/cfrun.hosts";
$DEBUG = 0;
# Customize this email address.  This is where the report will be sent.
$MAILADDR="cfengine\@yoursite.com";
$TIMEOUT=300;

$errorstrings="Couldn't|failed";
$downhosts="";
$messages="";

GetOptions("cfrun=s" => \$CFRUN,
	   "cfhosts=s" => \$CFHOSTS,
	   "nomail" => \$NOMAIL,
	   "smtpserver" => \$SMTP_SERVER,
	   "timeout=i" => \$TIMEOUT,
	   "mailaddr=s" => \$MAILADDR,
	   "debug" => \$DEBUG);

open(CFHOSTS,$CFHOSTS) || die "Can't open $CFHOSTS: $!\n";

$SIG{ALRM} = sub {die "Operation timed out"; };

while(<CFHOSTS>) {
    $host = $_;
    chomp $host;

    print "********* Trying host $host\n" if $DEBUG;

    eval {
	alarm($TIMEOUT);
	$output = `$CFRUN -f $CFHOSTS $host 2>&1`;
	alarm(0);
    };

    alarm(0);

    print "******** OUTPUT: $host **********\n$output\n\n\n" if $DEBUG;

    if(($output =~ m/$errorstrings/) || ($output eq "") || ($@ eq "Operation timed out")) {
	$downhosts .= "$host\n";
	$messages .= "************** $host ***************\n$output\n\n";
	if($@ eq "Operation timed out") {
	    $messages .= "[OPERATION TIMED OUT]";
	}
    }
}

$outputstring =  "The following hosts are down:\n$downhosts\n\n";
$outputstring .= "Full details for these hosts follow:\n$messages\n\n";

print $outputstring;

if(! $NOMAIL) {
    $mailer = Mail::Mailer->new("sendmail");
# Customize this next line, to include a valid return email address. 
    $mailer->open({ From    =>   "CFEngine Audit Service <cfengine\@yoursite.org>",
		    To      =>   $MAILADDR,
		    Subject =>   "CFEngine Audit Report",
		}) or die "Can't send mail: $!\n";
    
    print $mailer $outputstring;
    $mailer->close();
}