#!/usr/bin/perl
#
# Nagios check plugin for lustre filesystems.
#
# Josh Malone (jmalone@nrao.edu) - July 2013
#
# Revision 3 - July 17, 2013

use lib "/usr/lib64/nagios/plugins";
use lib "/usr/lib/nagios/plugins";
use lib ".";
use utils qw($TIMEOUT %ERRORS &print_revision &support);
use Switch;
use Getopt::Std;

$exstate=$ERRORS{'OK'};
$message="OK: All filesystems okay";
%pctused=();
%bused=();
%btotal=();

$warnpct=70;
$critpct=85;

##########

getopts('w:c:htsv');
if ($opt_w) { $warnpct=$opt_w; }
if ($opt_c) { $critpct=$opt_c; }

if ($opt_h) {
    # Help message
    print << "EOHELP";
Usage:  check_lustre.pl [-w warn%] [-c crit%]

	-w: Warning space threshold (percent used)
	-c: Critical space threshold (percent used)
	-t: Ignore OSTs that are temporarily unavailable
EOHELP
    exit 0;
}

# Set alarm to time out possibly hung `lfs df` command
alarm 40;
$SIG{ALRM} = sub {
    print "CRITICAL: lustre lfs operation timed out\n";
    exit $ERRORS{'CRITICAL'};
};

# Make sure lustre FS is loaded and available
if ( ! -d "/proc/fs/lustre" ) {
    print("CRITICAL: lustre filesystem not available!\n");
    exit $ERRORS{'CRITICAL'};
}

open (PROC, "/proc/fs/lustre/health_check") or die("Cannot read lustre health_check - is lustre installed?");
while(<PROC>) {
    chomp();
    if ($_ ne "healthy") {
    	$message="CRITICAL: Lustre not healthy";
	$exstate=$ERRORS{'CRITICAL'};
    }
}
close (PROC);

if ($opt_s) {
    open (LFS, "-") 
} else {
    open (LFS, "lfs df |") or die("Cannot read from lfs!");
}
while (<LFS>) {
    chomp();
    print "$_\n" if ($opt_v);
    @parts=split(' ');
    switch ($_) {
	case /Resource temporarily unavailable/ {
		next unless ($opt_t);
	}
	case /unavailable/ {
		$exstate=$ERRORS{'CRITICAL'};
		$message="CRITICAL: ".join(' ',@parts);
	}
	case / summary/ {
		$fs=$parts[6];
		$fs=~s/.*\///;
		$btotal{$fs}=$parts[2]*1024;
		$bused{$fs}=$parts[3]*1024;
		$tmp=$parts[5];
		$tmp=~s/\%//;
		$pctused{$fs}=$parts[5];
		if ($parts[5] > $critpct) {
		    # Don't overwrite a more serious OST error
		    if( $exstate==$ERRORS{'OK'} ) {
		    	$message=sprintf("CRITICAL: Filesystem %s is %d%% full", $fs, $parts[5]);
			$exstate=$ERRORS{'CRITICAL'};
		    }
		}
		if ($parts[5] > $warnpct) {
		    if( $exstate==$ERRORS{'OK'} ) {
		    	$message=sprintf("WARNING: Filesystem %s is %d%% full", $fs, $parts[5]);
			$exstate=$ERRORS{'WARNING'};
		    }
		}
	}
    }
}
$perf='';
foreach $fs (keys(%pctused)) {
    $perf.=sprintf(" %s_percent_used=%s;%d;%d", $fs, $pctused{$fs}, $warnpct, $critpct );
    $perf.=sprintf(" %s_used=%dBytes;%d;%d", $fs, $bused{$fs}, $btotal{$fs}*$warnpct, $btotal{$fs}*$critpct );
}
print $message." |$perf\n";
exit $exstate;

close(LFS);