Contents of /trunk/nl.nikhef.ndpf.tools/nagios-ndpf-sensors/scripts/check_qstat

#!/usr/bin/perl -w

use strict;
use Getopt::Long;

my $verbose=0;
my $help;
my $hostname;

my $TIMEOUT = 60;

GetOptions("help" => \$help,
           "hostname|H:s" => \$hostname, 
           "verbose|v+" => \$verbose );

my %ERRORS=(OK=>0,
            WARNING=>1,
            CRITICAL=>2,
            UNKNOWN=>3,
            DEPENDENT=>4);

my $QSTAT = "/usr/bin/qstat";
unless ( -x $QSTAT ) {
    print "$QSTAT: $!";
    return $ERRORS{UNKNOWN};
}

my $result = $ERRORS{"UNKNOWN"};
my $message = "";

# Just in case of problems, let's not hang Nagios
$SIG{'ALRM'} = sub {
     print ("ERROR: No response from $hostname querying qstat (alarm timeout)\n");
     exit $ERRORS{"CRITICAL"};
};

alarm($TIMEOUT);

##############################################################################
# output of qstat
#
my $errors    = 0;
my $warnings  = 0;

my $cmd = "$QSTAT";
open QSTAT, "$cmd |";
my $ret = $?;

my $perfdata = "";

if ( $ret == 0 ) {
    my ( $errstr, $warnstr, $okstr );

    my %count = ( Q => 0, R => 0 );
    while ( <QSTAT> ) {
        next unless ( /^\d+/ );             # only count lines with job ID
        my $state = (split /\s+/)[0,4];

        $count{$state}++;
    }

    # OK: message contains no details
    $message = "OK; $count{R} running, $count{Q} queued";
    $result = $ERRORS{"OK"};
    $perfdata = "'queued'=$count{Q};;;;; 'running'=$count{R};;;;";
}
else {
    # Failed to get output of pbsnodes: UNKNOWN
    $message = "UNKNOWN: could not determine result of $QSTAT";
    $result = $ERRORS{"UNKNOWN"};
}

alarm(0);


close QSTAT;

# Write output and return exit code;
print "$message|$perfdata\n";
exit($result);

1	#!/usr/bin/perl -w
2
3	use strict;
4	use Getopt::Long;
5
6	my $verbose=0;
7	my $help;
8	my $hostname;
9
10	my $TIMEOUT = 60;
11
12	GetOptions("help" => \$help,
13	"hostname\|H:s" => \$hostname,
14	"verbose\|v+" => \$verbose );
15
16	my %ERRORS=(OK=>0,
17	WARNING=>1,
18	CRITICAL=>2,
19	UNKNOWN=>3,
20	DEPENDENT=>4);
21
22	my $QSTAT = "/usr/bin/qstat";
23	unless ( -x $QSTAT ) {
24	print "$QSTAT: $!";
25	return $ERRORS{UNKNOWN};
26	}
27
28	my $result = $ERRORS{"UNKNOWN"};
29	my $message = "";
30
31	# Just in case of problems, let's not hang Nagios
32	$SIG{'ALRM'} = sub {
33	print ("ERROR: No response from $hostname querying qstat (alarm timeout)\n");
34	exit $ERRORS{"CRITICAL"};
35	};
36
37	alarm($TIMEOUT);
38
39	##############################################################################
40	# output of qstat
41	#
42	my $errors = 0;
43	my $warnings = 0;
44
45	my $cmd = "$QSTAT";
46	open QSTAT, "$cmd \|";
47	my $ret = $?;
48
49	my $perfdata = "";
50
51	if ( $ret == 0 ) {
52	my ( $errstr, $warnstr, $okstr );
53
54	my %count = ( Q => 0, R => 0 );
55	while ( <QSTAT> ) {
56	next unless ( /^\d+/ ); # only count lines with job ID
57	my $state = (split /\s+/)[0,4];
58
59	$count{$state}++;
60	}
61
62	# OK: message contains no details
63	$message = "OK; $count{R} running, $count{Q} queued";
64	$result = $ERRORS{"OK"};
65	$perfdata = "'queued'=$count{Q};;;;; 'running'=$count{R};;;;";
66	}
67	else {
68	# Failed to get output of pbsnodes: UNKNOWN
69	$message = "UNKNOWN: could not determine result of $QSTAT";
70	$result = $ERRORS{"UNKNOWN"};
71	}
72
73	alarm(0);
74
75
76	close QSTAT;
77
78	# Write output and return exit code;
79	print "$message\|$perfdata\n";
80	exit($result);
81
grid.support@nikhef.nl	ViewVC Help
Powered by ViewVC 1.1.28