check_serveriron_cpu

#!/usr/bin/perl -w
# vim:ts=4
#
# Check a Foundry / Brocade ServerIron WSM for CPU usage
# Jay Reitz 2012
#
# Usage -
# check_serveriron_cpu [-v] -H switchname [-C community] [-s slotNumber]
#
# Version 0.1

use strict;
use Net::SNMP;
use Getopt::Std;
use vars qw/$opt_c $opt_w $opt_h $opt_d $opt_H $opt_C $opt_s/;

my $CRITACTIVE = 80;
my $WARNACTIVE = 60;
my $MIBFOUNDRY = '1.3.6.1.4.1.1991';
my $MIBCPUVALUE = '1.1.2.11.1.1.4';
my $TIMEOUT = 4;
my $DEBUG = 0;
my $SWITCH = '';
my $COMMUNITY = 'public';
my $SLOT = 1;

my %processorTypes = (management => 1, barrel => 2);
my @processorTimespans = (5, 300);

my $cpu = {};
my $hot = {};

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

####
sub dohelp()
{
    print "Usage: $0 [-d] -H switchname [-C community] [-s slotNumber] [-w warnPercent] [-c critPercent]\n";
    print "Check the processor utilization of a WSM module in a Brocade/Foundry ServerIron";
    print "  -d\tdebug\n";
    print "  -H\thostname\n";
    print "  -C\tcommunity\n";
    print "  -s\tslot number (default: 1)\n";
    print "  -w\twarning threshold (default: $WARNACTIVE)\n";
    print "  -c\tcritical threshold (default: $CRITACTIVE)\n";
    print "\nExample:\n";
    print "$0 -H balancer1 -C mang -s 1\n";

    exit 0;
}
####

sub assureSanity()
{
    if (!$SWITCH) {
        print "ERROR: Must specify Foundry switch name.\n";
        exit $ERRORS{"UNKNOWN"};
    }
    if (!$COMMUNITY) {
        print "ERROR: Must specify SNMP community string.\n";
        exit $ERRORS{"UNKNOWN"};
    }
    if ($WARNACTIVE > $CRITACTIVE) {
        print "ERROR: Warning threshold must be lower than critical threshold.\n";
        exit $ERRORS{"UNKNOWN"};
    }
    if ($DEBUG) {
        use Data::Dumper;
        print "Switch: $SWITCH\n";
        print "Community: $COMMUNITY\n";
        print "Slot: $SLOT\n";
    }
}

sub getCpuMibs()
{
    my $mibs = [];
    while (my (undef, $type) = each %processorTypes) {
        foreach my $span (@processorTimespans) {
            push @$mibs, "$MIBFOUNDRY.$MIBCPUVALUE.$SLOT.$type.$span";
        }
    }

    return $mibs;
}

sub readFoundryValues()
{
    print "Starting SNMP\n" if ($DEBUG);
    my ($snmp,$snmperr) = Net::SNMP->session(-hostname=>$SWITCH,
        -community=>$COMMUNITY, -timeout=>$TIMEOUT, -retries=>2);
    if ($snmperr) {
        print "ERROR: $snmperr";
        exit $ERRORS{"UNKNOWN"};
    }

    #$snmp->snmp_debug() if $DEBUG;

    # gather CPU stats
    my $resp = $snmp->get_request(-varbindlist=> getCpuMibs());
    if (!$resp) {
        print "ERROR: Cannot read cpu OIDs";
        exit $ERRORS{"UNKNOWN"};
    }

    print Dumper($resp) if $DEBUG;

    my ($k, $slot, $proc, $span, $util);
    foreach my $oid (keys %{$resp}) {
        $oid =~ /$MIBCPUVALUE\.(\d+)\.(\d+)\.(\d+)/;
        #print "$oid => ". $resp->{$oid} ."\n" if ($DEBUG);

        $slot = $1;
        $proc = $2;
        $span = $3;
        $util = int($resp->{$oid}) / 100;
        $k = "$slot:$proc";

        unless (exists $cpu->{$k}) {
            $cpu->{$k} = {
                slot     => $slot,
                proc     => $proc
            };
        }

        # set the CPU utilization for this timespan
        $cpu->{$k}->{"util_$span"} = $util;

        # if this is the highest utilization value, store it
        if (!exists $hot->{key} || $util > $hot->{util}) {
            $hot = { key  => $k,
                     slot => $slot,
                     proc => $proc,
                     util => $util
                   };
        }
    }

    $snmp->close();
}

sub listCpus()
{
    print "Listing all CPUs...\n" if ($DEBUG);
    print "Slot Type  5 second      5 minute  hottest?\n";
    foreach (keys %$cpu) {
        printf "%-4s %-3s %5d            %3d  %7s\n",
            $cpu->{$_}->{slot}, $cpu->{$_}->{proc}, $cpu->{$_}->{util_5}, $cpu->{$_}->{util_300}, ($hot->{key} eq $_) ? 'yes' : '';
    }
}

sub getPerf()
{
    my @p = ();
    foreach my $k (sort keys %$cpu) {
        push @p, getShortName($cpu->{$k}) ."=". $cpu->{$k}->{util_300} ."%;$CRITACTIVE;$WARNACTIVE";
    }
    return '|' . join(' ', @p) ."\n";
}

sub getMsg()
{
    return getShortName($hot) .' processor is '. $hot->{util} . "% utilized"
}

sub getShortName() {
    my $c = shift @_;
    if ($c->{proc} == 1) {
        return 'management';
    } else {
        return "barrel". ($c->{proc} - 1);
    }
}

######
# main

getopts('dhH:C:s:w:c:');
dohelp if ($opt_h);
$DEBUG = 1 if ($opt_d);
$COMMUNITY = $opt_C if ($opt_C);
$SWITCH = $opt_H if ($opt_H);
$SLOT = $opt_s if ($opt_s);
$CRITACTIVE = $opt_c if ($opt_c);
$WARNACTIVE = $opt_w if ($opt_w);

assureSanity();
readFoundryValues();
listCpus if ($DEBUG);

# do output
if ($hot->{util} >= $CRITACTIVE) {
    print "CRIT: ". getMsg() . getPerf();
    exit $ERRORS{"CRITICAL"};
} elsif ($hot->{util} >= $WARNACTIVE) {
    print "WARN: ". getMsg() . getPerf();
    exit $ERRORS{"WARN"};
} else {
    print "OK: ". getMsg() . getPerf();
    exit $ERRORS{"OK"};
}