#!/usr/bin/perl
#-------------------------------------------------------------------------------
#
#   Programm        : check_cpu_usage
#
#-------------------------------------------------------------------------------
#
#   Beschreibung    : Nagios check script to check the CPU Usage by monitoring
#                     /proc/stat.
#
#   Author          : Marek Zavesicky
#   Version         : $Revision: $
#   Erstellt        : 2013/08/26
#   Letztes Update  : $Date: $
#
#   $Id: $
#   Change history  :
#                     $Log: $
#
#   This program is free software; you can redistribute it and/or modify it
#   under the terms of the GNU General Public License as published by the
#   Free Software Foundation; either version 2 of the License, or (at your
#   option) any later version.  See <http://www.fsf.org/copyleft/gpl.txt>.
#
#   This program is distributed in the hope that it will be useful, but
#   WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
#   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
#   for more details.
#
#-------------------------------------------------------------------------------

#-------------------------------------------------------------------------------
#                     Pragma
#-------------------------------------------------------------------------------
use strict;
use warnings;

#-------------------------------------------------------------------------------
#                     Used Modules
#-------------------------------------------------------------------------------
use Monitoring::Plugin;
use Data::Dumper;

#-------------------------------------------------------------------------------
#   Function        : readStatFile
#-------------------------------------------------------------------------------
#   Description     : Parses the stat file and extract the counters, then store
#                     them in a hash
#
#   Input           : $np           Nagios::Plugins reference
#                     $file         A file with status informations
#                     $names        Array that has the order of the fields
#   Output          : $stat         A hash object with the newdata stats
#-------------------------------------------------------------------------------
sub readStatFile
{
    my $np = shift;
    my $file = shift;
    my $names = shift;
    my $stat = ();

    open( FILE, "<", $file ) or $np->nagios_die( "cant open file $file" );

    while ( <FILE> )
    {
        next unless ( $_ =~ /^cpu/ );

        my ( $name, @cpudata ) = split( '\s+', $_ );
        $stat->{ $name } = \@cpudata;
    }
    close( FILE );

    return $stat;
}

#-------------------------------------------------------------------------------
#   Function        : writeStatFile
#-------------------------------------------------------------------------------
#   Description     : Write the olddata object to the gapfile
#
#   Input           : $np           Nagios::Plugins reference
#                     $objects      the data objects
#                     $names        Array that has the order of the fields
#   Output          : -
#-------------------------------------------------------------------------------
sub writeStatFile
{
    my $np = shift;
    my $objects = shift;
    my $names = shift;

    open( FILE, ">", $np->opts->gapfile ) or $np->nagios_die( "cant open file $np->opts->gapfile" );
    foreach ( keys %{ $objects->{ newdata } } )
    {
        my $string = $_ . " ";
        $string .= join( " ", @{ $objects->{ newdata }->{ $_ } } );
        print FILE $string . "\n";
    }
    close( FILE );
}

#-------------------------------------------------------------------------------
#   Function        : calculateData
#-------------------------------------------------------------------------------
#   Description     : Calculate the diference between old and new data and store
#                     the result in an array. Summarize all differences and
#                     calculate the percentage of each value.
#
#   Input           : $np           Nagios::Plugins reference
#                     $objects      the data objects
#                     $names        Array that has the order of the fields
#   Output          : -
#-------------------------------------------------------------------------------
sub calculateData
{
    my $name = shift;
    my $objects = shift;
    my $names = shift;
    my @results;
    my $sum = 0;
    my $cpu = ();

    for ( my $i = 0; $i < scalar @$names; $i++ )
    {
        if ( defined( $objects->{ olddata }->{ $name }->[ $i ] ) )
        {
            push( @results, $objects->{ newdata }->{ $name }->[ $i ] - $objects->{ olddata }->{ $name }->[ $i ] );
            $sum += $results[ $i ];
        }
        else
        {
            push( @results, 0 );
        }
    }

    for ( my $i = 0; $i < scalar @$names; $i++ )
    {
        # in case the gapfile is empty
        if ( $sum == 0 )
        {
            $cpu->[ $i ] = 0;
            next;
        }
        $cpu->[ $i ] = $results[ $i ] * 100 / $sum;
    }
    return $cpu;
}

#-------------------------------------------------------------------------------
#   Function        : processData
#-------------------------------------------------------------------------------
#   Description     : Process the hashes of cpu's.
#
#   Input           : $np           Nagios::Plugins reference
#                     $objects      the data objects
#                     $names        Array that has the order of the fields
#   Output          : -
#-------------------------------------------------------------------------------
sub processData
{
    my $np = shift;
    my $objects = shift;
    my $names = shift;
    my $percent = ();

    $percent->{ cpu } = calculateData( "cpu", $objects, $names );

    if ( $np->opts->details )
    {
        foreach ( sort keys %{ $objects->{ olddata } } )
        {
            next if ( $_ eq 'cpu' );

            $percent->{ $_ } = calculateData( $_, $objects, $names );
        }
    }
    return $percent;
}

#-------------------------------------------------------------------------------
#   Function        : compareData
#-------------------------------------------------------------------------------
#   Description     : Compare data with threshold if needed and generate the
#                     performance data string.
#
#   Input           : $np           Nagios::Plugins reference
#                     $objects      the data objects
#                     $names        Array that has the order of the fields
#   Output          : -
#-------------------------------------------------------------------------------
sub compareData
{
    my $np          = shift;
    my $objects     = shift;
    my $names       = shift;
    my $warnings    = [ split( ',', $np->opts->warning ) ];
    my $criticals   = [ split( ',', $np->opts->critical ) ];
    my $string      = "";
    my $result      = 0;
    my $res;

    foreach ( sort keys %{ $objects->{ olddata } } )
    {
        # if details are not needed, grab the next line
        unless ( $np->opts->details )
        {
            next unless ( $_ eq 'cpu' );
        }

        # compute all columns and check the thresholds
        $string .= uc( $_ );
        for ( my $i = 0; $i < scalar @$names; $i++ )
        {
            if ( defined( $$warnings[ $i ] ) and defined( $$criticals[ $i ] ) and ( $$warnings[ $i ] ne "none" ) and ( $$criticals[ $i ] ne "none" ) )
            {
                $res = $np->check_threshold( check      => $objects->{ percent }->{ $_ }->[ $i ],
                                             warning    => $$warnings[ $i ],
                                             critical   => $$criticals[ $i ] );
                # Evaluate the new return code, do not overwrite higher severity by lower
                $result = ( $res > $result ? $res : $result );
                $np->add_perfdata(
                    label       => $_ . "_" . $$names[ $i ],
                    value       => $objects->{ percent }->{ $_ }->[ $i ],
                    warning     => $$warnings[ $i ],
                    critical    => $$criticals[ $i ],
                    uom         => "%"
                );
            }
            else
            {
                $np->add_perfdata(
                    label       => $_ . "_" . $$names[ $i ],
                    value       => $objects->{ percent }->{ $_ }->[ $i ],
                    uom         => "%"
                );
            }
            # create the message string
            if ( $$names[ $i ] =~ /^user/ or $$names[ $i ] =~ /^system/  or $$names[ $i ] =~ /^idle/ or $$names[ $i ] =~ /^iowait/ or $$names[ $i ] =~ /^steal/ )
            {
                $string .= sprintf( " (%s=%3.2f)", $$names[ $i ], $objects->{ percent }->{ $_ }->[ $i ] );
            }
        }
        $string .= ", ";
    }

    # strip the last two characters from string
    return ( $result, substr( $string, 0, -2 ) );
}

#-------------------------------------------------------------------------------
#   Main
#-------------------------------------------------------------------------------
my $objects = ();
my $names = [];
my $np = Monitoring::Plugin->new(
    shortname =>    "Linux CPU Usage",
    usage =>        "Usage: %s < arguments > " .
                    "arguments: \n" .
                    "   [ -t|--timeout=<timeout> ]      timeout\n" .
                    "   [ -c|--critical=<threshold> ]   critical threshold\n" .
                    "   [ -w|--warning=<threshold> ]    warning threshold\n" .
                    "   [ -s|--statfile=<file> ]        name of the stat file (default /proc/stat)\n" .
                    "   [ -g|--gapfile=<file> ]         name of the gap file (default /tmp/check_cpu_usage.gap.tmp)\n" .
                    "   [ -n|--names=<list> ]           comma separated list of names representing the column in the stats file\n" .
                    "   [ -d|--details ]                show detailed information for each core\n",
    plugin  =>      'check_cpu_usage'
);

$np->add_arg(
    spec =>         'warning|w=s',
    help =>         "--warning -w\n   a list of threshold for warning in the same order as names\n" .
                    "   (default none,none,none,none,none,none,none,none,none,none,none,none,none,none)",
    default =>      "none,none,none,none,none,none,none,none,none,none,none,none,none,none",
    required =>     0
);

$np->add_arg(
    spec =>         'critical|c=s',
    help =>         "--critical -c\n   a list of threshold for critical in the same order as names\n" .
                    "   (default none,none,none,none,none,none,none,none,none,none,none,none,none,none)",
    default =>      "none,none,none,none,none,none,none,none,none,none,none,none,none,none",
    required =>     0
);

$np->add_arg(
    spec =>         'statfile|s=s',
    help =>         "--statfile -s\n   name of the stat file (default /proc/stat)",
    default =>      "/proc/stat",
    required =>     0
);

$np->add_arg(
    spec =>         'gapfile|g=s',
    help =>         "--gapfile -g\n   name of the gap file (default /tmp/check_cpu_usage.gap.tmp)",
    default =>      "/tmp/check_cpu_usage.gap.tmp",
    required =>     0
);

$np->add_arg(
    spec =>         'details|d',
    help =>         "--details -d\n   show detailed information for each core",
    required =>     0
);


$np->add_arg(
    spec =>         'names|n=s',
    help =>         "--names -n\n   a comma separated list of names representing the column in the stats file. See 'man proc' for details\n" .
                    "   (default user,nice,system,idle,iowait,irq,softirq,steal,guest,guest_nice,nyd1,nyd2,nyd3)",
    default =>      "user,nice,system,idle,iowait,irq,softirq,steal,guest,guest_nice,nyd1,nyd2,nyd3",
    required =>     0
);

# read the cli options
$np->getopts;

# rearange scalar lists into arrays
$names = [ split( ',', $np->opts->names ) ];

# read data from stats and store it to a hash
$objects->{ newdata } = readStatFile( $np, $np->opts->statfile, $names );

if ( -e $np->opts->gapfile )
{
    # read data from stored file and store it to a hash
    $objects->{ olddata } = readStatFile( $np, $np->opts->gapfile, $names );

    # calculate difference and percent for the data
    $objects->{ percent } = processData( $np, $objects, $names );
}

# write the new data to the gapfile
writeStatFile( $np, $objects, $names );

print Dumper( $names, $objects ) if ( $np->opts->verbose );

# compare data with warning and critical threshold and exit
$np->nagios_exit( compareData( $np, $objects, $names ) );