scripts/xymon/plugins/client/ext/smart

#!/usr/bin/perl
# $Id: sensors 70 2011-11-25 09:21:18Z skazi $
# Author: Jacek Tomasiak <jacek.tomasiak@gmail.com>
# https://github.com/skazi0/xymon-plugins/blob/master/client/ext/smart

use strict;
# add script's directory to module search path for Hobbit.pm on non-debian systems
use FindBin;
use lib $FindBin::Bin;

use Hobbit;
use Data::Dumper;

my $bb = new Hobbit('smart');

my $temp_disk_list = "$ENV{'XYMONTMP'}/$ENV{'MACHINEDOTS'}.smart.drivedb.list";
my @disks = ();
my %olderr = {};

my $CACHETIME = 10; # minutes
my $CACHEFILE = "$ENV{'XYMONTMP'}/$ENV{'MACHINEDOTS'}.smart.cache";

&load_config("$ENV{'XYMONTMP'}/logfetch.$ENV{'MACHINEDOTS'}.cfg");

my @disks_stat = stat($temp_disk_list);
my $disks_mtime = scalar @disks_stat ? $disks_stat[9] : 0;
#
# Regenerate disks list if the file is too old (600 minutes)
if (time() - $disks_mtime > 600)
{
    unlink $temp_disk_list;
}

if (-e $temp_disk_list) {
    # Should use the existing file
}
else {
    # Create a file with the list of disks
    system("ls -1 /dev/sd* | grep -vE '[0-9]' > $temp_disk_list") == 0
        or die "system command to create $temp_disk_list failed: $?";
}

# fallback to disk detection if nothing defined in the config
unless (@disks) {
    ## Put temp_disk_list content to disks array
    open(my $fh, '<:encoding(UTF-8)', $temp_disk_list)
        or die "Could not open file '$temp_disk_list' $!";
    while (my $row = <$fh>) {
        chomp $row;
        push(@disks, "$row");
    }
}

my @stat = stat($CACHEFILE);
my $mtime = scalar @stat ? $stat[9] : 0;
# regenerate sensors cache if outdated
if (time() - $mtime > $CACHETIME * 60)
{
    open(OUT, ">$CACHEFILE") or die "cannot open $CACHEFILE";

    foreach my $name (@disks)
    {
        print OUT ('=' x 20) . " $name " . ('=' x 20) . "\n";
        my @output = `sudo smartctl -AHi -l error -l selftest $name 2>&1` or die;
        my $ncv = '';
        my $newerr = 1;
        my $ponhours = undef;
        my $lasttest = undef;
        foreach my $line (@output)
        {
            # skip header
            next if ($line =~ /smartctl|Copyright|Home page|===/);

            if ($line =~ /.*overall-health.*:\s*(.*)/)
            {
                my $lstatus = ($1 eq 'PASSED') ? 'green' : 'red';
                print OUT "&$lstatus $line";
            }
            elsif ($line =~ /^\s*(\d+)\s+(\S+)\s+\S+\s+(\d+)\s+(\d+)\s+(\d+)\s+\S+\s+\S+\s+(\S+)\s+(.*)$/)
            {
                my ($aname, $value, $worst, $thresh, $failure, $raw) = ($2, $3, $4, $5, $6, $7);
                my $lstatus = 'green';
                if ($aname =~ /Current_Pending_Sector|Offline_Uncorrectable/ and int($raw) > 0)
                {
                    $lstatus = 'yellow';
                }
                elsif ($aname =~ /Power_On_Hours/)
                {
                    $ponhours = $raw;
                }
                if ($failure =~ /FAIL/) {
                    $lstatus = 'red';
                }

                print OUT "&$lstatus $line";

                $ncv .= "$name-$aname-value : $value\n";
                $ncv .= "$name-$aname-worst : $worst\n";
                $ncv .= "$name-$aname-thresh : $thresh\n";
                $ncv .= "$name-$aname-raw : $raw\n";
            }
            elsif ($line =~ /^\s*No Errors Logged\s*$/)
            {
                $newerr = 0;
                print OUT "&green $line";
            }
            elsif ($line =~ /Error Count:\s*(\d+)/)
            {
                $newerr = $1 - $olderr{$name};
                my $lstatus = $newerr > 0 ? 'red' : 'green';
                print OUT "&$lstatus $line"
            }
            elsif ($line =~ /^\s*Error \d+ occurred/)
            {
                my $lstatus = $newerr > 0 ? 'red' : 'green';
                print OUT "&$lstatus $line"
            }
            elsif ($line =~ /^\s*#\s*\d+\s+(Conveyance offline|Extended offline|Short offline|Extended captive)\s+(.*)\s+\d+%\s+(\d+)/)
            {
                my $status = $2;
                my $lifetime = $3;
                my $lstatus = 'red';
                $lasttest = $lifetime if (!defined($lasttest));
                $lstatus = 'yellow' if ($status =~ /Aborted by host|Interrupted \(host reset\)/);
                $lstatus = 'green' if ($status =~ /Completed without error|Self-test routine in progress|Interrupted \(host reset\)/);
                print OUT "&$lstatus $line";
            }
            else
            {
                print OUT "   $line";
            }
        }
        # test status footer
        my $lasttestage = $ponhours % 65536 - $lasttest;
        my $lasttestmsg = "$lasttestage hours ago";
        my $lasttestcolor = 'green';
        if (!defined($lasttest))
        {
                $lasttestcolor = 'yellow';
                $lasttestmsg = 'no test performed';
        }
        elsif ($lasttestage > 24 * 7)
        {
                $lasttestcolor = 'red';
        }
        elsif ($lasttestage > 24 * 2)
        {
                $lasttestcolor = 'yellow';
        }
        print OUT "&$lasttestcolor Last Self-test: $lasttestmsg\n";

        # hidden output for ncv
        print OUT "<!--\n$ncv\n-->\n";
    }

    close OUT;
}

# send cached content
{
    open IN, $CACHEFILE or die "cannot open $CACHEFILE";
    while (my $line = <IN>)
    {
        if ($line =~ /^\s*&(\S+)/)
        {
            $bb->color_print($1, $line);
        }
        else
        {
            $bb->print($line);
        }
    }
    close IN;
}


$bb->send;

sub load_config
{
    my $path = shift;

    open C, "<$path" or return;
#    print "loading config from $path\n";
    while (my $line = <C>)
    {
        next if ($line =~ /^\s*#/);
        if ($line =~ /DISKS\s*=\s*['"](.*?)["']/)
        {
            @disks = split(/\s+/, $1);
        }
        if ($line =~ /SMARTOLDERROR\[([\w\/]+)\]\s+(\d+)/)
        {
           $olderr{$1} = $2;
        }
        if ($line =~ /SMARTCACHETIME=(\d+)/)
        {
           $CACHETIME = $1;
        }
    }
    close C;
}