contrib-munin/plugins/disk/raid

#!/usr/bin/perl -w
# 
# (c) 2007 Nathan Rutman nathan@clusterfs.com
# 
# Plugin to monitor RAID status 
#
# Results are % of healthy drives in a raid device
# and % rebuilt of devices that are resyncing. 
#
#%# family=contrib
#%# capabilities=autoconf

if ($ARGV[0] and $ARGV[0] eq "autoconf") {
    if (-r "/proc/mdstat" and `grep md /proc/mdstat`) {
        print "yes\n";
        exit 0;
    } else {
        print "no RAID devices\n";
        exit 1;
    }
}

if ( $ARGV[0] and $ARGV[0] eq "config" ) {
    print "graph_title RAID status\n";
    print "graph_category disk\n";
    print "graph_info This graph monitors RAID disk health.  Values are percentage of healthy drives in each raid group.  Degraded devices are marked Critical.\n";
    print "graph_args --base 1000 -l 0\n";
    print "graph_vlabel % healthy/rebuilt\n";
    print "graph_scale  no\n";
}

open(my $mdstat, "/proc/mdstat");
my(@text) = <$mdstat>;
# contents of <$mdstat> may be changed at next reading, so fetch the contents at a time
close($mdstat);

my($devinfo_re, $devstat_re, $action_re) = (
    '(md\d+)\s+:\s+active\s+(\(auto-read-only\)\s+|)(\w+)\s+(.*)',
    '.*\[(\d+)\/(\d+)]\s+\[(\w+)]',
    '(.*(check|resync)\s=\s+(\d+\.\d+)%)',
);

my($dev, $type, $members, $nmem, $nact, $status, $proc);
while (@text) {
    my $line = shift @text;
    if ($line =~ /$devinfo_re/) {
        # first line should like "active raid1 sda1[0] sdc1[2] sdb1[1]"
        $dev =  $1;
        $type = $3;
        $members = $4;
        
        $line = shift @text;
        if ($line =~ /$devstat_re/) {
            # second line should like "123456 blocks super 1.2 [2/2] [UU]"
            $nmem = $1;
            $nact = $2;
            $status = $3;
        }
        else {
            # sencond line did not exist on /proc/mdstat
            next;
        }
        
        $line = shift @text;
        if ($line =~ /$action_re/) {
            # third line should like " [==>..................]  check = 10.0% (12345/123456) finish=123min speed=12345/sec"
            # this line will appear only when the array is in action
            $proc = $3;
        }
        else {
            # array is not in action
            unshift(@text, $line);
        }
    }
    else {
        # skip until first line is found
        next;
    }

    if ( $ARGV[0] and $ARGV[0] eq "config" ) {
        print "$dev.label $dev\n";
        print "$dev.info $type $members\n";
        # 100: means less than 100
        # Because of an unfound bug, sometimes reported as 99.XX even when OS reports 100.
        print "$dev.critical 98:\n";
        print $dev, "_rebuild.label $dev rebuilt\n";
        print $dev, "_rebuild.info $type\n";
        # Because of an unfound bug, sometimes reported as 99.XX even when OS reports 100.
        print $dev, "_rebuild.critical 98:\n";
        print $dev, "_check.label $dev check/resync \n";
        print $dev, "_check.info $type\n";
    } else {
        my $pct = 100 * $nact / $nmem;
        my $rpct = 100;
        if ( $pct < 100 ) {
            my @output = `/sbin/mdadm -D /dev/$dev | grep Rebuild`;
            if( $output[0] and $output[0] =~ /([0-9]+)% complete/ ) {
                    $rpct = $1;
            } else {
                    $rpct = 0;
            }
        }
        if ( $proc ) {
            $cpct = $proc;
        } else {
            $cpct = 0;
        }
        print "$dev.value $pct\n";
        print $dev, "_rebuild.value $rpct\n";
        print $dev, "_check.value $cpct\n";
    }
}

exit 0;