mdstat: Check if a certain amount of spares are present

lib/App/Monitoring/Plugin/CheckRaid/Plugins/mdstat.pm

@@ -161,6 +161,11 @@ sub check {
 	my (@status);
 	my @md = $this->parse;
 
+	my @spare_options = ();
+
+	@spare_options = split(/\,/, $this->{options}{mdstat_spare_count})
+		if (exists $this->{options}{mdstat_spare_count});
+
 	foreach (@md) {
 		my %md = %$_;
 
@@ -171,6 +176,26 @@ sub check {
 
 		# failed disks
 		my @fd = map { $_->{dev} } grep { $_->{flags} =~ /F/ } @{$md{disks}};
+		# spare disks
+		my @sd = map { $_->{dev} } grep { $_->{flags} =~ /S/ } @{$md{disks}};
+
+		my $spare_count = 0;
+		OPTION_LOOP:
+		{
+			foreach my $i (0 .. $#spare_options)
+			{
+				my ($disk, $value) = split(/:/, $spare_options[$i]);
+				for(@md)
+				{
+					if ($md{dev} eq $disk)
+					{
+						$spare_count = $value;
+						splice(@spare_options, $i, 1);
+						last OPTION_LOOP;
+					}
+				}
+			}
+		}
 
 		# raid0 is just there or its not. raid0 can't degrade.
 		# same for linear, no $md_status available
@@ -194,12 +219,27 @@ sub check {
 			# FIXME: this is same as above?
 			$this->warning;
 			$s .= "hot-spare failure:". join(",", @fd) .":$md{status}";
+		} elsif (@sd < $spare_count)
+		{
+			$this->warning;
+			$s .= "Array ".$md{dev}." should have ".$spare_count." spares, but has only ".(0+@sd)." spares";
 		} else {
 			$s .= "$md{status}";
 		}
 		push(@status, $s);
 	}
 
+	if (scalar @spare_options > 0)
+	{
+		$this->critical;
+		foreach (@spare_options)
+		{
+			my ($disk, $value) = split(/:/, $_);
+			my $s = "$disk is defined in spare_count option but could not be found!";
+			push(@status, $s);
+		}
+	}
+
 	return unless @status;
 
 	# denote this plugin as ran ok

t/check_mdstat.t

@@ -7,7 +7,7 @@ BEGIN {
 use strict;
 use warnings;
 use constant INACTIVE_TESTS => 2;
-use constant ACTIVE_TESTS => 19;
+use constant ACTIVE_TESTS => 26;
 use Test::More tests => 1 + ACTIVE_TESTS * 6 + INACTIVE_TESTS * 2;
 use test;
 
@@ -95,6 +95,42 @@ my @tests = (
 		active => 1,
 		message => 'md1(1.82 TiB raid1):hot-spare failure:sdh1:UUUUU, md0(119.18 GiB raid1):UU',
 	},
+	{ input => 'pr185_0', status => OK,
+		mdstat_spare_count => 'md1:1',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):UUUUU',
+	},
+	{ input => 'pr185_0', status => WARNING,
+		mdstat_spare_count => 'md1:2',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):Array md1 should have 2 spares, but has only 1 spares',
+	},
+	{ input => 'pr185_0', status => CRITICAL,
+		mdstat_spare_count => 'md2:2',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):UUUUU, md2 is defined in spare_count option but could not be found!',
+	},
+	{ input => 'pr185_1', status => OK,
+		mdstat_spare_count => 'md1:1,md2:2',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):UUUUU, md2(3.49 TiB raid5):UUUUU',
+	},
+	{ input => 'pr185_1', status => WARNING,
+		mdstat_spare_count => 'md1:1,md2:3',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):UUUUU, md2(3.49 TiB raid5):Array md2 should have 3 spares, but has only 2 spares',
+	},
+	{ input => 'pr185_1', status => WARNING,
+		mdstat_spare_count => 'md1:2,md2:3',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):Array md1 should have 2 spares, but has only 1 spares, md2(3.49 TiB raid5):Array md2 should have 3 spares, but has only 2 spares',
+	},
+	{ input => 'pr185_1', status => CRITICAL,
+		mdstat_spare_count => 'md1:1,md2:2,md3:1',
+		active => 1,
+		message => 'md1(3.49 TiB raid5):UUUUU, md2(3.49 TiB raid5):UUUUU, md3 is defined in spare_count option but could not be found!',
+	},
+
 );
 
 # test that plugin can be created
@@ -108,6 +144,10 @@ foreach my $test (@tests) {
 	if (defined $test->{check_status}) {
 		$options{check_status} = $test->{check_status};
 	}
+	if (defined $test->{mdstat_spare_count}) {
+		$options{mdstat_spare_count} = $test->{mdstat_spare_count};
+	}
+
 	my $plugin = mdstat->new(
 		commands => {
 			'mdstat' => ['<', TESTDIR . '/data/mdstat/' . $test->{input}],

t/data/mdstat/pr185_0

@@ -0,0 +1,6 @@
+Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5] [raid4] [raid10] 
+md1 : active raid5 sdh[3] sdg[2] sdj[6] sde[1] sdk[5](S) sdd[0]
+      3750244352 blocks super 1.2 level 5, 512k chunk, algorithm 2 [5/5] [UUUUU]
+      bitmap: 4/7 pages [16KB], 65536KB chunk
+
+unused devices: <none>

t/data/mdstat/pr185_1

@@ -0,0 +1,10 @@
+Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5] [raid4] [raid10] 
+md1 : active raid5 sdh[3] sdg[2] sdj[6] sde[1] sdk[5](S) sdd[0]
+      3750244352 blocks super 1.2 level 5, 512k chunk, algorithm 2 [5/5] [UUUUU]
+      bitmap: 4/7 pages [16KB], 65536KB chunk
+
+md2 : active raid5 sda[3] sdb[2] sdc[6] sdf[1] sdi[5](S) sdl[0](S) sdm[7]
+      3750244352 blocks super 1.2 level 5, 512k chunk, algorithm 2 [5/5] [UUUUU]
+      bitmap: 4/7 pages [16KB], 65536KB chunk
+
+unused devices: <none>