[Mon-commit] mon/mon.d raid.monitor,NONE,1.1

[Augie S. <as...@us...>]

Update of /cvsroot/mon/mon/mon.d
In directory 23jxhf1.ch3.sourceforge.com:/tmp/cvs-serv25390

Added Files:
	raid.monitor 
Log Message:
Adding raid.monitor the unified software/hardware RAID monitor.

--- NEW FILE: raid.monitor ---
#!/usr/bin/perl

# raid.monitor
# A unified RAID monitor to monitor both software and hardware RAID.
# Works with Linux software RAID, 3ware's tw_cli tool, and Areca's areca_cli tool.
# Augie Schwer <au...@co...>

use strict;
use warnings;

my $MDSTAT= '/proc/mdstat';
my $TWCLI = `which tw_cli 2> /dev/null`;
my $ARCCLI= `which areca_cli 2> /dev/null`;

chomp $TWCLI;
chomp $ARCCLI;

my $mdstat = '';

# Check for any failed software RAID.
if ( -f $MDSTAT )
{
	$mdstat = `egrep "\[.*_.*\]" $MDSTAT`;
	$mdstat = '' if $mdstat =~ 'read_ahead not set'; # Fix for Red Hat 7.3 boxes.
}
if ( $mdstat )
{
	print "Failure of one or more software RAID devices\n";
        exit 1;
}

# Check for any 3ware RAID failures.
if ( -x $TWCLI )
{
	my %controllers = ();
	# Handle multiple controllers.
	my @controllers = `$TWCLI info | grep '^c' | awk {'print \$1'}`;
	my (@units,@ports);

	# Iterate through controllers and find broken ports and units.
	for my $controller (@controllers)
	{
		chomp $controller;
		@{$controllers{$controller}->{'failed_units'}} = `$TWCLI info $controller | grep -v 'NOT-PRESENT' | grep '^u' | awk {'print \$3'} | grep -v 'OK' | grep -v 'VERIFYING'`;
		@{$controllers{$controller}->{'failed_ports'}} = `$TWCLI info $controller | grep -v 'NOT-PRESENT' | grep '^p' | awk {'print \$2'} | grep -v 'OK' | grep -v 'VERIFYING'`;
	}

	# Display failure and exit with an error if there are any failed ports or units.
	for my $controller (@controllers)
	{
		if ( @{$controllers{$controller}->{'failed_units'}} or @{$controllers{$controller}->{'failed_ports'}} )
		{
			print "Failure of one or more of the hardware RAID devices.\n";
			exit 1;
		}
	}
}

# Check for any Areca RAID failures.
if ( -x $ARCCLI )
{
	my @failed_raid_sets = `$ARCCLI rsf info | grep 'Raid Set' | grep -v 'Normal'`;
	# Display failure and exit with an error if there were any failed Raid Sets.
	if ( @failed_raid_sets )
	{
			print "Failure of one or more of the hardware RAID devices.\n";
			exit 1;
	}
}

# Otherwise everything was OK.
print "Success\n";
exit 0;