| 1 |
stephdl |
1.1 |
--- smeserver-zabbix-agent-0.1/root/etc/e-smith/templates/etc/sudoers/00zabbixAgentAlias.megaraid_parser 2009-04-24 22:28:07.000000000 +0200 |
| 2 |
|
|
+++ smeserver-zabbix-agent-0.1/root/etc/e-smith/templates/etc/sudoers/00zabbixAgentAlias 2009-04-26 18:43:19.000000000 +0200 |
| 3 |
|
|
@@ -1,7 +1,7 @@ |
| 4 |
|
|
{ |
| 5 |
|
|
my $runasroot = '/usr/bin/mysqladmin status, /sbin/e-smith/db yum_updates show, /var/lib/zabbix/bin/sensors *'; |
| 6 |
|
|
if ( -x '/opt/MegaRAID/MegaCli/MegaCli' ){ |
| 7 |
|
|
- $runasroot .= ', /opt/MegaRAID/MegaCli/MegaCli -ldinfo -Lall -Aall'; |
| 8 |
|
|
+ $runasroot .= ', /opt/MegaRAID/MegaCli/MegaCli *'; |
| 9 |
|
|
} |
| 10 |
|
|
$runasroot .= "\n"; |
| 11 |
|
|
$OUT .= 'Cmnd_Alias ZABBIX_AGENT_ROOT = '.$runasroot |
| 12 |
|
|
--- smeserver-zabbix-agent-0.1/root/etc/e-smith/templates/etc/zabbix/zabbix_agentd.conf/90UserParameters_megaRaid.megaraid_parser 2009-04-24 22:28:07.000000000 +0200 |
| 13 |
|
|
+++ smeserver-zabbix-agent-0.1/root/etc/e-smith/templates/etc/zabbix/zabbix_agentd.conf/90UserParameters_megaRaid 2009-04-26 19:39:57.000000000 +0200 |
| 14 |
|
|
@@ -15,13 +15,12 @@ |
| 15 |
|
|
# Show Value: As is |
| 16 |
|
|
|
| 17 |
|
|
# The value reported is like: |
| 18 |
|
|
-# State: Optimal |
| 19 |
|
|
-# State: Degraded |
| 20 |
|
|
+# State: OK: 0:0:RAID-1:2 drives:68GB:Optimal 0:1:RAID-5:4 drives:837GB:Optimal Drives:7 |
| 21 |
|
|
# |
| 22 |
|
|
|
| 23 |
|
|
# Tips: You can add a simple trigger on this check like: |
| 24 |
|
|
-# \{ hostname:raid.mega.status.str( Degraded ) \}=1 |
| 25 |
|
|
-UserParameter=raid.mega.status,/usr/bin/sudo /opt/MegaRAID/MegaCli/MegaCli -ldinfo -Lall -Aall | grep "^State:" |
| 26 |
|
|
+# \{ hostname:raid.mega.status.str( OK ) \}=0 |
| 27 |
|
|
+UserParameter=raid.mega.status,/var/lib/zabbix/bin/megaraid-parser.pl |
| 28 |
|
|
|
| 29 |
|
|
HERE |
| 30 |
|
|
} |
| 31 |
|
|
--- smeserver-zabbix-agent-0.1/root/var/lib/zabbix/bin/megaraid-parser.pl.megaraid_parser 2009-04-26 18:39:10.000000000 +0200 |
| 32 |
|
|
+++ smeserver-zabbix-agent-0.1/root/var/lib/zabbix/bin/megaraid-parser.pl 2009-04-26 18:38:39.000000000 +0200 |
| 33 |
|
|
@@ -0,0 +1,194 @@ |
| 34 |
|
|
+#!/usr/bin/perl -w |
| 35 |
|
|
+ |
| 36 |
|
|
+# check_megaraid_sas Nagios plugin |
| 37 |
|
|
+# Copyright (C) 2007 Jonathan Delgado, delgado@molbio.mgh.harvard.edu |
| 38 |
|
|
+# |
| 39 |
|
|
+# This program is free software; you can redistribute it and/or |
| 40 |
|
|
+# modify it under the terms of the GNU General Public License |
| 41 |
|
|
+# as published by the Free Software Foundation; either version 2 |
| 42 |
|
|
+# of the License, or (at your option) any later version. |
| 43 |
|
|
+# |
| 44 |
|
|
+# This program is distributed in the hope that it will be useful, |
| 45 |
|
|
+# but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 46 |
|
|
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 47 |
|
|
+# GNU General Public License for more details. |
| 48 |
|
|
+# |
| 49 |
|
|
+# You should have received a copy of the GNU General Public License |
| 50 |
|
|
+# along with this program; if not, write to the Free Software |
| 51 |
|
|
+# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. |
| 52 |
|
|
+# |
| 53 |
|
|
+# |
| 54 |
|
|
+# Nagios plugin to monitor the status of volumes attached to a LSI Megaraid SAS |
| 55 |
|
|
+# controller, such as the Dell PERC5/i and PERC5/e. If you have any hotspares |
| 56 |
|
|
+# attached to the controller, you can specify the number you should expect to |
| 57 |
|
|
+# find with the '-s' flag. |
| 58 |
|
|
+# |
| 59 |
|
|
+# The paths for the Nagios plugins lib and MegaCli may need to me changed. |
| 60 |
|
|
+# |
| 61 |
|
|
+# $Author: delgado $ |
| 62 |
|
|
+# $Revision: #3 $ $Date: 2007/06/07 $ |
| 63 |
|
|
+ |
| 64 |
|
|
+# Slightly modified by Daniel B. for SME Server integration with zabbix |
| 65 |
|
|
+# 23 Apr 2009 |
| 66 |
|
|
+ |
| 67 |
|
|
+use strict; |
| 68 |
|
|
+use Getopt::Std; |
| 69 |
|
|
+ |
| 70 |
|
|
+our($opt_h, $opt_s); |
| 71 |
|
|
+ |
| 72 |
|
|
+ |
| 73 |
|
|
+getopts('hs:'); |
| 74 |
|
|
+ |
| 75 |
|
|
+if ( $opt_h ) { |
| 76 |
|
|
+ print "usage w/o hotspare: $0\n"; |
| 77 |
|
|
+ print " w/ hotspare: $0 -s <number of hotspares>\n"; |
| 78 |
|
|
+ exit; |
| 79 |
|
|
+} |
| 80 |
|
|
+ |
| 81 |
|
|
+my $megacli = '/usr/bin/sudo /opt/MegaRAID/MegaCli/MegaCli'; |
| 82 |
|
|
+ |
| 83 |
|
|
+## Return codes for Nagios |
| 84 |
|
|
+my %ERRORS=('OK'=>0,'WARNING'=>1,'CRITICAL'=>2,'UNKNOWN'=>3,'DEPENDENT'=>4); |
| 85 |
|
|
+ |
| 86 |
|
|
+my ($adapters); |
| 87 |
|
|
+my $hotspares = 0; |
| 88 |
|
|
+my $hotsparecount = 0; |
| 89 |
|
|
+my $pdbad = 0; |
| 90 |
|
|
+my $pdcount = 0; |
| 91 |
|
|
+my $pderrors = 0; |
| 92 |
|
|
+my $result = ''; |
| 93 |
|
|
+my $status = 'OK'; |
| 94 |
|
|
+ |
| 95 |
|
|
+sub max_state ($$) { |
| 96 |
|
|
+ my ($current, $compare) = @_; |
| 97 |
|
|
+ |
| 98 |
|
|
+ if (($compare eq 'CRITICAL') || ($compare eq 'CRITICAL')) { |
| 99 |
|
|
+ return 'CRITICAL'; |
| 100 |
|
|
+ } elsif ($compare eq 'OK') { |
| 101 |
|
|
+ return $current; |
| 102 |
|
|
+ } elsif ($compare eq 'WARNING') { |
| 103 |
|
|
+ return 'WARNING'; |
| 104 |
|
|
+ } elsif (($compare eq 'UNKNOWN') && ($current eq 'OK')) { |
| 105 |
|
|
+ return 'UNKNOWN'; |
| 106 |
|
|
+ } else { |
| 107 |
|
|
+ return $current; |
| 108 |
|
|
+ } |
| 109 |
|
|
+} |
| 110 |
|
|
+ |
| 111 |
|
|
+ |
| 112 |
|
|
+if ( $opt_s ) { |
| 113 |
|
|
+ $hotspares = $opt_s; |
| 114 |
|
|
+} |
| 115 |
|
|
+ |
| 116 |
|
|
+# Get the number of RAID controllers we have |
| 117 |
|
|
+open (ADPCOUNT, "$megacli -adpCount |") |
| 118 |
|
|
+ || die "error: Could not execute MegaCli -adpCount"; |
| 119 |
|
|
+ |
| 120 |
|
|
+while (<ADPCOUNT>) { |
| 121 |
|
|
+ if ( m/Controller Count:\s*(\d+)/ ) { |
| 122 |
|
|
+ $adapters = $1; |
| 123 |
|
|
+ last; |
| 124 |
|
|
+ } |
| 125 |
|
|
+} |
| 126 |
|
|
+close ADPCOUNT; |
| 127 |
|
|
+ |
| 128 |
|
|
+ADAPTER: for ( my $adp = 0; $adp < $adapters; $adp++ ) { |
| 129 |
|
|
+ # Get the number of logical drives on this adapter |
| 130 |
|
|
+ open (LDGETNUM, "$megacli -LdGetNum -a$adp |") |
| 131 |
|
|
+ || die "error: Could not execute $megacli -LdGetNum -a$adp"; |
| 132 |
|
|
+ |
| 133 |
|
|
+ my ($ldnum); |
| 134 |
|
|
+ while (<LDGETNUM>) { |
| 135 |
|
|
+ if ( m/Number of Virtual drives configured on adapter \d:\s*(\d+)/ ) { |
| 136 |
|
|
+ $ldnum = $1; |
| 137 |
|
|
+ last; |
| 138 |
|
|
+ } |
| 139 |
|
|
+ } |
| 140 |
|
|
+ close LDGETNUM; |
| 141 |
|
|
+ |
| 142 |
|
|
+ LDISK: for ( my $ld = 0; $ld < $ldnum; $ld++ ) { |
| 143 |
|
|
+ # Get info on this particular logical drive |
| 144 |
|
|
+ open (LDINFO, "$megacli -LdInfo -L$ld -a$adp |") |
| 145 |
|
|
+ || die "error: Could not execute $megacli -LdInfo -L$ld -a$adp"; |
| 146 |
|
|
+ |
| 147 |
|
|
+ my ($size, $unit, $raidlevel, $ldpdcount, $state); |
| 148 |
|
|
+ while (<LDINFO>) { |
| 149 |
|
|
+ if ( m/Size:\s*((\d+)(MB|GB|TB))/ ) { |
| 150 |
|
|
+ $size = $2; |
| 151 |
|
|
+ $unit = $3; |
| 152 |
|
|
+ # Adjust MB to GB if that's what we got |
| 153 |
|
|
+ if ( $unit eq 'MB' ) { |
| 154 |
|
|
+ $size = sprintf( "%.0f", ($size / 1024) ); |
| 155 |
|
|
+ $unit= 'GB'; |
| 156 |
|
|
+ } |
| 157 |
|
|
+ } elsif ( m/State:\s*(\w+)/ ) { |
| 158 |
|
|
+ $state = $1; |
| 159 |
|
|
+ if ( $state ne 'Optimal' ) { |
| 160 |
|
|
+ $status = 'CRITICAL'; |
| 161 |
|
|
+ } |
| 162 |
|
|
+ } elsif ( m/Number Of Drives:\s*(\d+)/ ) { |
| 163 |
|
|
+ $ldpdcount = $1; |
| 164 |
|
|
+ } elsif ( m/RAID Level: Primary-(\d)/ ) { |
| 165 |
|
|
+ $raidlevel = $1; |
| 166 |
|
|
+ } |
| 167 |
|
|
+ } |
| 168 |
|
|
+ close LDINFO; |
| 169 |
|
|
+ |
| 170 |
|
|
+ $result .= "$adp:$ld:RAID-$raidlevel:$ldpdcount drives:$size$unit:$state "; |
| 171 |
|
|
+ |
| 172 |
|
|
+ } #LDISK |
| 173 |
|
|
+ close LDINFO; |
| 174 |
|
|
+ |
| 175 |
|
|
+ # Get info on physical disks for this adapter |
| 176 |
|
|
+ open (PDLIST, "$megacli -PdList -a$adp |") |
| 177 |
|
|
+ || die "error: Could not execute $megacli -PdList -a$adp"; |
| 178 |
|
|
+ |
| 179 |
|
|
+ my ($slotnumber,$fwstate); |
| 180 |
|
|
+ PDISKS: while (<PDLIST>) { |
| 181 |
|
|
+ if ( m/Slot Number:\s*(\d+)/ ) { |
| 182 |
|
|
+ $slotnumber = $1; |
| 183 |
|
|
+ $pdcount++ unless ( $slotnumber == 255 ); |
| 184 |
|
|
+ } elsif ( m/Error Count:\s*(\d+)/ ) { |
| 185 |
|
|
+ $pderrors += $1; |
| 186 |
|
|
+ } elsif ( m/Predictive Failure Count:\s*(\d+)/ ) { |
| 187 |
|
|
+ $pderrors += $1; |
| 188 |
|
|
+ } elsif ( m/Firmware state:\s*(\w+)/ ) { |
| 189 |
|
|
+ $fwstate = $1; |
| 190 |
|
|
+ if ( $fwstate eq 'Hotspare' ) { |
| 191 |
|
|
+ $hotsparecount++; |
| 192 |
|
|
+ } elsif ( $fwstate eq 'Online' ) { |
| 193 |
|
|
+ # Do nothing |
| 194 |
|
|
+ } elsif ( $slotnumber != 255 ) { |
| 195 |
|
|
+ $pdbad++; |
| 196 |
|
|
+ $status = 'CRITICAL'; |
| 197 |
|
|
+ } |
| 198 |
|
|
+ } |
| 199 |
|
|
+ } #PDISKS |
| 200 |
|
|
+ close PDLIST; |
| 201 |
|
|
+} |
| 202 |
|
|
+ |
| 203 |
|
|
+$result .= "Drives:$pdcount "; |
| 204 |
|
|
+ |
| 205 |
|
|
+# Any bad disks? |
| 206 |
|
|
+if ( $pdbad ) { |
| 207 |
|
|
+ $result .= "$pdbad Bad Drives "; |
| 208 |
|
|
+} |
| 209 |
|
|
+ |
| 210 |
|
|
+# Were there any errors? |
| 211 |
|
|
+if ( $pderrors ) { |
| 212 |
|
|
+ $result .= "($pderrors Errors) "; |
| 213 |
|
|
+ $status = max_state($status, 'WARNING'); |
| 214 |
|
|
+} |
| 215 |
|
|
+ |
| 216 |
|
|
+# Do we have as many hotspares as expected (if any) |
| 217 |
|
|
+if ( $hotspares ) { |
| 218 |
|
|
+ if ( $hotsparecount < $hotspares ) { |
| 219 |
|
|
+ $status = max_state($status, 'WARNING'); |
| 220 |
|
|
+ $result .= "Hotspare(s):$hotsparecount (of $hotspares)"; |
| 221 |
|
|
+ } else { |
| 222 |
|
|
+ $result .= "Hotspare(s):$hotsparecount"; |
| 223 |
|
|
+ } |
| 224 |
|
|
+} |
| 225 |
|
|
+ |
| 226 |
|
|
+print STDOUT "$status: $result\n"; |
| 227 |
|
|
+exit $ERRORS{$status}; |
Parent Directory
| 
Revision Log
| 
Revision Graph
