Initial import from http://exchange.nagios.org/directory/Uncategorized/Check-SMART-status-modified/details
Benjamin Renard commited on 2011-05-13 13:02:32
Showing 1 changed files, with 324 additions and 0 deletions.
- check_smart 0000000..03d7917
| ... | ... |
@@ -0,0 +1,324 @@ |
| 1 |
+#!/usr/bin/perl -w |
|
| 2 |
+# Check SMART status of ATA/SCSI disks, returning any usable metrics as perfdata. |
|
| 3 |
+# For usage information, run ./check_smart -h |
|
| 4 |
+# |
|
| 5 |
+# This script was created under contract for the US Government and is therefore Public Domain |
|
| 6 |
+# |
|
| 7 |
+# Changes and Modifications |
|
| 8 |
+# ========================= |
|
| 9 |
+# Feb 3, 2009: Kurt Yoder - initial version of script 1.0 |
|
| 10 |
+# Jan 27, 2010: Philippe Genonceaux - modifications for compatibility with megaraid, use smartmontool version >= 5.39 |
|
| 11 |
+# Add this line to /etc/sudoers: "nagios ALL=(root) NOPASSWD: /usr/sbin/smartctl" |
|
| 12 |
+ |
|
| 13 |
+use strict; |
|
| 14 |
+use Getopt::Long; |
|
| 15 |
+ |
|
| 16 |
+use File::Basename qw(basename); |
|
| 17 |
+my $basename = basename($0); |
|
| 18 |
+ |
|
| 19 |
+my $revision = '$Revision: 1.0.1 $'; |
|
| 20 |
+ |
|
| 21 |
+use lib '/usr/lib/nagios/plugins/'; |
|
| 22 |
+use utils qw(%ERRORS &print_revision &support &usage); |
|
| 23 |
+ |
|
| 24 |
+$ENV{'PATH'}='/bin:/usr/bin:/sbin:/usr/sbin';
|
|
| 25 |
+$ENV{'BASH_ENV'}='';
|
|
| 26 |
+$ENV{'ENV'}='';
|
|
| 27 |
+ |
|
| 28 |
+use vars qw($opt_d $opt_debug $opt_h $opt_i $opt_n $opt_v); |
|
| 29 |
+Getopt::Long::Configure('bundling');
|
|
| 30 |
+GetOptions( |
|
| 31 |
+ "debug" => \$opt_debug, |
|
| 32 |
+ "d=s" => \$opt_d, "device=s" => \$opt_d, |
|
| 33 |
+ "h" => \$opt_h, "help" => \$opt_h, |
|
| 34 |
+ "i=s" => \$opt_i, "interface=s" => \$opt_i, |
|
| 35 |
+ "n=s" => \$opt_n, "number=s" => \$opt_n, |
|
| 36 |
+ "v" => \$opt_v, "version" => \$opt_v, |
|
| 37 |
+); |
|
| 38 |
+ |
|
| 39 |
+if ($opt_v) {
|
|
| 40 |
+ print_revision($basename,$revision); |
|
| 41 |
+ exit $ERRORS{'OK'};
|
|
| 42 |
+} |
|
| 43 |
+ |
|
| 44 |
+if ($opt_h) {
|
|
| 45 |
+ print_help(); |
|
| 46 |
+ exit $ERRORS{'OK'};
|
|
| 47 |
+} |
|
| 48 |
+my ($device, $interface, $number) = qw//; |
|
| 49 |
+if ($opt_d) {
|
|
| 50 |
+ unless($opt_i){
|
|
| 51 |
+ print "must specify an interface for $opt_d using -i/--interface!\n\n"; |
|
| 52 |
+ print_help(); |
|
| 53 |
+ exit $ERRORS{'UNKNOWN'};
|
|
| 54 |
+ } |
|
| 55 |
+ |
|
| 56 |
+ if (-b $opt_d){
|
|
| 57 |
+ $device = $opt_d; |
|
| 58 |
+ } |
|
| 59 |
+ else{
|
|
| 60 |
+ print "$opt_d is not a valid block device!\n\n"; |
|
| 61 |
+ print_help(); |
|
| 62 |
+ exit $ERRORS{'UNKNOWN'};
|
|
| 63 |
+ } |
|
| 64 |
+ |
|
| 65 |
+ if(grep {$opt_i eq $_} ('ata', 'scsi', 'megaraid')){
|
|
| 66 |
+ $interface = $opt_i; |
|
| 67 |
+ if($interface eq 'megaraid'){
|
|
| 68 |
+ if(defined($opt_n)){
|
|
| 69 |
+ $number = $opt_n; |
|
| 70 |
+ $interface = $opt_i.",".$number; |
|
| 71 |
+ } |
|
| 72 |
+ else{
|
|
| 73 |
+ print "must specify a physical disk number within the MegaRAID controller!\n\n"; |
|
| 74 |
+ print_help(); |
|
| 75 |
+ exit $ERRORS{'UNKNOWN'};
|
|
| 76 |
+ } |
|
| 77 |
+ } |
|
| 78 |
+ } |
|
| 79 |
+ else{
|
|
| 80 |
+ print "invalid interface $opt_i for $opt_d!\n\n"; |
|
| 81 |
+ print_help(); |
|
| 82 |
+ exit $ERRORS{'UNKNOWN'};
|
|
| 83 |
+ } |
|
| 84 |
+} |
|
| 85 |
+else{
|
|
| 86 |
+ print "must specify a device!\n\n"; |
|
| 87 |
+ print_help(); |
|
| 88 |
+ exit $ERRORS{'UNKNOWN'};
|
|
| 89 |
+} |
|
| 90 |
+my $smart_command = '/usr/bin/sudo /usr/sbin/smartctl'; |
|
| 91 |
+my @error_messages = qw//; |
|
| 92 |
+my $exit_status = 'OK'; |
|
| 93 |
+ |
|
| 94 |
+ |
|
| 95 |
+warn "###########################################################\n" if $opt_debug; |
|
| 96 |
+warn "(debug) CHECK 1: getting overall SMART health status\n" if $opt_debug; |
|
| 97 |
+warn "###########################################################\n\n\n" if $opt_debug; |
|
| 98 |
+ |
|
| 99 |
+my $full_command = "$smart_command -d $interface -H $device"; |
|
| 100 |
+warn "(debug) executing:\n$full_command\n\n" if $opt_debug; |
|
| 101 |
+ |
|
| 102 |
+my @output = `$full_command`; |
|
| 103 |
+warn "(debug) output:\n@output\n\n" if $opt_debug; |
|
| 104 |
+ |
|
| 105 |
+# parse ata output, looking for "health status: passed" |
|
| 106 |
+my $found_status = 0; |
|
| 107 |
+my $line_str = 'SMART overall-health self-assessment test result: '; # ATA SMART line |
|
| 108 |
+my $ok_str = 'PASSED'; # ATA SMART OK string |
|
| 109 |
+ |
|
| 110 |
+if ($interface eq 'megaraid'.",".$number or 'scsi'){
|
|
| 111 |
+ $line_str = 'SMART Health Status: '; # SCSI OR MEGARAID SMART line |
|
| 112 |
+ $ok_str = 'OK'; #SCSI OR MEGARAID SMART OK string |
|
| 113 |
+} |
|
| 114 |
+ |
|
| 115 |
+foreach my $line (@output){
|
|
| 116 |
+ if($line =~ /$line_str(.+)/){
|
|
| 117 |
+ $found_status = 1; |
|
| 118 |
+ warn "(debug) parsing line:\n$line\n\n" if $opt_debug; |
|
| 119 |
+ if ($1 eq $ok_str) {
|
|
| 120 |
+ warn "(debug) found string '$ok_str'; status OK\n\n" if $opt_debug; |
|
| 121 |
+ } |
|
| 122 |
+ else {
|
|
| 123 |
+ warn "(debug) no '$ok_str' status; failing\n\n" if $opt_debug; |
|
| 124 |
+ push(@error_messages, "Health status: $1"); |
|
| 125 |
+ escalate_status('CRITICAL');
|
|
| 126 |
+ } |
|
| 127 |
+ } |
|
| 128 |
+} |
|
| 129 |
+ |
|
| 130 |
+unless ($found_status) {
|
|
| 131 |
+ push(@error_messages, 'No health status line found'); |
|
| 132 |
+ escalate_status('UNKNOWN');
|
|
| 133 |
+} |
|
| 134 |
+ |
|
| 135 |
+ |
|
| 136 |
+warn "###########################################################\n" if $opt_debug; |
|
| 137 |
+warn "(debug) CHECK 2: getting silent SMART health check\n" if $opt_debug; |
|
| 138 |
+warn "###########################################################\n\n\n" if $opt_debug; |
|
| 139 |
+ |
|
| 140 |
+$full_command = "$smart_command -d $interface -q silent -A $device"; |
|
| 141 |
+warn "(debug) executing:\n$full_command\n\n" if $opt_debug; |
|
| 142 |
+ |
|
| 143 |
+system($full_command); |
|
| 144 |
+my $return_code = $?; |
|
| 145 |
+warn "(debug) exit code:\n$return_code\n\n" if $opt_debug; |
|
| 146 |
+ |
|
| 147 |
+if ($return_code & 0x01) {
|
|
| 148 |
+ push(@error_messages, 'Commandline parse failure'); |
|
| 149 |
+ escalate_status('UNKNOWN');
|
|
| 150 |
+} |
|
| 151 |
+if ($return_code & 0x02) {
|
|
| 152 |
+ push(@error_messages, 'Device could not be opened'); |
|
| 153 |
+ escalate_status('UNKNOWN');
|
|
| 154 |
+} |
|
| 155 |
+if ($return_code & 0x04) {
|
|
| 156 |
+ push(@error_messages, 'Checksum failure'); |
|
| 157 |
+ escalate_status('WARNING');
|
|
| 158 |
+} |
|
| 159 |
+if ($return_code & 0x08) {
|
|
| 160 |
+ push(@error_messages, 'Disk is failing'); |
|
| 161 |
+ escalate_status('CRITICAL');
|
|
| 162 |
+} |
|
| 163 |
+if ($return_code & 0x10) {
|
|
| 164 |
+ push(@error_messages, 'Disk is in prefail'); |
|
| 165 |
+ escalate_status('WARNING');
|
|
| 166 |
+} |
|
| 167 |
+if ($return_code & 0x20) {
|
|
| 168 |
+ push(@error_messages, 'Disk may be close to failure'); |
|
| 169 |
+ escalate_status('WARNING');
|
|
| 170 |
+} |
|
| 171 |
+if ($return_code & 0x40) {
|
|
| 172 |
+ push(@error_messages, 'Error log contains errors'); |
|
| 173 |
+ escalate_status('WARNING');
|
|
| 174 |
+} |
|
| 175 |
+if ($return_code & 0x80) {
|
|
| 176 |
+ push(@error_messages, 'Self-test log contains errors'); |
|
| 177 |
+ escalate_status('WARNING');
|
|
| 178 |
+} |
|
| 179 |
+if ($return_code && !$exit_status) {
|
|
| 180 |
+ push(@error_messages, 'Unknown return code'); |
|
| 181 |
+ escalate_status('CRITICAL');
|
|
| 182 |
+} |
|
| 183 |
+ |
|
| 184 |
+if ($return_code) {
|
|
| 185 |
+ warn "(debug) non-zero exit code, generating error condition\n\n" if $opt_debug; |
|
| 186 |
+} |
|
| 187 |
+else {
|
|
| 188 |
+ warn "(debug) zero exit code, status OK\n\n" if $opt_debug; |
|
| 189 |
+} |
|
| 190 |
+ |
|
| 191 |
+ |
|
| 192 |
+warn "###########################################################\n" if $opt_debug; |
|
| 193 |
+warn "(debug) CHECK 3: getting detailed statistics\n" if $opt_debug; |
|
| 194 |
+warn "(debug) information contains a few more potential trouble spots\n" if $opt_debug; |
|
| 195 |
+warn "(debug) plus, we can also use the information for perfdata/graphing\n" if $opt_debug; |
|
| 196 |
+warn "###########################################################\n\n\n" if $opt_debug; |
|
| 197 |
+ |
|
| 198 |
+$full_command = "$smart_command -d $interface -A $device"; |
|
| 199 |
+warn "(debug) executing:\n$full_command\n\n" if $opt_debug; |
|
| 200 |
+@output = `$full_command`; |
|
| 201 |
+warn "(debug) output:\n@output\n\n" if $opt_debug; |
|
| 202 |
+my @perfdata = qw//; |
|
| 203 |
+ |
|
| 204 |
+# separate metric-gathering and output analysis for ATA vs SCSI SMART output |
|
| 205 |
+if ($interface eq 'ata'){
|
|
| 206 |
+ foreach my $line(@output){
|
|
| 207 |
+ # get lines that look like this: |
|
| 208 |
+ # 9 Power_On_Minutes 0x0032 241 241 000 Old_age Always - 113h+12m |
|
| 209 |
+ next unless $line =~ /^\s*\d+\s(\S+)\s+(?:\S+\s+){6}(\S+)\s+(\d+)/;
|
|
| 210 |
+ my ($attribute_name, $when_failed, $raw_value) = ($1, $2, $3); |
|
| 211 |
+ if ($when_failed ne '-'){
|
|
| 212 |
+ push(@error_messages, "Attribute $attribute_name failed at $when_failed"); |
|
| 213 |
+ escalate_status('WARNING');
|
|
| 214 |
+ warn "(debug) parsed SMART attribute $attribute_name with error condition:\n$when_failed\n\n" if $opt_debug; |
|
| 215 |
+ } |
|
| 216 |
+ # some attributes produce questionable data; no need to graph them |
|
| 217 |
+ if (grep {$_ eq $attribute_name} ('Unknown_Attribute', 'Power_On_Minutes') ){
|
|
| 218 |
+ next; |
|
| 219 |
+ } |
|
| 220 |
+ push (@perfdata, "$attribute_name=$raw_value"); |
|
| 221 |
+ |
|
| 222 |
+ # do some manual checks |
|
| 223 |
+ if ( ($attribute_name eq 'Current_Pending_Sector') && $raw_value ) {
|
|
| 224 |
+ push(@error_messages, "Sectors pending re-allocation"); |
|
| 225 |
+ escalate_status('WARNING');
|
|
| 226 |
+ warn "(debug) Current_Pending_Sector is non-zero ($raw_value)\n\n" if $opt_debug; |
|
| 227 |
+ } |
|
| 228 |
+ } |
|
| 229 |
+} |
|
| 230 |
+else{
|
|
| 231 |
+ my ($current_temperature, $max_temperature, $current_start_stop, $max_start_stop) = qw//; |
|
| 232 |
+ foreach my $line(@output){
|
|
| 233 |
+ if ($line =~ /Current Drive Temperature:\s+(\d+)/){
|
|
| 234 |
+ $current_temperature = $1; |
|
| 235 |
+ } |
|
| 236 |
+ elsif ($line =~ /Drive Trip Temperature:\s+(\d+)/){
|
|
| 237 |
+ $max_temperature = $1; |
|
| 238 |
+ } |
|
| 239 |
+ elsif ($line =~ /Current start stop count:\s+(\d+)/){
|
|
| 240 |
+ $current_start_stop = $1; |
|
| 241 |
+ } |
|
| 242 |
+ elsif ($line =~ /Recommended maximum start stop count:\s+(\d+)/){
|
|
| 243 |
+ $max_start_stop = $1; |
|
| 244 |
+ } |
|
| 245 |
+ elsif ($line =~ /Elements in grown defect list:\s+(\d+)/){
|
|
| 246 |
+ push (@perfdata, "defect_list=$1"); |
|
| 247 |
+ } |
|
| 248 |
+ elsif ($line =~ /Blocks sent to initiator =\s+(\d+)/){
|
|
| 249 |
+ push (@perfdata, "sent_blocks=$1"); |
|
| 250 |
+ } |
|
| 251 |
+ } |
|
| 252 |
+ if($current_temperature){
|
|
| 253 |
+ if($max_temperature){
|
|
| 254 |
+ push (@perfdata, "temperature=$current_temperature;;$max_temperature"); |
|
| 255 |
+ if($current_temperature > $max_temperature){
|
|
| 256 |
+ warn "(debug) Disk temperature is greater than max ($current_temperature > $max_temperature)\n\n" if $opt_debug; |
|
| 257 |
+ push(@error_messages, 'Disk temperature is higher than maximum'); |
|
| 258 |
+ escalate_status('CRITICAL');
|
|
| 259 |
+ } |
|
| 260 |
+ } |
|
| 261 |
+ else{
|
|
| 262 |
+ push (@perfdata, "temperature=$current_temperature"); |
|
| 263 |
+ } |
|
| 264 |
+ } |
|
| 265 |
+ if($current_start_stop){
|
|
| 266 |
+ if($max_start_stop){
|
|
| 267 |
+ push (@perfdata, "start_stop=$current_start_stop;$max_start_stop"); |
|
| 268 |
+ if($current_start_stop > $max_start_stop){
|
|
| 269 |
+ warn "(debug) Disk start_stop is greater than max ($current_start_stop > $max_start_stop)\n\n" if $opt_debug; |
|
| 270 |
+ push(@error_messages, 'Disk start_stop is higher than maximum'); |
|
| 271 |
+ escalate_status('WARNING');
|
|
| 272 |
+ } |
|
| 273 |
+ } |
|
| 274 |
+ else{
|
|
| 275 |
+ push (@perfdata, "start_stop=$current_start_stop"); |
|
| 276 |
+ } |
|
| 277 |
+ } |
|
| 278 |
+} |
|
| 279 |
+warn "(debug) gathered perfdata:\n@perfdata\n\n" if $opt_debug; |
|
| 280 |
+my $perf_string = join(' ', @perfdata);
|
|
| 281 |
+ |
|
| 282 |
+warn "###########################################################\n" if $opt_debug; |
|
| 283 |
+warn "(debug) FINAL STATUS: $exit_status\n" if $opt_debug; |
|
| 284 |
+warn "###########################################################\n\n\n" if $opt_debug; |
|
| 285 |
+ |
|
| 286 |
+warn "(debug) final status/output:\n" if $opt_debug; |
|
| 287 |
+ |
|
| 288 |
+my $status_string = ''; |
|
| 289 |
+ |
|
| 290 |
+if($exit_status ne 'OK'){
|
|
| 291 |
+ $status_string = "$exit_status: ".join(', ', @error_messages);
|
|
| 292 |
+} |
|
| 293 |
+else {
|
|
| 294 |
+ $status_string = "OK: no SMART errors detected"; |
|
| 295 |
+} |
|
| 296 |
+ |
|
| 297 |
+print "$status_string|$perf_string\n"; |
|
| 298 |
+exit $ERRORS{$exit_status};
|
|
| 299 |
+ |
|
| 300 |
+sub print_help {
|
|
| 301 |
+ print_revision($basename,$revision); |
|
| 302 |
+ print "Usage: $basename (--device=<SMART device> --interface=(ata|scsi)|-h|-v) [--debug]\n"; |
|
| 303 |
+ print " --debug: show debugging information\n"; |
|
| 304 |
+ print " -d/--device: a device to be SMART monitored, eg /dev/sda\n"; |
|
| 305 |
+ print " -i/--interface: ata, scsi, megaraid, depending upon the device's interface type\n"; |
|
| 306 |
+ print " -n/--number: where in the argument megaraid, it is the physical disk number within the MegaRAID controller\n"; |
|
| 307 |
+ print " -h/--help: this help\n"; |
|
| 308 |
+ print " -v/--version: Version number\n"; |
|
| 309 |
+ support(); |
|
| 310 |
+} |
|
| 311 |
+ |
|
| 312 |
+# escalate an exit status IFF it's more severe than the previous exit status |
|
| 313 |
+sub escalate_status {
|
|
| 314 |
+ my $requested_status = shift; |
|
| 315 |
+ # no test for 'CRITICAL'; automatically escalates upwards |
|
| 316 |
+ if ($requested_status eq 'WARNING') {
|
|
| 317 |
+ return if $exit_status eq 'CRITICAL'; |
|
| 318 |
+ } |
|
| 319 |
+ if ($requested_status eq 'UNKNOWN') {
|
|
| 320 |
+ return if $exit_status eq 'WARNING'; |
|
| 321 |
+ return if $exit_status eq 'CRITICAL'; |
|
| 322 |
+ } |
|
| 323 |
+ $exit_status = $requested_status; |
|
| 324 |
+} |
|
| 0 | 325 |