contrib/check_apc_ups.pl

   1 #! /usr/bin/perl -wT
   2 #
   3 # Check_apc_ups - Check APC UPS status via SNMP
   4 # Shamelessly copied from check_breeze.pl
   5 #
   6 # To do:
   7 # - Send SNMP queries directly, instead of forking `snmpget`.
   8 # - Make the status less verbose.  Maybe we can send an "onLine, time
   9 #   remaining: hh:mm:ss" if all is well, and a list of specific problems
  10 #   if something is broken.
  11
  12 use strict;
  13 use Getopt::Long;
  14 use vars qw($opt_V $opt_h $opt_H $opt_T $opt_t $opt_R $opt_r
  15   $opt_L $opt_l $PROGNAME);
  16 use lib "/usr/local/nagios/libexec";
  17 use utils qw(%ERRORS &print_revision &support &usage);
  18
  19 sub print_help ();
  20 sub print_usage ();
  21 sub get_snmp_int_val ($);
  22 sub escalate_exitval ($);
  23
  24 $ENV{'PATH'}='';
  25 $ENV{'BASH_ENV'}='';
  26 $ENV{'ENV'}='';
  27
  28 Getopt::Long::Configure('bundling');
  29 GetOptions
  30         ("V"   => \$opt_V, "version"            => \$opt_V,
  31          "h"   => \$opt_h, "help"               => \$opt_h,
  32          "T=s" => \$opt_T, "temp-critical"      => \$opt_T,
  33          "t=s" => \$opt_t, "temp-warning"       => \$opt_t,
  34          "R=s" => \$opt_R, "runtime-critical"   => \$opt_R,
  35          "r=s" => \$opt_r, "runtime-warning"    => \$opt_r,
  36          "L=s" => \$opt_L, "load-critical"      => \$opt_L,
  37          "l=s" => \$opt_l, "load-warning"       => \$opt_l,
  38          "H=s" => \$opt_H, "hostname=s"         => \$opt_H);
  39
  40 if ($opt_V) {
  41         print_revision($PROGNAME,'$Revision$');
  42         exit $ERRORS{'OK'};
  43 }
  44
  45 if ($opt_h) {print_help(); exit $ERRORS{'OK'};}
  46
  47 ($opt_H) || ($opt_H = shift) || usage("Host name/address not specified\n");
  48 my $host = $1 if ($opt_H =~ /([-.A-Za-z0-9]+)/);
  49 ($host) || usage("Invalid host: $opt_H\n");
  50
  51 # Defaults
  52
  53 $opt_R *= 60 * 100 if (defined $opt_R); # Convert minutes to secs/100
  54 $opt_r *= 60 * 100 if (defined $opt_R);
  55
  56 my $tempcrit    = $opt_T || 60;
  57 my $tempwarn    = $opt_t || 40;
  58 my $runtimecrit = $opt_R || 30 * 60 * 100;     # Secs / 100
  59 my $runtimewarn = $opt_r || 60 * 60 * 100;
  60 my $loadcrit    = $opt_L || 85;
  61 my $loadwarn    = $opt_l || 50;
  62
  63 if ($tempcrit !~ /\d+/) { usage ("Invalid critical temperature threshold.\n"); }
  64 if ($tempwarn !~ /\d+/) { usage ("Invalid critical temperature threshold.\n"); }
  65
  66 if ($runtimecrit !~ /\d+/) {
  67   usage ("Invalid critical run time threshold.\n");
  68 }
  69 if ($runtimewarn !~ /\d+/) {
  70   usage ("Invalid warning run time threshold.\n");
  71 }
  72
  73 if ($loadcrit !~ /\d+/ || $loadcrit < 0 || $loadcrit > 100) {
  74   usage ("Invalid critical load threshold.\n");
  75 }
  76 if ($loadwarn !~ /\d+/ || $loadwarn < 0 || $loadwarn > 100) {
  77   usage ("Invalid warning load threshold.\n");
  78 }
  79
  80
  81 # APC UPS OIDs
  82 # APC MIBs are available at ftp://ftp.apcftp.com/software/pnetmib/mib
  83 my $upsBasicOutputStatus          = ".1.3.6.1.4.1.318.1.1.1.4.1.1.0";
  84 my $upsBasicBatteryStatus         = ".1.3.6.1.4.1.318.1.1.1.2.1.1.0";
  85 my $upsAdvInputLineFailCause      = ".1.3.6.1.4.1.318.1.1.1.3.2.5.0";
  86 my $upsAdvBatteryTemperature      = ".1.3.6.1.4.1.318.1.1.1.2.2.2.0";
  87 my $upsAdvBatteryRunTimeRemaining = ".1.3.6.1.4.1.318.1.1.1.2.2.3.0";
  88 my $upsAdvBatteryReplaceIndicator = ".1.3.6.1.4.1.318.1.1.1.2.2.4.0";
  89 my $upsAdvOutputLoad              = ".1.3.6.1.4.1.318.1.1.1.4.2.3.0";
  90 my $upsAdvTestDiagnosticsResults  = ".1.3.6.1.4.1.318.1.1.1.7.2.3.0";
  91
  92 my @outputStatVals = (
  93   [ undef, undef ],                                     # pad 0
  94   [ undef, undef ],                                     # pad 1
  95   [ "onLine",                   $ERRORS{'OK'} ],        # 2
  96   [ "onBattery",                $ERRORS{'WARNING'} ],   # 3
  97   [ "onSmartBoost",             $ERRORS{'WARNING'} ],   # 4
  98   [ "timedSleeping",            $ERRORS{'WARNING'} ],   # 5
  99   [ "softwareBypass",           $ERRORS{'WARNING'} ],   # 6
 100   [ "off",                      $ERRORS{'CRITICAL'} ],  # 7
 101   [ "rebooting",                $ERRORS{'WARNING'} ],   # 8
 102   [ "switchedBypass",           $ERRORS{'WARNING'} ],   # 9
 103   [ "hardwareFailureBypass",    $ERRORS{'CRITICAL'} ],  # 10
 104   [ "sleepingUntilPowerReturn", $ERRORS{'CRITICAL'} ],  # 11
 105   [ "onSmartTrim",              $ERRORS{'WARNING'} ],   # 12
 106 );
 107
 108 my @failCauseVals = (
 109   undef,
 110   "noTransfer",
 111   "highLineVoltage",
 112   "brownout",
 113   "blackout",
 114   "smallMomentarySag",
 115   "deepMomentarySag",
 116   "smallMomentarySpike",
 117   "largeMomentarySpike",
 118   "selfTest",
 119   "rateOfVoltageChnage",
 120 );
 121
 122 my @battStatVals = (
 123   [ undef, undef ],                             # pad 0
 124   [ undef, undef ],                             # pad 1
 125   [ "batteryNormal",    $ERRORS{'OK'} ],        # 2
 126   [ "batteryLow",       $ERRORS{'CRITICAL'} ],  # 3
 127 );
 128
 129 my @battReplVals = (
 130   [ undef, undef ],                                     # pad 0
 131   [ "noBatteryNeedsReplacing",  $ERRORS{'OK'} ],        # 1
 132   [ "batteryNeedsReplacing",    $ERRORS{'CRITICAL'} ],  # 2
 133 );
 134
 135 my @diagnosticsResultsVals = (
 136   [ undef, undef ],                             # pad 0
 137   [ "OK",               $ERRORS{'OK'} ],        # 1
 138   [ "failed",           $ERRORS{'CRITICAL'} ],  # 2
 139   [ "invalidTest",      $ERRORS{'CRITICAL'} ],  # 3
 140   [ "testInProgress",   $ERRORS{'OK'} ],        # 4
 141 );
 142
 143 my $exitval     = $ERRORS{'UNKNOWN'};
 144 my $data;
 145 my $onbattery   = 3;
 146
 147 $data = get_snmp_int_val( $upsBasicOutputStatus );
 148
 149 print "Output status: ";
 150 if (defined ($data) && defined ($outputStatVals[$data][0])) {
 151   print "$outputStatVals[$data][0] | ";
 152   escalate_exitval($outputStatVals[$data][1]);
 153 } else {
 154   print "unknown | ";
 155 }
 156
 157 $data = get_snmp_int_val( $upsAdvBatteryRunTimeRemaining );
 158
 159 print "Rem time: ";
 160 if (defined ($data)) {
 161   my $hrs  = int($data / (60 * 60 * 100)); # Data is hundredths of a second
 162   my $mins = int($data / (60 * 100)) % 60;
 163   my $secs = ($data % 100) / 100;
 164   printf "%d:%02d:%05.2f | ", $hrs, $mins, $secs;
 165   if ($data <= $runtimecrit) {
 166     escalate_exitval($ERRORS{'CRITICAL'});
 167   } elsif ($data <= $runtimewarn) {
 168     escalate_exitval($ERRORS{'WARNING'});
 169   } else {
 170     escalate_exitval($ERRORS{'OK'});
 171   }
 172 } else {
 173   print "unknown | ";
 174 }
 175
 176 $data = get_snmp_int_val( $upsBasicBatteryStatus );
 177
 178 print "Battery status: ";
 179 if (defined ($data) && defined ($battStatVals[$data][0])) {
 180   my $failcause = "unknown";
 181   my $fc = get_snmp_int_val( $upsAdvInputLineFailCause );
 182   if ($data == $onbattery) {
 183     if (defined ($failCauseVals[$fc])) { $failcause = $failCauseVals[$fc]; }
 184     print "$battStatVals[$data][0] ($failcause) | ";
 185   } else {
 186     print "$battStatVals[$data][0] | ";
 187   }
 188   escalate_exitval($battStatVals[$data][1]);
 189 } else {
 190   print "unknown | ";
 191 }
 192
 193 $data = get_snmp_int_val( $upsAdvBatteryTemperature );
 194
 195 print "Battery temp(C): ";
 196 if (defined ($data)) {
 197   print "$data | ";
 198   if ($data >= $tempcrit) {
 199     escalate_exitval($ERRORS{'CRITICAL'});
 200   } elsif ($data >= $tempwarn) {
 201     escalate_exitval($ERRORS{'WARNING'});
 202   } else {
 203     escalate_exitval($ERRORS{'OK'});
 204   }
 205 } else {
 206   print "unknown | ";
 207 }
 208
 209 $data = get_snmp_int_val( $upsAdvBatteryReplaceIndicator );
 210
 211 print "Battery repl: ";
 212 if (defined ($data) && defined ($battReplVals[$data][0])) {
 213   print "$battReplVals[$data][0] | ";
 214   escalate_exitval($battReplVals[$data][1]);
 215 } else {
 216   print "unknown | ";
 217 }
 218
 219 $data = get_snmp_int_val( $upsAdvOutputLoad );
 220
 221 print "Output load (%): ";
 222 if (defined ($data)) {
 223   print "$data | ";
 224   if ($data >= $loadcrit) {
 225     escalate_exitval($ERRORS{'CRITICAL'});
 226   } elsif ($data >= $loadwarn) {
 227     escalate_exitval($ERRORS{'WARNING'});
 228   } else {
 229     escalate_exitval($ERRORS{'OK'});
 230   }
 231 } else {
 232   print "unknown | ";
 233 }
 234
 235 $data = get_snmp_int_val( $upsAdvTestDiagnosticsResults );
 236
 237 print "Diag result: ";
 238 if (defined ($data) && defined ($diagnosticsResultsVals[$data][0])) {
 239   print "$diagnosticsResultsVals[$data][0]\n";
 240   escalate_exitval($diagnosticsResultsVals[$data][1]);
 241 } else {
 242   print "unknown\n";
 243 }
 244
 245
 246 exit $exitval;
 247
 248
 249 sub print_usage () {
 250         print "Usage: $PROGNAME -H <host> -T temp -t temp -R minutes -r minutes\n";
 251         print "  -L percent -l percent\n";
 252 }
 253
 254 sub print_help () {
 255         print_revision($PROGNAME,'$Revision$');
 256         print "Copyright (c) 2001 Gerald Combs/Jeffrey Blank/Karl DeBisschop
 257
 258 This plugin reports the status of an APC UPS equipped with an SNMP management
 259 module.
 260
 261 ";
 262         print_usage();
 263         print "
 264 -H, --hostname=HOST
 265    Name or IP address of host to check
 266 -T --temp-critical
 267    Battery degrees C above which a CRITICAL status will result (default: 60)
 268 -t --temp-warning
 269    Battery degrees C above which a WARNING status will result (default: 40)
 270 -R --runtime-critical
 271    Minutes remaining below which a CRITICAL status will result (default: 30)
 272 -r --runtime-warning
 273    Minutes remaining below which a WARNING status will result (default: 60)
 274 -L --load-critical
 275    Output load pct above which a CRITICAL status will result (default: 85
 276 -l --load-warning
 277    Output load pct above which a WARNING status will result (default: 50
 278
 279 ";
 280         support();
 281 }
 282
 283 sub get_snmp_int_val ($) {
 284   my $val=0;
 285   my $oid = shift(@_);
 286
 287   $val = `/usr/bin/snmpget $host public $oid 2> /dev/null`;
 288   my @test = split(/ /,$val,3);
 289
 290   return undef unless (defined ($test[2]));
 291
 292   if ($test[2] =~ /\(\d+\)/) {  # Later versions of UCD SNMP
 293     ($val) = ($test[2] =~ /\((\d+)\)/);
 294   } elsif ($test[2] =~ /: \d+/) {
 295     ($val) = ($test[2] =~ /: (\d+)/);
 296   } else {
 297     $val = $test[2];
 298   }
 299
 300   return $val;
 301 }
 302
 303 sub escalate_exitval ($) {
 304   my $newval = shift(@_);
 305
 306   if ($newval > $exitval) { $exitval = $newval; }
 307 }