contrib/check_pcpmetric.py

   1 #! /usr/bin/env python
   2 #
   3 # Nagios client for checking Performance Co-Pilot metrics
   4 #
   5 #
   6
   7 from sys import argv,exit
   8 import popen2, getopt, string, types
   9
  10 DEBUG=0
  11
  12 nagios_pcpclient_version = 0.01
  13 PMVAL='/usr/bin/pmval'
  14 COMMANDLINE=PMVAL + " -s 1"
  15 METRIC='undefined'
  16 CRITICAL=0
  17 WARNING=0
  18
  19 def usage():
  20         print "Usage:", argv[0], "[options]"
  21         print "Options:"
  22         print "\t[-H host]\tHostname to contact"
  23         print "\t[-m metric]\tPCP metric to check"
  24         print "\t[-i instance]\tPCP metric instance"
  25         print "\t[-w warn]\tIssue warning alert if value is larger than this"
  26         print "\t[-c critical]\tIssue critical alert value is larger than this"
  27         print "\t[-V]\t\tProgram version"
  28         print "\t[-h]\t\tThis helptext"
  29         print ""
  30         print "F.ex. to check 5 minute loadaverage, warn if the load is above 2,"
  31         print "and give critical warning if it's above 10:"
  32         print "\n\t%", argv[0], " -i 5 -m kernel.all.load -w 2 -c 10"
  33         print ""
  34         print "A list of all PCP metrics can be found with the command 'pminfo'."
  35         print "A list of all instances within a metric can be found with 'pminfo -f metric'."
  36         print "F.ex. to see all available instances of 'filesys.full' execute:"
  37         print "\n\t% pminfo -f filesys.full"
  38         print "\tfilesys.full"
  39         print """\t\tinst [0 or "/dev/root"] value 45.35514044640914"""
  40         print """\t\tinst [1 or "/dev/sda1"] value 46.74285959344712"""
  41         print """\t\tinst [2 or "/dev/sdb1"] value 0.807766570678168"""
  42         print ""
  43         print "And the command to have nagios monitor the /dev/sda1 filesystem would be:"
  44         print "\n\t", argv[0], " -i /dev/sda1 -m filesys.full -w 70 -c 90"
  45
  46
  47 opts, args = getopt.getopt(argv[1:],'hH:c:w:m:i:V')
  48 for opt in opts:
  49         key,value = opt
  50         if key == '-H':
  51                 COMMANDLINE = COMMANDLINE + " -h " + value
  52         elif key == '-m':
  53                 METRIC=value
  54         elif key == '-i':
  55                 COMMANDLINE = COMMANDLINE + " -i " + value
  56         elif key == '-c':
  57                 CRITICAL = value
  58         elif key == '-w':
  59                 WARNING = value
  60         elif key == '-h':
  61                 usage()
  62                 exit(0)
  63         elif key == '-V':
  64                 print "Nagios Performance CoPilot client v%.2f" % nagios_pcpclient_version
  65                 print "Written by Jan-Frode Myklebust <janfrode@parallab.uib.no>"
  66                 exit(0)
  67
  68 if METRIC == 'undefined':
  69         usage()
  70         exit(3)
  71
  72 COMMANDLINE = COMMANDLINE + " " + METRIC
  73 if DEBUG: print COMMANDLINE
  74 p=popen2.Popen4(COMMANDLINE)
  75 exitcode=p.wait()
  76
  77 # Get the last line of output from 'pmval':
  78 buffer = p.fromchild.readline()
  79 while (buffer != ''):
  80         output=buffer
  81         buffer = p.fromchild.readline()
  82
  83 returndata = string.split(output)[0]
  84
  85
  86 # Confirm that we have gotten a float, and not
  87 # some errormessage in the returndata. If not,
  88 # print the error, and give the UNKNOWN exit code:
  89
  90 try:
  91         retval = string.atof(returndata)
  92 except ValueError, e:
  93         print e
  94         exit(3)
  95
  96 if (retval < WARNING):
  97         EXITCODE=0
  98 elif (retval > CRITICAL):
  99         EXITCODE=2
 100 elif (retval > WARNING):
 101         EXITCODE=1
 102 else:
 103         EXITCODE=3
 104
 105 print retval
 106 exit(EXITCODE)