libexec/check_shinken.py

   1 #!/usr/bin/env python
   2 ################################################
   3 #check_shinken.py
   4 ################################################
   5
   6 from optparse import OptionParser
   7 try:
   8     import shinken.pyro_wrapper as pyro
   9     from shinken.pyro_wrapper import Pyro
  10 except ImportError:
  11     print 'CRITICAL : check_shinken requires the Python Pyro and the shinken.pyro_wrapper module. Please install it.'
  12     raise SystemExit, CRITICAL
  13
  14 # Exit statuses recognized by Nagios and thus by Shinken
  15 UNKNOWN = -1
  16 OK = 0
  17 WARNING = 1
  18 CRITICAL = 2
  19 #Name of the Pyro Object we are searching
  20 PYRO_OBJECT = 'ForArbiter'
  21 daemon_types = ['arbiter', 'broker', 'scheduler', 'poller', 'reactionner']
  22
  23 #The following functions just count the number of deamons with the associated parameters
  24 def count_total(result):
  25     num = 0
  26     for cur in result:
  27         num += 1
  28     return num
  29
  30 def count_alive(result):
  31     num = 0
  32     for cur in result:
  33         if result[cur]['alive'] == True:
  34             num += 1
  35     return num
  36
  37 def count_total_spare(result):
  38     num = 0
  39     for cur in result:
  40         if result[cur]['spare'] == True:
  41             num += 1
  42     return num
  43
  44 def count_alive_spare(result):
  45     num = 0
  46     for cur in result:
  47         if (result[cur]['spare'] == True and result[cur]['alive'] == True):
  48             num += 1
  49     return num
  50
  51 #this function prints the names of the daemons fallen
  52 def find_dead(result):
  53     dead_ones = ''
  54     for cur in result:
  55         if (result[cur]['alive'] == False):
  56             dead_ones = dead_ones+' '+cur
  57     return dead_ones
  58
  59 def check_deamons_numbers(result, target):
  60     total_number = count_total(result)
  61     alive_number = count_alive(result)
  62     total_spare_number = count_total_spare(result)
  63     alive_spare_number = count_alive_spare(result)
  64     dead_number = total_number - alive_number
  65     dead_list = find_dead(result)
  66     #TODO : perfdata to graph deamons would be nice (in big HA architectures)
  67     #if alive_number <= critical, then we have a big problem
  68     if alive_number <= options.critical:
  69         print "CRITICAL - %d/%d %s(s) UP" % (alive_number, total_number, target)
  70         raise SystemExit, CRITICAL
  71     #We are not in a case where there is no more daemons, but are there daemons down?
  72     elif dead_number >= options.warning:
  73         print "WARNING - %d/%d %s(s) DOWN :%s" % (dead_number, total_number, target, dead_list)
  74         raise SystemExit, WARNING
  75         #Everything seems fine. But that's no surprise, is it?
  76     else :
  77         print "OK - %d/%d %s(s) UP, with %d/%d spare(s) UP" % (alive_number, total_number, target, alive_spare_number, total_spare_number)
  78         raise SystemExit, OK
  79
  80 # Adding options. None are required, check_shinken will use shinken defaults
  81 #TODO : Add more control in args problem and usage than the default OptionParser one
  82 parser = OptionParser()
  83 parser.add_option('-a', '--hostname', dest='hostname', default='127.0.0.1')
  84 parser.add_option('-p', '--portnumber', dest='portnum', default=7770)
  85 parser.add_option('-s', '--ssl', dest='ssl', default=False)
  86 #TODO : Add a list of correct values for target and don't authorize anything else
  87 parser.add_option('-t', '--target', dest='target')
  88 parser.add_option('-d', '--daemonname', dest='daemon', default='')
  89 #In HA architectures, a warning should be displayed if there's one daemon down
  90 parser.add_option('-w','--warning', dest='warning', default = 1)
  91 #If no deamon is left, display a critical (but shinken will be probably dead already)
  92 parser.add_option('-c', '--critical', dest='critical', default = 0)
  93
  94 #Retrieving options
  95 options, args = parser.parse_args()
  96 #TODO : for now, helpme doesn't work as desired
  97 options.helpme = False
  98
  99 # Check for required option target
 100 if not getattr(options, 'target'):
 101     print 'CRITICAL - target is not specified; You must specify which daemons you want to check!'
 102     raise SystemExit, CRITICAL
 103 elif options.target not in daemon_types:
 104     print 'CRITICAL - target %s is not a Shinken daemon!' % options.target
 105     raise SystemExit, CRITICAL
 106
 107 uri = pyro.create_uri(options.hostname, options.portnum, PYRO_OBJECT , options.ssl)
 108 if options.daemon:
 109     #We just want a check for a single satellite daemon
 110     #Only OK or CRITICAL here
 111     daemon_name = options.daemon
 112     result = Pyro.core.getProxyForURI(uri).get_satellite_status(options.target, daemon_name)
 113     if result:
 114         if result['alive']:
 115             print 'OK - %s alive' % daemon_name
 116             raise SystemExit, OK
 117         else:
 118             print 'CRITICAL - %s down' % daemon_name
 119             raise SystemExit, CRITICAL
 120     else:
 121         print 'UNKNOWN - %s status could not be retrieved' % daemon_name
 122         raise SystemExit, UNKNOWN
 123 else:
 124     #If no daemonname is specified, we want a general overview of the "target" daemons
 125     result = {}
 126     daemon_list = Pyro.core.getProxyForURI(uri).get_satellite_list(options.target)
 127     for daemon_name in daemon_list:
 128         #Getting individual daemon and putting status info in the result dictionnary
 129         result[daemon_name] = Pyro.core.getProxyForURI(uri).get_satellite_status(options.target, daemon_name)
 130     #Now we have all data
 131     if result:
 132         check_deamons_numbers(result, options.target)
 133     else :
 134         print 'UNKNOWN - Arbiter could not retrieve status for %s' % options.target
 135         raise SystemExit, UNKNOWN