nagios/configurator/comon_query.py

   1 #!/usr/bin/env python
   2
   3 #
   4 # This module checks for a planetlab node by
   5 # (*) connecting to the comon central query interface
   6 # (*) retrieving the latest info for a given node
   7 #
   8
   9 import sys
  10 import socket
  11 import re
  12 import string
  13 import urllib2
  14
  15 import nagios
  16
  17 # default server
  18 SERVER='summer.cs.princeton.edu'
  19
  20 NodeUnknownException="NodeUnknownByComon"
  21
  22 ####################
  23 def check (node,server=None):
  24
  25     if (server == None):
  26         server = SERVER
  27
  28     try:
  29
  30         # compute node IP number
  31         node_address = compute_address (node)
  32 #        print 'int address',node_address
  33         # build URL, connect and return attributes dict
  34         node_dict = query_node (server,node_address)
  35         for key in node_dict.keys():
  36             print key,node_dict[key]
  37         # interpret
  38         return  interpret (node_dict)
  39
  40     except NodeUnknownException:
  41         return nagios.UNKNOWN
  42     except Exception,e:
  43         print "comon_query.check got exception",e
  44         return nagios.UNKNOWN
  45
  46 ##########
  47 re_dec="([0-9]{1,3})"
  48 re_ipsep="\."
  49 re_ip=(re_dec+re_ipsep)*3+re_dec
  50 ma_ip = re.compile (re_ip)
  51
  52 def compute_address (nodename):
  53     ip=socket.gethostbyname(nodename)
  54     ints=map(int,ma_ip.match(ip).groups())
  55     res=ints[0]
  56     for i in range(1,4):
  57         res = (res*256)+ints[i]
  58     return res
  59
  60 ##########
  61 URL_FORMAT="http://%s/status/tabulator.cgi"
  62 ARGS_FORMAT="table=table_nodeviewshort&select='%s'"
  63 FILTER_FORMAT="address==%d"
  64 CSV_FORMAT="&format=formatcsv"
  65
  66 def filter_address (address):
  67     return FILTER_FORMAT%address
  68 def filter_node (nodename):
  69     return filter_address(compute_address(nodename))
  70
  71 def full_url (server, filter):
  72     return (URL_FORMAT%server
  73             + '?'
  74             + ARGS_FORMAT%filter)
  75
  76 def full_url_csv (server,filter):
  77     return full_url(server,filter)+CSV_FORMAT
  78
  79
  80 # mention field here means we'll parse it and keep it
  81 # see store_dict below
  82 FIELDS_FOCUS={
  83     'resptime': 'float',
  84     'sshstatus':'int',
  85     'bootstate':'string',
  86     }
  87
  88 NOTHING_MATCHED='nothing matched select statement'
  89 ##########
  90 def query_node (server,address):
  91     filter=filter_address(address)
  92     full_url=full_url_csv(server,filter)
  93     req = urllib2.urlopen(full_url)
  94     # let's parse this manually
  95     headers=map(string.strip,req.readline().split(','))
  96     # handle the case where the node is unknown to comon
  97     values=req.readline()
  98     if values.find(NOTHING_MATCHED) != -1:
  99         raise NodeUnknownException
 100
 101     values=values.split(',')
 102
 103     print 'h',headers
 104     print 'v', values
 105
 106     dict={}
 107 #    store_dict=lambda key,val: dict[key]=val
 108 #    -> SyntaxError: can't assign to lambda
 109     def store_dict (key,val):
 110         if FIELDS_FOCUS.has_key(key):
 111             format = FIELDS_FOCUS[key]
 112             if format == 'float':
 113                 dict[key]=float(val)
 114             elif format == 'int':
 115                 dict[key]=int(val)
 116             elif format == 'string':
 117                 dict[key]=string.strip(val)
 118     map (store_dict, headers, values)
 119
 120     return dict
 121
 122 ##########
 123 # function for decomposing a number along units
 124 # tuple [n] must be a multiple of tuple[n+1]
 125 # e.g.
 126 #   1223456, (10000, 500, 10)  -> (122, 68, 5)
 127 # coz 1223456 = 122*10000 + 68*500 + 5*10 + 6
 128 def split_number (n,tuple):
 129     result=()
 130     for i in range(0,len(tuple)):
 131         base=tuple[i]
 132         result+=(n/base,)
 133         n=n%base
 134     return result
 135
 136 ###
 137 MINUTE=60
 138 HOUR=60*MINUTE
 139 DAY=HOUR*24
 140 WEEK=DAY*7
 141 MONTH=DAY*30
 142
 143 # from a delay in seconds, returns a human-readable string
 144 def seconds_printable (seconds):
 145     month,week,day,hour,minute = split_number(seconds,
 146                                               (MONTH,WEEK,DAY,HOUR,MINUTE))
 147     if month != 0:
 148         return "%d month(s), %d weeks, %d day(s)"%(month,week,day)
 149     elif week !=0:
 150         return "%d weeks, %d day(s), %d hour(s)"%(week,day,hour)
 151     elif day != 0:
 152         return "%d day(s), %d hour(s)"%(day,hour)
 153     elif hour != 0:
 154         return "%d hour(s), %d minute(s)"%(hour,minute)
 155     else:
 156         return "%d minute(s)"%(minute)
 157
 158 ##########
 159 def interpret (dict):
 160     # check sshstatus is null
 161     sshstatus=dict['sshstatus']
 162     if sshstatus != 0:
 163         print 'No response to comon/ssh for %s'%seconds_printable(sshstatus)
 164         if sshstatus >= 10*MINUTE:
 165             return nagios.KO
 166         else:
 167             return nagios.WARNING
 168     else:
 169         # let's focus on resptime
 170         resptime = dict['resptime']
 171         print "Response time as measured by comon = %.2f s"%resptime
 172         if resptime >= 10.0:
 173             return nagios.KO
 174         elif resptime >= 5.0:
 175             return nagios.WARNING
 176         else:
 177             return nagios.OK
 178
 179 #################### quick test
 180 def usage():
 181     print "Usage comon_query.py node"
 182     sys.exit(1)
 183
 184 if __name__=='__main__':
 185     if len(sys.argv) != 2:
 186         usage()
 187     print 'comon_query.check would return %d'%check(sys.argv[1])
 188 #     print 'get >%s<',seconds_printable(int(sys.argv[1]))