comon.py

   1 #
   2 # Copyright (c) 2004  The Trustees of Princeton University (Trustees).
   3 #
   4 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
   5 #
   6 # $Id: $
   7 #
   8 # Get CoMon data, unsorted, in CSV, and create a huge hash.
   9 #
  10
  11
  12 import urllib2
  13 import httplib
  14 import time
  15 import Queue
  16 import logging
  17 from threading import *
  18 #httplib.HTTPConnection.debuglevel = 1
  19
  20 logger = logging.getLogger("monitor")
  21
  22 # Time between comon refresh
  23 COSLEEP=1200
  24
  25 # CoMon
  26 COMONURL = "http://summer.cs.princeton.edu/status/tabulator.cgi?table=table_nodeview"
  27
  28
  29 class Comon(Thread):
  30         """
  31         cdb is the comon database (dictionary)
  32         all buckets is a queue of all problem nodes. This gets sent to rt to find
  33         tickets open for host.
  34         """
  35         def __init__(self, cdb, allbuckets):
  36                 self.codata = cdb
  37                 self.updated = time.time()
  38                 self.allbuckets = allbuckets
  39                 self.comonbkts = {"down" : "resptime%20==%200%20&&%20keyok==null",
  40                         "ssh": "sshstatus%20%3E%202h",
  41                         "clock_drift": "drift%20%3E%201m",
  42                         "dns": "dns1udp%20%3E%2080%20&&%20dns2udp%20%3E%2080",
  43                         "disk": "resptime%20%3E%200%20&&%20gbfree%20%3C%205",
  44                         "filerw": "filerw%3E0",
  45                         "dbg" : "keyok==0"}
  46                 Thread.__init__(self)
  47
  48         def __tohash(self,rawdata):
  49                 # First line Comon returns is list of keys with respect to index
  50                 keys = rawdata.readline().rstrip().split(", ")
  51                 host = []
  52                 hash = {}
  53                 try:
  54                         for line in rawdata.readlines():
  55                                 host = line.rstrip().split(", ")
  56                                 tmp = {}
  57                                 for i in range(1,len(keys)):
  58                                         tmp[keys[i]]=host[i]
  59                                 hash[host[0]]=tmp
  60                         logger.debug("Retrieved %s hosts" % len(hash.keys()))
  61                 except Exception, err:
  62                         logger.debug("No hosts retrieved")
  63                         return {}
  64                 return hash
  65
  66         # Update individual buckekts.  Hostnames only.
  67         def updatebkts(self):
  68                 for (bkt,url) in self.comonbkts.items():
  69                         tmp = self.coget(COMONURL + "&format=formatcsv&select='" + url + "'").keys()
  70                         setattr(self, bkt, tmp)
  71
  72         # Update ALL node information
  73         def updatedb(self):
  74                 # Get time of update
  75                 self.updated = time.time()
  76                 # Make a Hash, put in self.
  77                 self.codata.update(self.coget(COMONURL + "&format=formatcsv"))
  78
  79         def coget(self,url):
  80                 rawdata = None
  81                 try:
  82                         logger.debug("Trying - " + url)
  83                         coserv = urllib2.Request(url)
  84                         coserv.add_header('User-Agent',
  85                                 'PL_Monitor +http://monitor.planet-lab.org/')
  86                         opener = urllib2.build_opener()
  87                         # Initial web get from summer.cs in CSV
  88                         rawdata = opener.open(coserv)
  89                 except urllib2.URLError, (err):
  90                         print "Attempting %s" %COMONURL
  91                         print "URL error (%s)" % (err)
  92                         rawdata = None
  93                 return self.__tohash(rawdata)
  94
  95         # Push nodes that are bad (in *a* bucket) into q(allbuckets)
  96         def push(self):
  97                 for bucket in self.comonbkts.keys():
  98                         for host in getattr(self,bucket):
  99                                 self.allbuckets.put(host)
 100
 101         def run(self):
 102                 while 1:
 103                         self.updatedb()
 104                         self.updatebkts()
 105                         self.push()
 106                         time.sleep(COSLEEP)
 107
 108         def __repr__(self):
 109             return self
 110
 111 def main():
 112         logger.setLevel(logging.DEBUG)
 113         ch = logging.StreamHandler()
 114         ch.setLevel(logging.DEBUG)
 115         formatter = logging.Formatter('%(message)s')
 116         ch.setFormatter(formatter)
 117         logger.addHandler(ch)
 118
 119
 120         t = Queue.Queue()
 121         cdb = {}
 122         a = Comon(cdb,t)
 123         print a.comonbkts
 124         a.start()
 125
 126         time.sleep(5)
 127         print a.down
 128
 129         time.sleep(5)
 130         #print cdb
 131         for host in cdb.keys():
 132                 if cdb[host]['keyok'] == "0":
 133                         print("%s \t Bootstate %s nodetype %s kernver %s keyok %s" %(host, cdb[host]['bootstate'], cdb[host]['nodetype'], cdb[host]['kernver'], cdb[host]['keyok']))
 134         #time.sleep(3)
 135         #a.push()
 136         #print a.filerw
 137         #print a.coget(COMONURL + "&format=formatcsv&select='" + a.comonbkts['filerw'])
 138
 139         os._exit(0)
 140 if __name__ == '__main__':
 141         import os
 142         try:
 143                 main()
 144         except KeyboardInterrupt:
 145                 print "Killed.  Exitting."
 146                 logger.info('Monitor Killed')
 147                 os._exit(0)