action.py

   1 #!/usr/bin/python
   2 #
   3 # Copyright (c) 2004  The Trustees of Princeton University (Trustees).
   4 #
   5 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
   6 # Stephen Soltesz <soltesz@cs.princeton.edu>
   7 #
   8 # $Id$
   9
  10 import sys
  11 from threading import *
  12 import time
  13 import logging
  14 import Queue
  15 from sets import Set
  16
  17 # Global config options
  18 import parser as parsermodule
  19
  20 parser = parsermodule.getParser()
  21
  22 parser.set_defaults(nodelist=None,
  23                                         cachert=False,
  24                                         cachenodes=False,
  25                                         blacklist=None,
  26                                         ticketlist=None)
  27
  28 parser.add_option("", "--nodelist", dest="nodelist",
  29                                         help="Read nodes to act on from specified file")
  30 parser.add_option("", "--cachert", action="store_true",
  31                                         help="Cache the RT database query")
  32 parser.add_option("", "--cachenodes", action="store_true",
  33                                         help="Cache node lookup from PLC")
  34 parser.add_option("", "--ticketlist", dest="ticketlist",
  35                                         help="Whitelist all RT tickets in this file")
  36 parser.add_option("", "--blacklist", dest="blacklist",
  37                                         help="Blacklist all nodes in this file")
  38
  39 config = parsermodule.parse_args(parser)
  40
  41 # daemonize and *pid
  42 #from util.process import *
  43
  44 # RT tickets
  45 import rt
  46 # Correlates input with policy to form actions
  47 import policy
  48 import database
  49 import plc
  50
  51 # Log to what
  52 LOG="./monitor.log"
  53
  54 # Time to refresh DB and remove unused entries
  55 RTSLEEP=7200 #2hrs
  56 # Time between policy enforce/update
  57 #POLSLEEP=43200 #12hrs
  58 POLSLEEP=10
  59
  60 # Global list of all running threads.  Any threads added to
  61 # list will be monitored.
  62 runningthreads = {}
  63 # Seconds between checking threads
  64 WATCHSLEEP = 10
  65
  66 # Set up Logging
  67 logger = logging.getLogger("monitor")
  68 logger.setLevel(logging.DEBUG)
  69 fh = logging.FileHandler(LOG, mode = 'a')
  70 fh.setLevel(logging.DEBUG)
  71 formatter = logging.Formatter('%(asctime)s %(levelname)s %(message)s')
  72 fh.setFormatter(formatter)
  73 logger.addHandler(fh)
  74
  75
  76 """
  77 Launches threads and adds them to the runningthreads global list.
  78 Assigns name for thread, starts.
  79 """
  80 def startThread(fnct, name):
  81                 runningthreads[name] = fnct
  82                 runningthreads[name].setName(name)
  83                 try:
  84                         logger.info("Starting thread " + name)
  85                         runningthreads[name].start()
  86                 except Exception, err:
  87                         logger.error("Thread: " + name + " " + error)
  88
  89
  90 """
  91 Watches threads and catches exceptions.  Each launched thread is
  92 watched and state is logged.
  93 """
  94 class ThreadWatcher(Thread):
  95         def __init__(self):
  96                 Thread.__init__(self)
  97
  98         def run(self):
  99                 while 1:
 100                         self.checkThreads()
 101                         time.sleep(WATCHSLEEP)
 102
 103         def checkThreads(self):
 104                 # Iterate through treads, compare with last running.
 105                 for thread in runningthreads.keys():
 106                         # If thread found dead, remove from queue
 107                         #print "found %s" % thread
 108                         if not runningthreads[thread].isAlive():
 109                                 logger.error("***********Thread died: %s**********" %(thread))
 110                                 del runningthreads[thread]
 111                 return len(runningthreads.keys())
 112
 113
 114 class Dummy(Thread):
 115         def __init__(self):
 116                 Thread.__init__(self)
 117
 118         def run(self):
 119                 time.sleep(5)
 120
 121 def dict_from_nodelist(nl):
 122         d = {}
 123         for host in nl:
 124                 h = host['hostname']
 125                 d[h] = host
 126         return d
 127
 128 """
 129 Start threads, do some housekeeping, then daemonize.
 130 """
 131 def main():
 132         # Defaults
 133         global status, logger
 134         global config
 135
 136         logger.info('Action Started')
 137         print 'Action Started'
 138
 139         #########  GET NODES    ########################################
 140         logger.info('Get Nodes from PLC')
 141         print "getnode from plc"
 142         l_plcnodes = database.if_cached_else(True,
 143                                                                 "l_plcnodes",
 144                                                                 lambda : plc.getNodes({'peer_id':None}))
 145
 146         s_plcnodenames = Set([x['hostname'] for x in l_plcnodes])
 147
 148         # List of nodes from a user-provided file.
 149         if config.nodelist:
 150                 file = config.nodelist
 151                 nodelist = config.getListFromFile(file)
 152                 #for node in nodelist:
 153                 #       print "%s" % node
 154
 155                 s_usernodes = Set(nodelist)
 156                 # SAFE nodes are in PLC and the list
 157                 s_safe_usernodes   = s_plcnodenames & s_usernodes
 158                 # UNSAFE nodes are in list but not in PLC. i.e. ignore them.
 159                 s_unsafe_usernodes = s_usernodes - s_plcnodenames
 160                 if len(s_unsafe_usernodes) > 0 :
 161                         for node in s_unsafe_usernodes:
 162                                 print "WARNING: User provided: %s but not found in PLC" % node
 163
 164                 l_nodes = filter(lambda x: x['hostname'] in s_safe_usernodes,l_plcnodes)
 165         else:
 166                 l_nodes = l_plcnodes
 167
 168         print "len of l_nodes: %d" % len(l_nodes)
 169         # Minus blacklisted ones..
 170         l_ticket_blacklist = database.if_cached_else(1,"l_ticket_blacklist",lambda : [])
 171
 172         l_blacklist = database.if_cached_else(1, "l_blacklist", lambda : [])
 173         l_nodes  = filter(lambda x : not x['hostname'] in l_blacklist, l_nodes)
 174
 175         #######  Get RT tickets    #########################################
 176         #logger.info('Get Tickets from RT')
 177         #t = commands.MyTimer()
 178         #ad_dbTickets = database.if_cached_else(config.cachert, "ad_dbTickets", rt.rt_tickets)
 179         #print "Getting tickets from RT took: %f sec" % t.diff() ; del t
 180
 181         logger.info('Start Action thread')
 182         ####### Action
 183         action = policy.Action( [node['hostname'] for node in l_nodes] )
 184         startThread(action,"action")
 185
 186
 187         tw = ThreadWatcher()
 188         while True:
 189                 if tw.checkThreads() == 0:
 190                         break
 191                 time.sleep(WATCHSLEEP)
 192
 193         logger.info('Action Exitting')
 194         sys.exit(0)
 195
 196 if __name__ == '__main__':
 197         try:
 198                 main()
 199         except KeyboardInterrupt:
 200                 print "Killed.  Exitting."
 201                 logger.info('Action Killed')
 202                 sys.exit(0)