diagnose.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: diagnose.py,v 1.1 2007/08/08 13:36:46 soltesz Exp $
   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 from config import config
  19 from optparse import OptionParser
  20 parser = OptionParser()
  21
  22 parser.set_defaults(nodelist=None,
  23                                         refresh=False,
  24                                         cachert=False,
  25                                         cachenodes=False,
  26                                         blacklist=None,
  27                                         ticketlist=None)
  28
  29 parser.add_option("", "--nodelist", dest="nodelist", metavar="filename",
  30                                         help="Read nodes to act on from specified file")
  31 parser.add_option("", "--refresh", action="store_true", dest="refresh",
  32                                         help="Refresh the cached values")
  33 parser.add_option("", "--cachert", action="store_true", dest="cachert",
  34                                         help="Cache the RT database query")
  35 parser.add_option("", "--cachenodes", action="store_true", dest="cachenodes",
  36                                         help="Cache node lookup from PLC")
  37 parser.add_option("", "--ticketlist", dest="ticketlist",
  38                                         help="Whitelist all RT tickets in this file")
  39 parser.add_option("", "--blacklist", dest="blacklist",
  40                                         help="Blacklist all nodes in this file")
  41
  42 config = config(parser)
  43 config.parse_args()
  44
  45 # daemonize and *pid
  46 #from util.process import *
  47
  48 # RT tickets
  49 import rt
  50 # Correlates input with policy to form actions
  51 import policy
  52 import moncommands
  53 import database
  54 import plc
  55 import syncplcdb
  56
  57 # Log to what
  58 LOG="./monitor.log"
  59
  60 # Time to refresh DB and remove unused entries
  61 RTSLEEP=7200 #2hrs
  62 # Time between policy enforce/update
  63 #POLSLEEP=43200 #12hrs
  64 POLSLEEP=10
  65
  66 # Global list of all running threads.  Any threads added to
  67 # list will be monitored.
  68 runningthreads = {}
  69 # Seconds between checking threads
  70 WATCHSLEEP = 5
  71
  72 # Set up Logging
  73 logger = logging.getLogger("monitor")
  74 logger.setLevel(logging.DEBUG)
  75 fh = logging.FileHandler(LOG, mode = 'a')
  76 fh.setLevel(logging.DEBUG)
  77 formatter = logging.Formatter('%(asctime)s %(levelname)s %(message)s')
  78 fh.setFormatter(formatter)
  79 logger.addHandler(fh)
  80
  81
  82 """
  83 Launches threads and adds them to the runningthreads global list.
  84 Assigns name for thread, starts.
  85 """
  86 def startThread(fnct, name):
  87                 runningthreads[name] = fnct
  88                 runningthreads[name].setName(name)
  89                 try:
  90                         logger.info("Starting thread " + name)
  91                         runningthreads[name].start()
  92                 except Exception, err:
  93                         logger.error("Thread: " + name + " " + error)
  94
  95
  96 """
  97 Watches threads and catches exceptions.  Each launched thread is
  98 watched and state is logged.
  99 """
 100 class ThreadWatcher(Thread):
 101         def __init__(self):
 102                 Thread.__init__(self)
 103
 104         def run(self):
 105                 while 1:
 106                         self.checkThreads()
 107                         time.sleep(WATCHSLEEP)
 108
 109         def checkThreads(self):
 110                 # Iterate through treads, compare with last running.
 111                 for thread in runningthreads.keys():
 112                         # If thread found dead, remove from queue
 113                         #print "found %s" % thread
 114                         if not runningthreads[thread].isAlive():
 115                                 logger.error("***********Thread died: %s**********" %(thread))
 116                                 del runningthreads[thread]
 117                 return len(runningthreads.keys())
 118
 119
 120 class Dummy(Thread):
 121         def __init__(self):
 122                 Thread.__init__(self)
 123
 124         def run(self):
 125                 time.sleep(5)
 126
 127 def dict_from_nodelist(nl):
 128         d = {}
 129         for host in nl:
 130                 h = host['hostname']
 131                 d[h] = host
 132         return d
 133
 134
 135
 136 """
 137 Start threads, do some housekeeping, then daemonize.
 138 """
 139 def main():
 140         # Defaults
 141         global status, logger
 142         global config
 143
 144         logger.info('Diagnose Started')
 145         print 'Diagnose Started'
 146
 147         ##########  VARIABLES   ########################################
 148         # Queue between Merge and RT
 149         toRT = Queue.Queue()
 150
 151         # Queue between RT and Diagnose
 152         fromRT = Queue.Queue()
 153
 154         #########  GET NODES    ########################################
 155         logger.info('Get Nodes from PLC')
 156         print "getnode from plc: %s %s %s" % (config.debug, config.cachenodes, config.refresh)
 157         l_plcnodes = database.if_cached_else_refresh(config.cachenodes,
 158                                                                 config.refresh, "l_plcnodes",
 159                                                                 lambda : syncplcdb.create_plcdb() )
 160
 161         s_plcnodenames = Set([x['hostname'] for x in l_plcnodes])
 162
 163         # List of nodes from a user-provided file.
 164         if config.nodelist:
 165                 file = config.nodelist
 166                 nodelist = config.getListFromFile(file)
 167
 168                 s_usernodes = Set(nodelist)
 169                 # SAFE nodes are in PLC and the list
 170                 s_safe_usernodes   = s_plcnodenames & s_usernodes
 171                 # UNSAFE nodes are list but not in PLC. i.e. do not monitor.
 172                 s_unsafe_usernodes = s_usernodes - s_plcnodenames
 173                 if len(s_unsafe_usernodes) > 0 :
 174                         for node in s_unsafe_usernodes:
 175                                 print "WARNING: User provided: %s but not found in PLC" % node
 176
 177                 l_nodes = filter(lambda x: x['hostname'] in s_safe_usernodes,l_plcnodes)
 178         else:
 179                 l_nodes = l_plcnodes
 180
 181         print "len of l_nodes: %d" % len(l_nodes)
 182         # Minus blacklisted ones..
 183         l_blacklist = database.if_cached_else(1, "l_blacklist", lambda : [])
 184         l_ticket_blacklist = database.if_cached_else(1,"l_ticket_blacklist",lambda : [])
 185         l_nodes  = filter(lambda x : not x['hostname'] in l_blacklist, l_nodes)
 186
 187         logger.info('Get Tickets from RT')
 188         #######  RT tickets    #########################################
 189         t = moncommands.MyTimer()
 190         ad_dbTickets = database.if_cached_else_refresh(config.cachert, config.refresh, "ad_dbTickets", rt.rt_tickets)
 191         if ad_dbTickets == "":
 192                 print "ad_dbTickets failed..."
 193                 sys.exit(1)
 194         print "Getting tickets from RT took: %f sec" % t.diff() ; del t
 195
 196         logger.info('Start Merge/RT/Diagnose threads')
 197         ####### Merge
 198         # only look at nodes in l_nodes
 199         merge = policy.Merge( [node['hostname'] for node in l_nodes], toRT)
 200         startThread(merge,"merge")
 201         ####### RT
 202         rt1   = rt.RT(ad_dbTickets, toRT, fromRT, l_ticket_blacklist)
 203         startThread(rt1,"rt1")
 204         ####### Diagnose
 205         diagnose = policy.Diagnose(fromRT)
 206         startThread(diagnose,"diagnose")
 207
 208
 209         tw = ThreadWatcher()
 210         while True:
 211                 if tw.checkThreads() == 0:
 212                         break
 213                 print "waiting... %s" % time.time()
 214                 time.sleep(WATCHSLEEP)
 215
 216         logger.info('Diagnose Exitting')
 217         sys.exit(0)
 218
 219 if __name__ == '__main__':
 220         try:
 221                 main()
 222         except KeyboardInterrupt:
 223                 print "Killed.  Exitting."
 224                 logger.info('Diagnose Killed')
 225                 sys.exit(0)