3 # Copyright (c) 2004 The Trustees of Princeton University (Trustees).
5 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
6 # Stephen Soltesz <soltesz@cs.princeton.edu>
8 # $Id: diagnose.py,v 1.1 2007/08/08 13:36:46 soltesz Exp $
11 from threading import *
17 # Global config options
18 from config import config
19 from optparse import OptionParser
20 parser = OptionParser()
22 parser.set_defaults(nodelist=None,
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")
42 config = config(parser)
46 #from util.process import *
50 # Correlates input with policy to form actions
60 # Time to refresh DB and remove unused entries
62 # Time between policy enforce/update
63 #POLSLEEP=43200 #12hrs
66 # Global list of all running threads. Any threads added to
67 # list will be monitored.
69 # Seconds between checking threads
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)
83 Launches threads and adds them to the runningthreads global list.
84 Assigns name for thread, starts.
86 def startThread(fnct, name):
87 runningthreads[name] = fnct
88 runningthreads[name].setName(name)
90 logger.info("Starting thread " + name)
91 runningthreads[name].start()
92 except Exception, err:
93 logger.error("Thread: " + name + " " + error)
97 Watches threads and catches exceptions. Each launched thread is
98 watched and state is logged.
100 class ThreadWatcher(Thread):
102 Thread.__init__(self)
107 time.sleep(WATCHSLEEP)
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())
122 Thread.__init__(self)
127 def dict_from_nodelist(nl):
137 Start threads, do some housekeeping, then daemonize.
141 global status, logger
144 logger.info('Diagnose Started')
145 print 'Diagnose Started'
147 ########## VARIABLES ########################################
148 # Queue between Merge and RT
151 # Queue between RT and Diagnose
152 fromRT = Queue.Queue()
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() )
161 s_plcnodenames = Set([x['hostname'] for x in l_plcnodes])
163 # List of nodes from a user-provided file.
165 file = config.nodelist
166 nodelist = config.getListFromFile(file)
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
177 l_nodes = filter(lambda x: x['hostname'] in s_safe_usernodes,l_plcnodes)
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)
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..."
194 print "Getting tickets from RT took: %f sec" % t.diff() ; del t
196 logger.info('Start Merge/RT/Diagnose threads')
198 # only look at nodes in l_nodes
199 merge = policy.Merge( [node['hostname'] for node in l_nodes], toRT)
200 startThread(merge,"merge")
202 rt1 = rt.RT(ad_dbTickets, toRT, fromRT, l_ticket_blacklist)
203 startThread(rt1,"rt1")
205 diagnose = policy.Diagnose(fromRT)
206 startThread(diagnose,"diagnose")
211 if tw.checkThreads() == 0:
213 print "waiting... %s" % time.time()
214 time.sleep(WATCHSLEEP)
216 logger.info('Diagnose Exitting')
219 if __name__ == '__main__':
222 except KeyboardInterrupt:
223 print "Killed. Exitting."
224 logger.info('Diagnose Killed')