nm_mom.py

   1 #!/usr/bin/python
   2 #
   3 # Restarts NM if NM does not respond to periodic probes.
   4 #
   5 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
   6 # Copyright (C) 2006 The Trustees of Princeton University
   7 #
   8 # $Id: $
   9 #
  10
  11 import syslog
  12 import os
  13 import sys
  14 import getopt
  15 import time
  16 import pickle
  17
  18 import socket
  19 import xmlrpclib
  20 import bwlimit
  21
  22 from sets import Set
  23
  24 # Utility functions
  25 from pl_mom import *
  26
  27
  28 def restartNM():
  29         try:
  30                 pid = os.fork()
  31                 if pid == 0:
  32                         bwlimit.run("/etc/init.d/pl_nm stop")
  33                         try:
  34                                 nmpid = bwlimit.run("/bin/ps -e -o pid,cmd | grep nm.py")[0].split(" ")[0]
  35                                 os.kill(int(nmpid), 9)
  36                                 time.sleep(3)
  37                                 print "NM  %s forcefully killed." %nmpid
  38                         except:
  39                                 print "NM %s killed." %nmpid
  40                         bwlimit.run("/etc/init.d/pl_nm restart")
  41                         print "NM restarted."
  42                         os._exit(0)
  43                 else:
  44                         os.waitpid(pid,0)
  45         except Exception, err:
  46                 print "Warning: Exception received while restarting NM:  %s" % err
  47
  48
  49 def main():
  50         syslog.openlog("nm_mom")
  51         sys.stdout = sys.stderr = Logger()
  52
  53         nm = NM()
  54         last_maxrate = 0
  55         try:
  56                 # Query something.  If it doesnt come back...
  57                 vals = nm.query("pl_netflow", [('nm_net_max_rate', last_maxrate)])
  58
  59         except Exception, err:
  60                 print "Warning: Exception received while querying NM:", err
  61                 restartNM()
  62
  63 if __name__ == '__main__':
  64     main()