policy.py

   1 #
   2 # Copyright (c) 2004  The Trustees of Princeton University (Trustees).
   3 #
   4 # Faiyaz Ahmed <faiyaza@cs.princeton.edu>
   5 #
   6 # $Id: policy.py,v 1.7 2007/01/11 21:39:07 faiyaza Exp $
   7 #
   8 # Policy Engine.
   9
  10 #from monitor import *
  11 from threading import *
  12 import time
  13 import logging
  14 import mailer
  15 import emailTxt
  16 import pickle
  17 import Queue
  18 import plc
  19 import reboot
  20 import config
  21
  22 DAT="./monitor.dat"
  23
  24 logger = logging.getLogger("monitor")
  25
  26 # Time to enforce policy
  27 POLSLEEP = 7200
  28
  29 # Where to email the summary
  30 SUMTO = "faiyaza@cs.princeton.edu"
  31 TECHEMAIL="tech-%s@sites.planet-lab.org"
  32 PIEMAIL="pi-%s@sites.planet-lab.org"
  33 SLICEMAIL="%s@slices.planet-lab.org"
  34 PLCEMAIL="support@planet-lab.org"
  35
  36 #Thresholds (DAYS)
  37 SPERDAY = 86400
  38 PITHRESH = 2 * SPERDAY
  39 SLICETHRESH = 5 * SPERDAY
  40 # Days before attempting rins again
  41 RINSTHRESH = 5 * SPERDAY
  42
  43 # Minimum number of nodes up before squeezing
  44 MINUP = 2
  45
  46 # IF:
  47 #  no SSH, down.
  48 #  bad disk, down
  49 #  DNS, kinda down (sick)
  50 #  clock, kinda down (sick)
  51 #  Full disk, going to be down
  52
  53 # Actions:
  54 #  Email
  55 #  suspend slice creation
  56 #  kill slices
  57 class Policy(Thread):
  58         def __init__(self, comonthread, sickNoTicket, emailed):
  59                 self.cmn = comonthread
  60                 # host - > (time of email, type of email)
  61                 self.emailed = emailed
  62                 # all sick nodes w/o tickets
  63                 self.sickNoTicket = sickNoTicket
  64                 # Sitess we've Squeezed.
  65                 self.squeezed = {}
  66                 Thread.__init__(self)
  67
  68
  69         '''
  70         What to do when node is in dbg (as reported by CoMon).
  71         '''
  72         def __actOnDebug(self, node):
  73                 # Check to see if we've done this before
  74                 if (node in self.emailed.keys()):
  75                         if (self.emailed[node][0] == "dbg"):
  76                                 delta = time.time() - self.emailed[node][1]
  77                                 if (delta <= RINSTHRESH ):
  78                                         # Don't mess with node if under Thresh.
  79                                         # Return, move on.
  80                                         logger.info("POLICY:  %s in dbg, but acted on %s days ago" % (node, delta // SPERDAY))
  81                                         return
  82                         logger.info("POLICY:  Node in dbg - " + node)
  83                         plc.nodeBootState(node, "rins")
  84                         # If it has a PCU
  85                         return reboot.reboot(node)
  86
  87         '''
  88         What to do when node is in dbg (as reported by CoMon).
  89         '''
  90         def __actOnFilerw(self, node):
  91                 target = [PLCEMAIL]
  92                 logger.info("POLICY:  Emailing PLC for " + node)
  93                 tmp = emailTxt.mailtxt.filerw
  94                 sbj = tmp[0] % {'hostname': node}
  95                 msg = tmp[1] % {'hostname': node}
  96                 mailer.email(sbj, msg, target)
  97                 self.emailed[node] = ("filerw", time.time())
  98
  99
 100         '''
 101         Acts on sick nodes.
 102         '''
 103         def actOnSick(self):
 104                 # Get list of nodes in debug from PLC
 105                 #dbgNodes = NodesDebug()
 106                 global TECHEMAIL, PIEMAIL
 107                 # Grab a node from the queue (pushed by rt thread).
 108                 node = self.sickNoTicket.get(block = True)
 109                 # Get the login base
 110                 loginbase = plc.siteId(node)
 111
 112                 # Princeton Backdoor
 113                 if loginbase == "princeton": return
 114
 115                 # Send appropriate message for node if in appropriate bucket.
 116                 # If we know where to send a message
 117                 if not loginbase:
 118                         logger.info("POLICY:  loginbase for %s not found" %node)
 119                 # And we didn't email already.
 120                 else:
 121                         # If first email, send to Tech
 122                         target = [TECHEMAIL % loginbase]
 123
 124                         # If disk is foobarred, PLC should check it.
 125                         if (node in self.cmn.filerw) and \
 126                         (node not in self.emailed.keys()):
 127                                 self.__actOnFilerw(node)
 128                                 return
 129
 130                         # If in dbg, set to rins, then reboot.  Inform PLC.
 131                         if (node in self.cmn.dbg):
 132                         # If reboot failure via PCU, POD and send email
 133                         # if contacted PCU, return
 134                                 if self.__actOnDebug(node):  return
 135
 136                         if (node in self.emailed.keys()) and \
 137                         (node not in self.cmn.filerw)    and \
 138                         (node not in self.cmn.clock_drift):
 139                                 # If we emailed before, how long ago?
 140                                 delta = time.time() - self.emailed[node][1]
 141                                 if delta < SPERDAY:
 142                                         logger.info("POLICY:  already acted on %s today." % node)
 143                                         return
 144
 145                                 logger.info("POLICY:  acted %s on %s days ago" % (node,
 146                                 delta // SPERDAY))
 147
 148                                 # If no luck with tech, email PI
 149                                 if (delta >= 1):
 150                                         target.append(PIEMAIL % loginbase)
 151
 152                                 # If more than PI thresh, but less than slicethresh
 153                                 if (delta >= PITHRESH) and (delta < SLICETHRESH):
 154                                         #remove slice creation if enough nodes arent up
 155                                         if not self.enoughUp(loginbase):
 156                                                 slices = plc.slices(loginbase)
 157                                                 if len(slices) >= 1:
 158                                                         for slice in slices:
 159                                                                 target.append(SLICEMAIL % slice)
 160                                                 logger.info("POLICY:  Removing slice creation from %s" % loginbase)
 161                                                 tmp = emailTxt.mailtxt.removedSliceCreation
 162                                                 sbj = tmp[0]
 163                                                 msg = tmp[1] % {'loginbase': loginbase}
 164                                                 plc.removeSliceCreation(node)
 165                                                 mailer.email(sbj, msg, target)
 166                                                 self.squeezed[loginbase] = (time.time(), "creation")
 167                                                 self.emailed[node] = ("creation", time.time())
 168                                                 return
 169
 170                                 # If more than PI thresh and slicethresh
 171                                 if (delta >= PITHRESH) and (delta > SLICETHRESH):
 172                                         target.append(PIEMAIL % loginbase)
 173                                         # Email slices at site.
 174                                         slices = plc.slices(loginbase)
 175                                         if len(slices) >= 1:
 176                                                 for slice in slices:
 177                                                         target.append(SLICEMAIL % slice)
 178                                         # If not enough up, freeze slices and email everyone.
 179                                         if not self.enoughUp(loginbase):
 180                                                 logger.info("POLICY:  Suspending %s slices." % loginbase)
 181                                                 tmp = emailTxt.mailtxt.suspendSlices
 182                                                 sbj = tmp[0]
 183                                                 msg = tmp[1] % {'loginbase': loginbase}
 184                                                 plc.suspendSlices(node)
 185                                                 self.squeezed[loginbase] = (time.time(), "freeze")
 186                                                 mailer.email(sbj, msg, target)
 187                                                 self.emailed[node] = ("freeze", time.time())
 188                                                 return
 189
 190                         # Find the bucket the node is in and send appropriate email
 191                         # to approriate list of people.
 192                         for bkt in self.cmn.comonbkts.keys():
 193                                 if (node in getattr(self.cmn, bkt)):
 194                                         # Send predefined message for that bucket.
 195                                         logger.info("POLICY: Emailing (%s) %s - %s"\
 196                                                 %(bkt, node, target))
 197                                         tmp = getattr(emailTxt.mailtxt, bkt)
 198                                         sbj = tmp[0] % {'hostname': node}
 199                                         msg = tmp[1] % {'hostname': node}
 200                                         mailer.email(sbj, msg, target)
 201                                         self.emailed[node] = (bkt , time.time())
 202                                         return
 203
 204
 205         '''
 206         Prints, logs, and emails status of up nodes, down nodes, and buckets.
 207         '''
 208         def status(self):
 209                 sub = "Monitor Summary"
 210                 msg = "\nThe following nodes were acted upon:  \n\n"
 211                 for (node, (type, date)) in self.emailed.items():
 212                         # Print only things acted on today.
 213                         if (time.gmtime(time.time())[2] == time.gmtime(date)[2]):
 214                                 msg +="%s\t(%s)\t%s\n" %(node, type, time.ctime(date))
 215                 msg +="\n\nThe following sites have been 'squeezed':\n\n"
 216                 for (loginbase, (date, type)) in self.squeezed.items():
 217                         # Print only things acted on today.
 218                         if (time.gmtime(time.time())[2] == time.gmtime(date)[2]):
 219                                 msg +="%s\t(%s)\t%s\n" %(loginbase, type, time.ctime(date))
 220                 mailer.email(sub, msg, [SUMTO])
 221                 logger.info(msg)
 222                 return
 223
 224         '''
 225         Store/Load state of emails.  When, where, what.
 226         '''
 227         def emailedStore(self, action):
 228                 try:
 229                         if action == "LOAD":
 230                                 f = open(DAT, "r+")
 231                                 logger.info("POLICY:  Found and reading " + DAT)
 232                                 self.emailed.update(pickle.load(f))
 233                         if action == "WRITE":
 234                                 f = open(DAT, "w")
 235                                 #logger.debug("Writing " + DAT)
 236                                 pickle.dump(self.emailed, f)
 237                         f.close()
 238                 except Exception, err:
 239                         logger.info("POLICY:  Problem with DAT, %s" %err)
 240
 241         '''
 242         Returns True if more than MINUP nodes are up at a site.
 243         '''
 244         def enoughUp(self, loginbase):
 245                 allsitenodes = plc.getSiteNodes(loginbase)
 246                 if len(allsitenodes) == 0:
 247                         logger.info("Node not in db")
 248                         return
 249
 250                 numnodes = len(allsitenodes)
 251                 sicknodes = []
 252                 # Get all sick nodes from comon
 253                 for bucket in self.cmn.comonbkts.keys():
 254                         for host in getattr(self.cmn, bucket):
 255                                 sicknodes.append(host)
 256                 # Diff.
 257                 for node in allsitenodes:
 258                         if node in sicknodes:
 259                                 numnodes -= 1
 260
 261                 if numnodes < MINUP:
 262                         logger.info(\
 263 "POLICY:  site with %s has nodes %s up." %(loginbase, numnodes))
 264                         return False
 265                 else:
 266                         return True
 267
 268
 269
 270
 271         def run(self):
 272                 while 1:
 273                         self.actOnSick()
 274                         self.emailedStore("WRITE")
 275
 276
 277 def main():
 278         logger.setLevel(logging.DEBUG)
 279         ch = logging.StreamHandler()
 280         ch.setLevel(logging.DEBUG)
 281         formatter = logging.Formatter('%(message)s')
 282         ch.setFormatter(formatter)
 283         logger.addHandler(ch)
 284
 285         #print NodesDebug()
 286         #tmp = Queue.Queue()
 287         #a = Policy(None, tmp)
 288         #a.emailedStore("LOAD")
 289         #print a.emailed
 290
 291         print plc.slices(plc.siteId("alice.cs.princeton.edu"))
 292         os._exit(0)
 293 if __name__ == '__main__':
 294         import os
 295         import plc
 296         try:
 297                 main()
 298         except KeyboardInterrupt:
 299                 print "Killed.  Exitting."
 300                 logger.info('Monitor Killed')
 301                 os._exit(0)