unified_model.py

   1 #!/usr/bin/python
   2
   3 from monitor import database
   4
   5 import plc
   6 api = plc.getAuthAPI()
   7
   8 import mailer
   9 import time
  10
  11 from model import *
  12 from const import *
  13 import util.file
  14 import config
  15
  16 def gethostlist(hostlist_file):
  17         return util.file.getListFromFile(hostlist_file)
  18
  19         #nodes = api.GetNodes({'peer_id' : None}, ['hostname'])
  20         #return [ n['hostname'] for n in nodes ]
  21
  22 def array_to_priority_map(array):
  23         """ Create a mapping where each entry of array is given a priority equal
  24         to its position in the array.  This is useful for subsequent use in the
  25         cmpMap() function."""
  26         map = {}
  27         count = 0
  28         for i in array:
  29                 map[i] = count
  30                 count += 1
  31         return map
  32
  33 def cmpValMap(v1, v2, map):
  34         if v1 in map and v2 in map and map[v1] < map[v2]:
  35                 return 1
  36         elif v1 in map and v2 in map and map[v1] > map[v2]:
  37                 return -1
  38         elif v1 in map and v2 in map:
  39                 return 0
  40         else:
  41                 raise Exception("No index %s or %s in map" % (v1, v2))
  42
  43 def cmpCategoryVal(v1, v2):
  44         map = array_to_priority_map([ None, 'ALPHA', 'PROD', 'OLDBOOTCD', 'UNKNOWN', 'FORCED', 'ERROR', ])
  45         return cmpValMap(v1,v2,map)
  46
  47
  48 class PCU:
  49         def __init__(self, hostname):
  50                 self.hostname = hostname
  51
  52         def reboot(self):
  53                 return True
  54         def available(self):
  55                 return True
  56         def previous_attempt(self):
  57                 return True
  58         def setValidMapping(self):
  59                 pass
  60
  61 class Penalty:
  62         def __init__(self, key, valuepattern, action):
  63                 pass
  64
  65 class PenaltyMap:
  66         def __init__(self):
  67                 pass
  68
  69         # connect one penalty to another, in a FSM diagram.  After one
  70         #       condition/penalty is applied, move to the next phase.
  71
  72
  73 #fb = database.dbLoad("findbad")
  74
  75 class RT(object):
  76         def __init__(self, ticket_id = None):
  77                 self.ticket_id = ticket_id
  78                 if self.ticket_id:
  79                         print "getting ticket status",
  80                         self.status = mailer.getTicketStatus(self.ticket_id)
  81                         print self.status
  82
  83         def setTicketStatus(self, status):
  84                 mailer.setTicketStatus(self.ticket_id, status)
  85                 self.status = mailer.getTicketStatus(self.ticket_id)
  86                 return True
  87
  88         def getTicketStatus(self):
  89                 if not self.status:
  90                         self.status = mailer.getTicketStatus(self.ticket_id)
  91                 return self.status
  92
  93         def closeTicket(self):
  94                 mailer.closeTicketViaRT(self.ticket_id, "Ticket CLOSED automatically by SiteAssist.")
  95
  96         def email(self, subject, body, to):
  97                 self.ticket_id = mailer.emailViaRT(subject, body, to, self.ticket_id)
  98                 return self.ticket_id
  99
 100 class Message(object):
 101         def __init__(self, subject, message, via_rt=True, ticket_id=None, **kwargs):
 102                 self.via_rt = via_rt
 103                 self.subject = subject
 104                 self.message = message
 105                 self.rt = RT(ticket_id)
 106
 107         def send(self, to):
 108                 if self.via_rt:
 109                         return self.rt.email(self.subject, self.message, to)
 110                 else:
 111                         return mailer.email(self.subject, self.message, to)
 112
 113 class Recent(object):
 114         def __init__(self, withintime):
 115                 self.withintime = withintime
 116
 117                 try:
 118                         self.time = self.__getattribute__('time')
 119                 except:
 120                         self.time = time.time()- 7*24*60*60
 121
 122                 #self.time = time.time()
 123                 #self.action_taken = False
 124
 125         def isRecent(self):
 126                 if self.time + self.withintime < time.time():
 127                         self.action_taken = False
 128
 129                 if self.time + self.withintime > time.time() and self.action_taken:
 130                         return True
 131                 else:
 132                         return False
 133
 134         def unsetRecent(self):
 135                 self.action_taken = False
 136                 self.time = time.time()
 137                 return True
 138
 139         def setRecent(self):
 140                 self.action_taken = True
 141                 self.time = time.time()
 142                 return True
 143
 144 class PersistFlags(Recent):
 145         def __new__(typ, id, *args, **kwargs):
 146                 if 'db' in kwargs:
 147                         db = kwargs['db']
 148                         del kwargs['db']
 149                 else:
 150                         db = "persistflags"
 151
 152                 try:
 153                         pm = database.dbLoad(db)
 154                 except:
 155                         database.dbDump(db, {})
 156                         pm = database.dbLoad(db)
 157                 #print pm
 158                 if id in pm:
 159                         obj = pm[id]
 160                 else:
 161                         obj = super(PersistFlags, typ).__new__(typ, *args, **kwargs)
 162                         for key in kwargs.keys():
 163                                 obj.__setattr__(key, kwargs[key])
 164                         obj.time = time.time()
 165                         obj.action_taken = False
 166
 167                 obj.db = db
 168                 return obj
 169
 170         def __init__(self, id, withintime, **kwargs):
 171                 self.id = id
 172                 Recent.__init__(self, withintime)
 173
 174         def save(self):
 175                 pm = database.dbLoad(self.db)
 176                 pm[self.id] = self
 177                 database.dbDump(self.db, pm)
 178
 179         def resetFlag(self, name):
 180                 self.__setattr__(name, False)
 181
 182         def setFlag(self, name):
 183                 self.__setattr__(name, True)
 184
 185         def getFlag(self, name):
 186                 try:
 187                         return self.__getattribute__(name)
 188                 except:
 189                         self.__setattr__(name, False)
 190                         return False
 191
 192         def resetRecentFlag(self, name):
 193                 self.resetFlag(name)
 194                 self.unsetRecent()
 195
 196         def setRecentFlag(self, name):
 197                 self.setFlag(name)
 198                 self.setRecent()
 199
 200         def getRecentFlag(self, name):
 201                 # if recent and flag set -> true
 202                 # else false
 203                 try:
 204                         return self.isRecent() & self.__getattribute__(name)
 205                 except:
 206                         self.__setattr__(name, False)
 207                         return False
 208
 209         def checkattr(self, name):
 210                 try:
 211                         x = self.__getattribute__(name)
 212                         return True
 213                 except:
 214                         return False
 215
 216
 217 class PersistMessage(Message):
 218         def __new__(typ, id, subject, message, via_rt, **kwargs):
 219                 if 'db' in kwargs:
 220                         db = kwargs['db']
 221                 else:
 222                         db = "persistmessages"
 223
 224                 try:
 225                         pm = database.dbLoad(db)
 226                 except:
 227                         database.dbDump(db, {})
 228                         pm = database.dbLoad(db)
 229
 230                 #print pm
 231                 if id in pm:
 232                         #print "Using existing object"
 233                         obj = pm[id]
 234                 else:
 235                         #print "creating new object"
 236                         obj = super(PersistMessage, typ).__new__(typ, [id, subject, message, via_rt], **kwargs)
 237                         obj.id = id
 238                         obj.actiontracker = Recent(3*60*60*24)
 239                         obj.ticket_id = None
 240
 241                 if 'ticket_id' in kwargs and kwargs['ticket_id'] is not None:
 242                         obj.ticket_id = kwargs['ticket_id']
 243
 244                 obj.db = db
 245                 return obj
 246
 247         def __init__(self, id, subject, message, via_rt=True, **kwargs):
 248                 print "initializing object: %s" % self.ticket_id
 249                 self.id = id
 250                 Message.__init__(self, subject, message, via_rt, self.ticket_id)
 251
 252         def reset(self):
 253                 self.actiontracker.unsetRecent()
 254
 255         def save(self):
 256                 pm = database.dbLoad(self.db)
 257                 pm[self.id] = self
 258                 database.dbDump(self.db, pm)
 259
 260         def send(self, to):
 261                 if not self.actiontracker.isRecent():
 262                         self.ticket_id = Message.send(self, to)
 263                         self.actiontracker.setRecent()
 264                         self.save()
 265                 else:
 266                         # NOTE: only send a new message every week, regardless.
 267                         print "Not sending to host b/c not within window of %s days" % (self.actiontracker.withintime // (60*60*24))
 268
 269 class MonitorMessage(object):
 270         def __new__(typ, id, *args, **kwargs):
 271                 if 'db' in kwargs:
 272                         db = kwargs['db']
 273                 else:
 274                         db = "monitormessages"
 275
 276                 try:
 277                         if 'reset' in kwargs and kwargs['reset'] == True:
 278                                 database.dbDump(db, {})
 279                         pm = database.dbLoad(db)
 280                 except:
 281                         database.dbDump(db, {})
 282                         pm = database.dbLoad(db)
 283
 284                 #print pm
 285                 if id in pm:
 286                         print "Using existing object"
 287                         obj = pm[id]
 288                 else:
 289                         print "creating new object"
 290                         obj = super(object, typ).__new__(typ, id, *args, **kwargs)
 291                         obj.id = id
 292                         obj.sp = PersistSitePenalty(id, 0)
 293
 294                 obj.db = db
 295                 return obj
 296
 297         def __init__(self, id, message):
 298                 pass
 299
 300
 301 class SitePenalty(object):
 302         penalty_map = []
 303         penalty_map.append( { 'name': 'noop',                   'enable'   : lambda host: None,
 304                                                                                                         'disable'  : lambda host: None } )
 305         penalty_map.append( { 'name': 'nocreate',               'enable'   : lambda host: plc.removeSliceCreation(host),
 306                                                                                                         'disable'  : lambda host: plc.enableSliceCreation(host) } )
 307         penalty_map.append( { 'name': 'suspendslices',  'enable'   : lambda host: plc.suspendSlices(host),
 308                                                                                                         'disable'  : lambda host: plc.enableSlices(host) } )
 309
 310         #def __init__(self, index=0, **kwargs):
 311         #       self.index = index
 312
 313         def get_penalties(self):
 314                 # TODO: get penalties actually applied to a node from PLC DB.
 315                 return [ n['name'] for n in SitePenalty.penalty_map ]
 316
 317         def increase(self):
 318                 self.index = self.index + 1
 319                 if self.index > len(SitePenalty.penalty_map)-1: self.index = len(SitePenalty.penalty_map)-1
 320                 return True
 321
 322         def decrease(self):
 323                 self.index = self.index - 1
 324                 if self.index < 0: self.index = 0
 325                 return True
 326
 327         def apply(self, host):
 328
 329                 for i in range(len(SitePenalty.penalty_map)-1,self.index,-1):
 330                         print "\tdisabling %s on %s" % (SitePenalty.penalty_map[i]['name'], host)
 331                         SitePenalty.penalty_map[i]['disable'](host)
 332
 333                 for i in range(0,self.index+1):
 334                         print "\tapplying %s on %s" % (SitePenalty.penalty_map[i]['name'], host)
 335                         SitePenalty.penalty_map[i]['enable'](host)
 336
 337                 return
 338
 339
 340
 341 class PersistSitePenalty(SitePenalty):
 342         def __new__(typ, id, index, **kwargs):
 343                 if 'db' in kwargs:
 344                         db = kwargs['db']
 345                 else:
 346                         db = "persistpenalties"
 347
 348                 try:
 349                         if 'reset' in kwargs and kwargs['reset'] == True:
 350                                 database.dbDump(db, {})
 351                         pm = database.dbLoad(db)
 352                 except:
 353                         database.dbDump(db, {})
 354                         pm = database.dbLoad(db)
 355
 356                 #print pm
 357                 if id in pm:
 358                         print "Using existing object"
 359                         obj = pm[id]
 360                 else:
 361                         print "creating new object"
 362                         obj = super(PersistSitePenalty, typ).__new__(typ, [index], **kwargs)
 363                         obj.id = id
 364                         obj.index = index
 365
 366                 obj.db = db
 367                 return obj
 368
 369         def __init__(self, id, index, **kwargs):
 370                 self.id = id
 371
 372         def save(self):
 373                 pm = database.dbLoad(self.db)
 374                 pm[self.id] = self
 375                 database.dbDump(self.db, pm)
 376
 377
 378 class Target:
 379         """
 380                 Each host has a target set of attributes.  Some may be set manually,
 381                 or others are set globally for the preferred target.
 382
 383                 For instance:
 384                         All nodes in the Alpha or Beta group would have constraints like:
 385                                 [ { 'state' : 'BOOT', 'kernel' : '2.6.22' } ]
 386         """
 387         def __init__(self, constraints):
 388                 self.constraints = constraints
 389
 390         def verify(self, data):
 391                 """
 392                         self.constraints is a list of key, value pairs.
 393                         # [ {... : ...}==AND , ... , ... , ] == OR
 394                 """
 395                 con_or_true = False
 396                 for con in self.constraints:
 397                         #print "con: %s" % con
 398                         con_and_true = True
 399                         for key in con.keys():
 400                                 #print "looking at key: %s" % key
 401                                 if key in data:
 402                                         #print "%s %s" % (con[key], data[key])
 403                                         con_and_true = con_and_true & (con[key] in data[key])
 404                                 elif key not in data:
 405                                         print "missing key %s" % key
 406                                         con_and_true = False
 407
 408                         con_or_true = con_or_true | con_and_true
 409
 410                 return con_or_true
 411
 412 class Record(object):
 413
 414         def __init__(self, hostname, data):
 415                 self.hostname = hostname
 416                 self.data = data
 417                 self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb")
 418                 self.loginbase = self.plcdb_hn2lb[self.hostname]
 419                 return
 420
 421
 422         def stageIswaitforever(self):
 423                 if 'waitforever' in self.data['stage']:
 424                         return True
 425                 else:
 426                         return False
 427
 428         def severity(self):
 429                 category = self.data['category']
 430                 prev_category = self.data['prev_category']
 431                 #print "SEVERITY: ", category, prev_category
 432                 val = cmpCategoryVal(category, prev_category)
 433                 return val
 434
 435         def improved(self):
 436                 return self.severity() > 0
 437
 438         def end_record(self):
 439                 return node_end_record(self.hostname)
 440
 441         def reset_stage(self):
 442                 self.data['stage'] = 'findbad'
 443                 return True
 444
 445         def getCategory(self):
 446                 return self.data['category'].lower()
 447
 448         def getState(self):
 449                 return self.data['state'].lower()
 450
 451         def getDaysDown(cls, diag_record):
 452                 daysdown = -1
 453                 if diag_record['comonstats']['uptime'] != "null":
 454                         daysdown = - int(float(diag_record['comonstats']['uptime'])) // (60*60*24)
 455                 #elif diag_record['comonstats']['sshstatus'] != "null":
 456                 #       daysdown = int(diag_record['comonstats']['sshstatus']) // (60*60*24)
 457                 #elif diag_record['comonstats']['lastcotop'] != "null":
 458                 #       daysdown = int(diag_record['comonstats']['lastcotop']) // (60*60*24)
 459                 else:
 460                         now = time.time()
 461                         last_contact = diag_record['plcnode']['last_contact']
 462                         if last_contact == None:
 463                                 # the node has never been up, so give it a break
 464                                 daysdown = -1
 465                         else:
 466                                 diff = now - last_contact
 467                                 daysdown = diff // (60*60*24)
 468                 return daysdown
 469         getDaysDown = classmethod(getDaysDown)
 470
 471         def getStrDaysDown(cls, diag_record):
 472                 daysdown = "unknown"
 473                 last_contact = diag_record['plcnode']['last_contact']
 474                 date_created = diag_record['plcnode']['date_created']
 475
 476                 if      diag_record['comonstats']['uptime'] != "null" and \
 477                         diag_record['comonstats']['uptime'] != "-1":
 478                         daysdown = int(float(diag_record['comonstats']['uptime'])) // (60*60*24)
 479                         daysdown = "%d days up" % daysdown
 480
 481                 elif last_contact is None:
 482                         if date_created is not None:
 483                                 now = time.time()
 484                                 diff = now - date_created
 485                                 daysdown = diff // (60*60*24)
 486                                 daysdown = "Never contacted PLC, created %s days ago" % daysdown
 487                         else:
 488                                 daysdown = "Never contacted PLC"
 489                 else:
 490                         now = time.time()
 491                         diff = now - last_contact
 492                         daysdown = diff // (60*60*24)
 493                         daysdown = "%s days down" % daysdown
 494                 return daysdown
 495         getStrDaysDown = classmethod(getStrDaysDown)
 496
 497         #def getStrDaysDown(cls, diag_record):
 498         #       daysdown = cls.getDaysDown(diag_record)
 499         #       if daysdown > 0:
 500         #               return "%d days down"%daysdown
 501         #       elif daysdown == -1:
 502         #               return "Never online"
 503         #       else:
 504         #               return "%d days up"% -daysdown
 505         #getStrDaysDown = classmethod(getStrDaysDown)
 506
 507         def takeAction(self):
 508                 pp = PersistSitePenalty(self.hostname, 0, db='persistpenalty_hostnames')
 509                 if 'improvement' in self.data['stage'] or self.improved() or \
 510                         'monitor-end-record' in self.data['stage']:
 511                         print "takeAction: decreasing penalty for %s"%self.hostname
 512                         pp.decrease()
 513                         pp.decrease()
 514                 else:
 515                         print "takeAction: increasing penalty for %s"%self.hostname
 516                         pp.increase()
 517                 pp.apply(self.hostname)
 518                 pp.save()
 519
 520         def _format_diaginfo(self):
 521                 info = self.data['info']
 522                 print "FORMAT : STAGE: ", self.data['stage']
 523                 if self.data['stage'] == 'monitor-end-record':
 524                         if info[2] == "ALPHA": info = (info[0], info[1], "PROD")
 525                         hlist = "    %s went from '%s' to '%s'\n" % (info[0], info[1], info[2])
 526                 else:
 527                         hlist = "    %s %s - %s\n" % (info[0], info[2], info[1]) #(node,ver,daysdn)
 528                 return hlist
 529         def saveAction(self):
 530                 if 'save-act-all' in self.data and self.data['save-act-all'] == True:
 531                         return True
 532                 else:
 533                         return False
 534
 535         def getMessage(self, ticket_id=None):
 536                 self.data['args']['hostname'] = self.hostname
 537                 self.data['args']['loginbase'] = self.loginbase
 538                 self.data['args']['hostname_list'] = self._format_diaginfo()
 539                 #print self.data['message']
 540                 if self.data['message']:
 541                         message = PersistMessage(self.hostname,
 542                                                                  self.data['message'][0] % self.data['args'],
 543                                                                  self.data['message'][1] % self.data['args'],
 544                                                                  True, db='monitor_persistmessages',
 545                                                                  ticket_id=ticket_id)
 546                         return message
 547                 else:
 548                         return None
 549
 550         def getContacts(self):
 551                 roles = self.data['email']
 552
 553                 if not config.mail and not config.debug and config.bcc:
 554                         roles = ADMIN
 555                 if config.mail and config.debug:
 556                         roles = ADMIN
 557
 558                 # build targets
 559                 contacts = []
 560                 if ADMIN & roles:
 561                         contacts += [config.email]
 562                 if TECH & roles:
 563                         contacts += [TECHEMAIL % self.loginbase]
 564                 if PI & roles:
 565                         contacts += [PIEMAIL % self.loginbase]
 566                 if USER & roles:
 567                         slices = plc.slices(self.loginbase)
 568                         if len(slices) >= 1:
 569                                 for slice in slices:
 570                                         contacts += [SLICEMAIL % slice]
 571                                 print "SLIC: %20s : %d slices" % (self.loginbase, len(slices))
 572                         else:
 573                                 print "SLIC: %20s : 0 slices" % self.loginbase
 574
 575                 return contacts
 576
 577
 578 class NodeRecord:
 579         def __init__(self, hostname, target):
 580                 self.hostname = hostname
 581                 self.ticket = None
 582                 self.target = target
 583                 #if hostname in fb['nodes']:
 584                 #       self.data = fb['nodes'][hostname]['values']
 585                 #else:
 586                 #       raise Exception("Hostname not in scan database")
 587
 588         def stageIswaitforever(self):
 589                 if 'waitforever' in self.data['stage']:
 590                         return True
 591                 else:
 592                         return False
 593
 594         def severity(self):
 595                 category = self.data['category']
 596                 prev_category = self.data['prev_category']
 597                 print "IMPROVED: ", category, prev_category
 598                 val = cmpCategoryVal(category, prev_category)
 599                 return val
 600
 601         def improved(self):
 602                 return self.severity() > 0
 603
 604         def end_record(self):
 605                 return node_end_record(self.hostname)
 606
 607         def reset_stage(self):
 608                 self.data['stage'] = 'findbad'
 609                 return True
 610
 611         def open_tickets(self):
 612                 if self.ticket and self.ticket.status['status'] == 'open':
 613                         return 1
 614                 return 0
 615         def setIntrospect(self):
 616                 pass
 617
 618         def email_notice(self):
 619                 message = self._get_message_for_condition()
 620                 message.send(self._get_contacts_for_condition())
 621                 return True
 622         def close_ticket(self):
 623                 if self.ticket:
 624                         self.ticket.closeTicket()
 625
 626         def exempt_from_penalties(self):
 627                 bl = database.dbLoad("l_blacklist")
 628                 return self.hostname in bl
 629
 630         def penalties(self):
 631                 return []
 632         def escellate_penalty(self):
 633                 return True
 634         def reduce_penalty(self):
 635                 return True
 636
 637
 638         def atTarget(self):
 639                 return self.target.verify(self.data)
 640
 641         def _get_condition(self):
 642                 return self.data['category'].lower()
 643
 644         def _get_stage(self):
 645                 "improvement"
 646                 "firstnotice_noop"
 647                 "secondnotice_noslicecreation"
 648                 "thirdnotice_disableslices"
 649
 650                 delta = current_time - self.data['time']
 651
 652         def _get_message_for_condition(self):
 653                 pass
 654         def _get_contacts_for_condition(self):
 655                 pass
 656
 657 class Action(MonRecord):
 658         def __init__(self, host, data):
 659                 self.host = host
 660                 MonRecord.__init__(self, data)
 661                 return
 662
 663         def deltaDays(self, delta):
 664                 t = datetime.fromtimestamp(self.__dict__['time'])
 665                 d = t + timedelta(delta)
 666                 self.__dict__['time'] = time.mktime(d.timetuple())
 667
 668 def node_end_record(node):
 669         act_all = database.dbLoad("act_all")
 670         if node not in act_all:
 671                 del act_all
 672                 return False
 673
 674         if len(act_all[node]) == 0:
 675                 del act_all
 676                 return False
 677
 678         pm = database.dbLoad("monitor_persistmessages")
 679         if node not in pm:
 680                 del pm
 681                 return False
 682         else:
 683                 print "deleting node record"
 684                 del pm[node]
 685                 database.dbDump("monitor_persistmessages", pm)
 686
 687         a = Action(node, act_all[node][0])
 688         a.delField('rt')
 689         a.delField('found_rt_ticket')
 690         a.delField('second-mail-at-oneweek')
 691         a.delField('second-mail-at-twoweeks')
 692         a.delField('first-found')
 693         rec = a.get()
 694         rec['action'] = ["close_rt"]
 695         rec['category'] = "ALPHA"       # assume that it's up...
 696         rec['stage'] = "monitor-end-record"
 697         rec['ticket_id'] = None
 698         rec['time'] = time.time() - 7*60*60*24
 699         act_all[node].insert(0,rec)
 700         database.dbDump("act_all", act_all)
 701         del act_all
 702         return True
 703
 704 if __name__ == "__main__":
 705         #r = RT()
 706         #r.email("test", "body of test message", ['database@cs.princeton.edu'])
 707         #from emailTxt import mailtxt
 708         print "loaded"
 709         #database.dbDump("persistmessages", {});
 710         #args = {'url_list': 'http://www.planet-lab.org/bootcds/planet1.usb\n','hostname': 'planet1','hostname_list': ' blahblah -  days down\n'}
 711         #m = PersistMessage("blue", "test 1", mailtxt.newdown_one[1] % args, True)
 712         #m.send(['soltesz@cs.utk.edu'])
 713         #m = PersistMessage("blue", "test 1 - part 2", mailtxt.newalphacd_one[1] % args, True)
 714         # TRICK timer to thinking some time has passed.
 715         #m.actiontracker.time = time.time() - 6*60*60*24
 716         #m.send(['soltesz@cs.utk.edu'])