clean_policy.py

   1 import config
   2 import database
   3 import time
   4 import mailer
   5 from unified_model import cmpCategoryVal
   6 import sys
   7 import emailTxt
   8 import string
   9
  10 from rt import is_host_in_rt_tickets
  11 import plc
  12
  13 # Time to enforce policy
  14 POLSLEEP = 7200
  15
  16 # Where to email the summary
  17 SUMTO = "soltesz@cs.princeton.edu"
  18
  19 from const import *
  20
  21 from unified_model import *
  22
  23 def get_ticket_id(record):
  24         if 'ticket_id' in record and record['ticket_id'] is not "" and record['ticket_id'] is not None:
  25                 return record['ticket_id']
  26         elif            'found_rt_ticket' in record and \
  27                  record['found_rt_ticket'] is not "" and \
  28                  record['found_rt_ticket'] is not None:
  29                 return record['found_rt_ticket']
  30         else:
  31                 return None
  32
  33 class MonitorMergeDiagnoseSendEscellate:
  34         act_all = None
  35         fb = None
  36
  37         def __init__(self, hostname, act):
  38                 self.hostname = hostname
  39                 self.act = act
  40                 self.plcdb_hn2lb = None
  41                 if self.plcdb_hn2lb is None:
  42                         self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb")
  43                 self.loginbase = self.plcdb_hn2lb[self.hostname]
  44                 return
  45
  46         def getFBRecord(self):
  47                 if MonitorMergeDiagnoseSendEscellate.fb == None:
  48                         MonitorMergeDiagnoseSendEscellate.fb = database.dbLoad("findbad")
  49
  50                 fb = MonitorMergeDiagnoseSendEscellate.fb
  51
  52                 if self.hostname in fb['nodes']:
  53                         fbnode = fb['nodes'][self.hostname]['values']
  54                 else:
  55                         raise Exception("Hostname %s not in scan database"% self.hostname)
  56                 return fbnode
  57
  58         def getActionRecord(self):
  59                 # update ticket status
  60                 if MonitorMergeDiagnoseSendEscellate.act_all == None:
  61                         MonitorMergeDiagnoseSendEscellate.act_all = database.dbLoad("act_all")
  62
  63                 act_all = MonitorMergeDiagnoseSendEscellate.act_all
  64
  65                 if self.hostname in act_all and len(act_all[self.hostname]) > 0:
  66                         actnode = act_all[self.hostname][0]
  67                 else:
  68                         actnode = None
  69                 return actnode
  70
  71         def getKernel(self, unamestr):
  72                 s = unamestr.split()
  73                 if len(s) > 2:
  74                         return s[2]
  75                 else:
  76                         return ""
  77
  78         def mergeRecord(self, fbnode, actnode):
  79                 fbnode['kernel'] = self.getKernel(fbnode['kernel'])
  80                 fbnode['stage'] = "findbad"
  81                 fbnode['message'] = None
  82                 fbnode['args'] = None
  83                 fbnode['info'] = None
  84                 fbnode['log'] = None
  85                 fbnode['time'] = time.time()
  86                 fbnode['email'] = TECH
  87                 fbnode['action'] = ['noop']
  88                 fbnode['date_created'] = time.time()
  89
  90                 if actnode is None: # there is no entry in act_all
  91                         actnode = {}
  92                         actnode.update(fbnode)
  93                         actnode['ticket_id'] = ""
  94                         actnode['prev_category'] = "ERROR"
  95                 else:
  96                         actnode['prev_category']= actnode['category']
  97                         actnode['comonstats']   = fbnode['comonstats']
  98                         actnode['category']             = fbnode['category']
  99                         actnode['state']                = fbnode['state']
 100                         actnode['kernel']               = fbnode['kernel']
 101                         actnode['bootcd']               = fbnode['bootcd']
 102                         actnode['plcnode']              = fbnode['plcnode']
 103                         ticket = get_ticket_id(actnode)
 104                         if ticket is None: actnode['ticket_id'] = ""
 105                         actnode['rt'] = mailer.getTicketStatus(ticket)
 106
 107                         #for key in actnode.keys():
 108                         #       print "%10s %s %s " % (key, "==", actnode[key])
 109                         #print "----------------------------"
 110
 111                 return actnode
 112
 113         def run(self):
 114                 fbnode = self.getFBRecord()
 115                 actnode= self.getActionRecord()
 116                 actrec = self.mergeRecord(fbnode, actnode)
 117                 record = Record(self.hostname, actrec)
 118                 diag   = self.diagnose(record)
 119                 if self.act and diag is not None:
 120                         self.action(record,diag)
 121
 122         def diagnose(self, record):
 123
 124                 diag = PersistFlags(record.hostname, 60*60*24, db='persist_diagnose_flags')
 125                 # NOTE: change record stage based on RT status.
 126                 #diag.setFlag('ResetStage')
 127                 if record.stageIswaitforever():
 128                         ticket = record.data['rt']
 129                         if 'new' in ticket['Status']:
 130                                 print "Resetting Stage!!!!!"
 131                         #       diag.setFlag('ResetStage')
 132                                 record.reset_stage()
 133                         #if diag.getFlag('ResetStage'):
 134                         #       print "diagnose: resetting stage"
 135                         #       diag.resetFlag('ResetStage')
 136
 137                         if 'resolved' in ticket['Status']:
 138                                 diag.setFlag('RTEndRecord')
 139
 140                 # NOTE: take category, and prepare action
 141                 category = record.getCategory()
 142                 if category == "error":
 143                         diag.setFlag('SendNodedown')
 144                         record.data['message_series'] = emailTxt.mailtxt.newdown
 145                         record.data['log'] = self.getDownLog(record)
 146
 147                 elif category == "prod" or category == "alpha":
 148                         state = record.getState()
 149                         if state == "boot":
 150                                 if record.severity() != 0:
 151                                         diag.setFlag('SendThankyou')
 152                                         print "RESETTING STAGE: improvement"
 153                                         record.data['stage'] = 'improvement'
 154                                         record.data['message_series'] = emailTxt.mailtxt.newthankyou
 155                                         record.data['log'] = self.getThankyouLog(record)
 156                                 else:
 157                                         # NOTE: do nothing, since we've already done the above.
 158                                         print "DIAGNOSED: %s is boot. no further action necessary." % record.hostname
 159                                         return None
 160                         elif state == "debug":
 161                                 pass
 162                         else:
 163                                 print "unknown state %s for host %s" % (state, self.hostname)
 164                 else:
 165                         print "unknown category: %s" % category
 166
 167
 168                 # TODO: how to not send email?...
 169                 record = self.checkStageAndTime(diag,record)
 170                 #if record:
 171                 print "diagnose: checkStageAndTime Returned Valid Record"
 172                 site = PersistFlags(self.loginbase, 1, db='site_persistflags')
 173
 174                 if site.status != "good":
 175                         print "diagnose: Setting site %s for 'squeeze'" % self.loginbase
 176                         diag.setFlag('Squeeze')
 177                 else:
 178                         print "diagnose: Setting site %s for 'backoff'" % self.loginbase
 179                         diag.setFlag('BackOff')
 180
 181                 diag.save()
 182                 return diag
 183                 #else:
 184                 #       print "checkStageAndTime Returned NULL Record"
 185                 #       return None
 186
 187         def action(self, record, diag):
 188
 189                 message = None
 190
 191                 #print record.data['stage']
 192                 #print "improvement" in record.data['stage']
 193                 #print self.getSendEmailFlag(record)
 194                 if self.getSendEmailFlag(record) or "monitor-end-record" in record.data['stage']:
 195                         print "action: getting message"
 196                         message = record.getMessage(record.data['ticket_id'])
 197                         if message:
 198                                 #message.reset()
 199                                 print "action: sending email"
 200                                 message.send(record.getContacts())
 201                                 #print "DEBUG NOT SENDING MESSAGE WHEN I SHOULD BE!!!!!"
 202                                 #print "DEBUG NOT SENDING MESSAGE WHEN I SHOULD BE!!!!!"
 203                                 #print "DEBUG NOT SENDING MESSAGE WHEN I SHOULD BE!!!!!"
 204                                 #print message
 205                                 if message.rt.ticket_id:
 206                                         print "action: setting record ticket_id"
 207                                         record.data['ticket_id'] = message.rt.ticket_id
 208
 209                         if (record.data['takeaction'] and diag.getFlag('Squeeze') ) or diag.getFlag('BackOff'):
 210                                 print "action: taking action"
 211                                 record.takeAction()
 212                                 diag.resetFlag('Squeeze')
 213                                 diag.resetFlag('BackOff')
 214                                 diag.save()
 215
 216                         if record.saveAction():
 217                                 print "action: saving act_all db"
 218                                 self.add_and_save_act_all(record)
 219                         else:
 220                                 print "action: NOT saving act_all db"
 221                                 print "stage: %s %s" % ( record.data['stage'], record.data['save-act-all'] )
 222
 223                         if record.improved() or diag.getFlag('RTEndRecord'):
 224                                 print "action: end record for %s" % self.hostname
 225                                 record.end_record()
 226                                 diag.setFlag('CloseRT')
 227                                 diag.resetFlag('RTEndRecord')
 228                                 diag.save()
 229                                 #return None
 230
 231                         if message:
 232                                 if diag.getFlag('CloseRT'):
 233                                         message.rt.closeTicket()
 234                                         diag.resetFlag('CloseRT')
 235                                         diag.save()
 236
 237                 else:
 238                         print "NOT sending email : %s %s" % (config.mail, record.data['rt'])
 239
 240                 return
 241
 242         def getSendEmailFlag(self, record):
 243                 if not config.mail:
 244                         return False
 245
 246                 # resend if open & created longer than 30 days ago.
 247                 if  'rt' in record.data and \
 248                         'Status' in record.data['rt'] and \
 249                         "open" in record.data['rt']['Status'] and \
 250                         record.data['rt']['Created'] > int(time.time() - 60*60*24*30):
 251                         # if created-time is greater than the thirty days ago from the current time
 252                         return False
 253
 254                 return True
 255
 256         def add_and_save_act_all(self, record):
 257                 self.act_all = database.dbLoad("act_all")
 258                 if self.hostname not in self.act_all:
 259                         self.act_all[self.hostname] = []
 260                 self.act_all[self.hostname].insert(0,record.data)
 261                 database.dbDump("act_all", self.act_all)
 262
 263         def getDownLog(self, record):
 264
 265                 record.data['args'] = {'nodename': self.hostname}
 266                 record.data['info'] = (self.hostname, Record.getStrDaysDown(record.data), "")
 267
 268                 #for key in record.data.keys():
 269                 #       print "%10s %s %s " % (key, "==", record.data[key])
 270
 271                 if record.data['ticket_id'] == "" and 'found_rt_ticket' in record.data:
 272                         log = "DOWN: %20s : %-40s == %20s %s" % \
 273                                 (self.loginbase, self.hostname, record.data['info'][1:], record.data['found_rt_ticket'])
 274                 else:
 275                         log = "DOWN: %20s : %-40s == %20s %s" % \
 276                                 (self.loginbase, self.hostname, record.data['info'][1:], record.data['ticket_id'])
 277                 return log
 278
 279         def getThankyouLog(self, record):
 280
 281                 record.data['args'] = {'nodename': self.hostname}
 282                 record.data['info'] = (self.hostname, record.data['prev_category'], record.data['category'])
 283
 284                 try:
 285                         if record.data['ticket_id'] == "" and 'found_rt_ticket' in record.data:
 286                                 log = "IMPR: %20s : %-40s == %20s %20s %s %s" % \
 287                                                 (self.loginbase, self.hostname, record.data['stage'],
 288                                                  record.data['prev_category'], record.data['category'], record.data['found_rt_ticket'])
 289                         else:
 290                                 log = "IMPR: %20s : %-40s == %20s %20s %s %s" % \
 291                                                 (self.loginbase, self.hostname, record.data['stage'],
 292                                                  record.data['prev_category'], record.data['category'], record.data['ticket_id'])
 293                 except:
 294                         log = "IMPR: %s improved to %s " % (self.hostname, record.data['category'])
 295                 return log
 296
 297         def checkStageAndTime(self, diag, record):
 298                 current_time = time.time()
 299                 delta = current_time - record.data['time']
 300                 #print record.data
 301                 if   'findbad' in record.data['stage']:
 302                         # The node is bad, and there's no previous record of it.
 303                         record.data['email'] = TECH
 304                         record.data['action'] = ['noop']
 305                         record.data['takeaction'] = False
 306                         record.data['message'] = record.data['message_series'][0]
 307                         record.data['stage'] = 'stage_actinoneweek'
 308                         record.data['save-act-all'] = True
 309
 310                 elif 'reboot_node' in record.data['stage']:
 311                         record.data['email'] = TECH
 312                         record.data['action'] = ['noop']
 313                         record.data['message'] = record.data['message_series'][0]
 314                         record.data['stage'] = 'stage_actinoneweek'
 315                         record.data['takeaction'] = False
 316                         record.data['save-act-all'] = False
 317
 318                 elif 'improvement' in record.data['stage']:
 319                         print "checkStageAndTime: backing off of %s" % self.hostname
 320                         record.data['action'] = ['close_rt']
 321                         record.data['takeaction'] = True
 322                         record.data['message'] = record.data['message_series'][0]
 323                         record.data['stage'] = 'monitor-end-record'
 324                         record.data['save-act-all'] = True
 325
 326                 elif 'actinoneweek' in record.data['stage']:
 327                         if delta >= 7 * SPERDAY:
 328                                 print "checkStageAndTime: transition to next stage actintwoweeks"
 329                                 record.data['email'] = TECH | PI
 330                                 record.data['stage'] = 'stage_actintwoweeks'
 331                                 record.data['message'] = record.data['message_series'][1]
 332                                 record.data['action'] = ['nocreate' ]
 333                                 record.data['time'] = current_time              # reset clock for waitforever
 334                                 record.data['takeaction'] = True
 335                                 record.data['save-act-all'] = True
 336                         elif delta >= 3* SPERDAY and not 'second-mail-at-oneweek' in record.data:
 337                                 print "checkStageAndTime: second message in one week"
 338                                 record.data['email'] = TECH
 339                                 record.data['message'] = record.data['message_series'][0]
 340                                 record.data['action'] = ['sendmailagain-waitforoneweekaction' ]
 341                                 record.data['second-mail-at-oneweek'] = True
 342                                 record.data['takeaction'] = False
 343                                 record.data['save-act-all'] = True
 344                         else:
 345                                 record.data['message'] = None
 346                                 record.data['action'] = ['waitforoneweekaction' ]
 347                                 record.data['takeaction'] = False
 348                                 record.data['save-act-all'] = False
 349                                 print "checkStageAndTime: ignoring this record for: %s" % self.hostname
 350                                 #return None                    # don't send if there's no action
 351
 352                 elif 'actintwoweeks' in record.data['stage']:
 353                         if delta >= 7 * SPERDAY:
 354                                 print "checkStageAndTime: transition to next stage waitforever"
 355                                 record.data['email'] = TECH | PI | USER
 356                                 record.data['stage'] = 'stage_waitforever'
 357                                 record.data['message'] = record.data['message_series'][2]
 358                                 record.data['action'] = ['suspendslices']
 359                                 record.data['time'] = current_time              # reset clock for waitforever
 360                                 record.data['takeaction'] = True
 361                                 record.data['save-act-all'] = True
 362                         elif delta >= 3* SPERDAY and not 'second-mail-at-twoweeks' in record.data:
 363                                 print "checkStageAndTime: second message in one week for stage two"
 364                                 record.data['email'] = TECH | PI
 365                                 record.data['message'] = record.data['message_series'][1]
 366                                 record.data['action'] = ['sendmailagain-waitfortwoweeksaction' ]
 367                                 record.data['second-mail-at-twoweeks'] = True
 368                                 record.data['takeaction'] = False
 369                                 record.data['save-act-all'] = True
 370                         else:
 371                                 record.data['message'] = None
 372                                 record.data['takeaction'] = False
 373                                 record.data['action'] = ['waitfortwoweeksaction']
 374                                 record.data['save-act-all'] = False
 375                                 print "checkStageAndTime: second message in one week for stage two"
 376                                 #return None                    # don't send if there's no action
 377
 378                 elif 'ticket_waitforever' in record.data['stage']:
 379                         record.data['email'] = TECH
 380                         record.data['takeaction'] = True
 381                         if 'first-found' not in record.data:
 382                                 record.data['first-found'] = True
 383                                 record.data['log'] += " firstfound"
 384                                 record.data['action'] = ['ticket_waitforever']
 385                                 record.data['message'] = None
 386                                 record.data['time'] = current_time
 387                                 record.data['save-act-all'] = True
 388                         else:
 389                                 if delta >= 7*SPERDAY:
 390                                         record.data['action'] = ['ticket_waitforever']
 391                                         record.data['message'] = None
 392                                         record.data['time'] = current_time              # reset clock
 393                                         record.data['save-act-all'] = True
 394                                 else:
 395                                         record.data['action'] = ['ticket_waitforever']
 396                                         record.data['message'] = None
 397                                         record.data['takeaction'] = False
 398                                         record.data['save-act-all'] = False
 399                                         return None
 400
 401                 elif 'waitforever' in record.data['stage']:
 402                         # more than 3 days since last action
 403                         # TODO: send only on weekdays.
 404                         # NOTE: expects that 'time' has been reset before entering waitforever stage
 405                         record.data['takeaction'] = True
 406                         if delta >= 3*SPERDAY:
 407                                 record.data['action'] = ['email-againwaitforever']
 408                                 record.data['message'] = record.data['message_series'][2]
 409                                 record.data['time'] = current_time              # reset clock
 410                                 record.data['save-act-all'] = True
 411                         else:
 412                                 record.data['action'] = ['waitforever']
 413                                 record.data['message'] = None
 414                                 record.data['takeaction'] = False
 415                                 record.data['save-act-all'] = False
 416                                 #return None                    # don't send if there's no action
 417
 418                 else:
 419                         # There is no action to be taken, possibly b/c the stage has
 420                         # already been performed, but diagnose picked it up again.
 421                         # two cases,
 422                         #       1. stage is unknown, or
 423                         #       2. delta is not big enough to bump it to the next stage.
 424                         # TODO: figure out which. for now assume 2.
 425                         print "UNKNOWN stage for %s; nothing done" % self.hostname
 426                         record.data['action'] = ['unknown']
 427                         record.data['message'] = record.data['message_series'][0]
 428
 429                         record.data['email'] = TECH
 430                         record.data['action'] = ['noop']
 431                         record.data['message'] = record.data['message_series'][0]
 432                         record.data['stage'] = 'stage_actinoneweek'
 433                         record.data['time'] = current_time              # reset clock
 434                         record.data['takeaction'] = False
 435                         record.data['save-act-all'] = True
 436
 437                 print "%s" % record.data['log'],
 438                 print "%15s" % record.data['action']
 439                 return record
 440