3 from monitor import database
14 def gethostlist(hostlist_file):
15 return util.file.getListFromFile(hostlist_file)
17 def array_to_priority_map(array):
18 """ Create a mapping where each entry of array is given a priority equal
19 to its position in the array. This is useful for subsequent use in the
28 def cmpValMap(v1, v2, map):
29 if v1 in map and v2 in map and map[v1] < map[v2]:
31 elif v1 in map and v2 in map and map[v1] > map[v2]:
33 elif v1 in map and v2 in map:
36 raise Exception("No index %s or %s in map" % (v1, v2))
38 def cmpCategoryVal(v1, v2):
39 # Terrible hack to manage migration to no more 'ALPHA' states.
40 if v1 == 'ALPHA': v1 = "PROD"
41 if v2 == 'ALPHA': v2 = "PROD"
42 #map = array_to_priority_map([ None, 'PROD', 'ALPHA', 'OLDBOOTCD', 'UNKNOWN', 'FORCED', 'ERROR', ])
43 map = array_to_priority_map([ None, 'ALPHA', 'PROD', 'OLDBOOTCD', 'UNKNOWN', 'FORCED', 'ERROR', ])
44 return cmpValMap(v1,v2,map)
48 def __init__(self, hostname):
49 self.hostname = hostname
55 def previous_attempt(self):
57 def setValidMapping(self):
61 def __init__(self, key, valuepattern, action):
68 # connect one penalty to another, in a FSM diagram. After one
69 # condition/penalty is applied, move to the next phase.
72 #fb = database.dbLoad("findbad")
75 def __init__(self, ticket_id = None):
76 self.ticket_id = ticket_id
78 print "getting ticket status",
79 self.status = mailer.getTicketStatus(self.ticket_id)
82 def setTicketStatus(self, status):
83 mailer.setTicketStatus(self.ticket_id, status)
84 self.status = mailer.getTicketStatus(self.ticket_id)
87 def getTicketStatus(self):
89 self.status = mailer.getTicketStatus(self.ticket_id)
92 def closeTicket(self):
93 mailer.closeTicketViaRT(self.ticket_id, "Ticket CLOSED automatically by SiteAssist.")
95 def email(self, subject, body, to):
96 self.ticket_id = mailer.emailViaRT(subject, body, to, self.ticket_id)
99 class Message(object):
100 def __init__(self, subject, message, via_rt=True, ticket_id=None, **kwargs):
102 self.subject = subject
103 self.message = message
104 self.rt = RT(ticket_id)
108 return self.rt.email(self.subject, self.message, to)
110 return mailer.email(self.subject, self.message, to)
112 class Recent(object):
113 def __init__(self, withintime):
114 self.withintime = withintime
117 self.time = self.__getattribute__('time')
119 self.time = time.time()- 7*24*60*60
121 #self.time = time.time()
122 #self.action_taken = False
125 if self.time + self.withintime < time.time():
126 self.action_taken = False
128 if self.time + self.withintime > time.time() and self.action_taken:
133 def unsetRecent(self):
134 self.action_taken = False
135 self.time = time.time()
139 self.action_taken = True
140 self.time = time.time()
143 class PersistFlags(Recent):
144 def __new__(typ, id, *args, **kwargs):
152 pm = database.dbLoad(db)
154 database.dbDump(db, {})
155 pm = database.dbLoad(db)
160 obj = super(PersistFlags, typ).__new__(typ, *args, **kwargs)
161 for key in kwargs.keys():
162 obj.__setattr__(key, kwargs[key])
163 obj.time = time.time()
164 obj.action_taken = False
169 def __init__(self, id, withintime, **kwargs):
171 Recent.__init__(self, withintime)
174 pm = database.dbLoad(self.db)
176 database.dbDump(self.db, pm)
178 def resetFlag(self, name):
179 self.__setattr__(name, False)
181 def setFlag(self, name):
182 self.__setattr__(name, True)
184 def getFlag(self, name):
186 return self.__getattribute__(name)
188 self.__setattr__(name, False)
191 def resetRecentFlag(self, name):
195 def setRecentFlag(self, name):
199 def getRecentFlag(self, name):
200 # if recent and flag set -> true
203 return self.isRecent() & self.__getattribute__(name)
205 self.__setattr__(name, False)
208 def checkattr(self, name):
210 x = self.__getattribute__(name)
216 class PersistMessage(Message):
217 def __new__(typ, id, subject, message, via_rt, **kwargs):
221 db = "persistmessages"
224 pm = database.dbLoad(db)
226 database.dbDump(db, {})
227 pm = database.dbLoad(db)
231 #print "Using existing object"
234 #print "creating new object"
235 obj = super(PersistMessage, typ).__new__(typ, [id, subject, message, via_rt], **kwargs)
237 obj.actiontracker = Recent(1*60*60*24)
240 if 'ticket_id' in kwargs and kwargs['ticket_id'] is not None:
241 obj.ticket_id = kwargs['ticket_id']
246 def __init__(self, id, subject, message, via_rt=True, **kwargs):
247 print "initializing object: %s" % self.ticket_id
249 Message.__init__(self, subject, message, via_rt, self.ticket_id)
252 self.actiontracker.unsetRecent()
255 pm = database.dbLoad(self.db)
257 database.dbDump(self.db, pm)
260 if not self.actiontracker.isRecent():
261 self.ticket_id = Message.send(self, to)
262 self.actiontracker.setRecent()
265 # NOTE: only send a new message every week, regardless.
266 # NOTE: can cause thank-you messages to be lost, for instance when node comes back online within window.
267 print "Not sending to host b/c not within window of %s days" % (self.actiontracker.withintime // (60*60*24))
269 class MonitorMessage(object):
270 def __new__(typ, id, *args, **kwargs):
274 db = "monitormessages"
277 if 'reset' in kwargs and kwargs['reset'] == True:
278 database.dbDump(db, {})
279 pm = database.dbLoad(db)
281 database.dbDump(db, {})
282 pm = database.dbLoad(db)
286 print "Using existing object"
289 print "creating new object"
290 obj = super(object, typ).__new__(typ, id, *args, **kwargs)
292 obj.sp = PersistSitePenalty(id, 0)
297 def __init__(self, id, message):
301 class SitePenalty(object):
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) } )
310 #def __init__(self, index=0, **kwargs):
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 ]
318 self.index = self.index + 1
319 if self.index > len(SitePenalty.penalty_map)-1: self.index = len(SitePenalty.penalty_map)-1
323 self.index = self.index - 1
324 if self.index < 0: self.index = 0
327 def apply(self, host):
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)
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)
341 class PersistSitePenalty(SitePenalty):
342 def __new__(typ, id, index, **kwargs):
346 db = "persistpenalties"
349 if 'reset' in kwargs and kwargs['reset'] == True:
350 database.dbDump(db, {})
351 pm = database.dbLoad(db)
353 database.dbDump(db, {})
354 pm = database.dbLoad(db)
358 print "Using existing object"
361 print "creating new object"
362 obj = super(PersistSitePenalty, typ).__new__(typ, [index], **kwargs)
369 def __init__(self, id, index, **kwargs):
373 pm = database.dbLoad(self.db)
375 database.dbDump(self.db, pm)
380 Each host has a target set of attributes. Some may be set manually,
381 or others are set globally for the preferred target.
384 All nodes in the Alpha or Beta group would have constraints like:
385 [ { 'state' : 'BOOT', 'kernel' : '2.6.22' } ]
387 def __init__(self, constraints):
388 self.constraints = constraints
390 def verify(self, data):
392 self.constraints is a list of key, value pairs.
393 # [ {... : ...}==AND , ... , ... , ] == OR
396 for con in self.constraints:
397 #print "con: %s" % con
399 for key in con.keys():
400 #print "looking at key: %s" % key
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
408 con_or_true = con_or_true | con_and_true
412 class Record(object):
414 def __init__(self, hostname, data):
415 self.hostname = hostname
417 self.plcdb_hn2lb = database.dbLoad("plcdb_hn2lb")
418 self.loginbase = self.plcdb_hn2lb[self.hostname]
422 def stageIswaitforever(self):
423 if 'waitforever' in self.data['stage']:
429 category = self.data['category']
430 prev_category = self.data['prev_category']
431 #print "SEVERITY: ", category, prev_category
432 val = cmpCategoryVal(category, prev_category)
436 return self.severity() > 0
438 def end_record(self):
439 return node_end_record(self.hostname)
441 def reset_stage(self):
442 self.data['stage'] = 'findbad'
445 def getCategory(self):
446 return self.data['category'].lower()
449 return self.data['state'].lower()
451 def getDaysDown(cls, diag_record):
453 if diag_record['comonstats']['uptime'] != "null" and diag_record['comonstats']['uptime'] != "-1":
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)
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
466 diff = now - last_contact
467 daysdown = diff // (60*60*24)
469 getDaysDown = classmethod(getDaysDown)
471 def getStrDaysDown(cls, diag_record):
473 last_contact = diag_record['plcnode']['last_contact']
474 date_created = diag_record['plcnode']['date_created']
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
481 elif last_contact is None:
482 if date_created is not None:
484 diff = now - date_created
485 daysdown = diff // (60*60*24)
486 daysdown = "Never contacted PLC, created %s days ago" % daysdown
488 daysdown = "Never contacted PLC"
491 diff = now - last_contact
492 daysdown = diff // (60*60*24)
493 daysdown = "%s days down" % daysdown
495 getStrDaysDown = classmethod(getStrDaysDown)
497 #def getStrDaysDown(cls, diag_record):
498 # daysdown = cls.getDaysDown(diag_record)
500 # return "%d days down"%daysdown
501 # elif daysdown == -1:
502 # return "Never online"
504 # return "%d days up"% -daysdown
505 #getStrDaysDown = classmethod(getStrDaysDown)
507 def takeAction(self, index=0):
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
515 print "takeAction: increasing penalty for %s"%self.hostname
518 pp.apply(self.hostname)
521 def _format_diaginfo(self):
522 info = self.data['info']
523 print "FORMAT : STAGE: ", self.data['stage']
524 if self.data['stage'] == 'monitor-end-record':
525 if info[2] == "ALPHA": info = (info[0], info[1], "PROD")
526 hlist = " %s went from '%s' to '%s'\n" % (info[0], info[1], info[2])
528 hlist = " %s %s - %s\n" % (info[0], info[2], info[1]) #(node,ver,daysdn)
530 def saveAction(self):
531 if 'save-act-all' in self.data and self.data['save-act-all'] == True:
536 def getMessage(self, ticket_id=None):
537 self.data['args']['hostname'] = self.hostname
538 self.data['args']['loginbase'] = self.loginbase
539 self.data['args']['hostname_list'] = self._format_diaginfo()
540 #print self.data['message']
541 if self.data['message']:
542 message = PersistMessage(self.hostname,
543 self.data['message'][0] % self.data['args'],
544 self.data['message'][1] % self.data['args'],
545 True, db='monitor_persistmessages',
547 if self.data['stage'] == "improvement":
553 def getContacts(self):
554 roles = self.data['email']
556 if not config.mail and not config.debug and config.bcc:
558 if config.mail and config.debug:
564 contacts += [config.email]
566 #contacts += [TECHEMAIL % self.loginbase]
567 contacts += plc.getTechEmails(self.loginbase)
569 #contacts += [PIEMAIL % self.loginbase]
570 contacts += plc.getPIEmails(self.loginbase)
572 contacts += plc.getSliceUserEmails(self.loginbase)
573 slices = plc.slices(self.loginbase)
575 #for slice in slices:
576 # contacts += [SLICEMAIL % slice]
577 print "SLIC: %20s : %d slices" % (self.loginbase, len(slices))
579 print "SLIC: %20s : 0 slices" % self.loginbase
585 def __init__(self, hostname, target):
586 self.hostname = hostname
589 #if hostname in fb['nodes']:
590 # self.data = fb['nodes'][hostname]['values']
592 # raise Exception("Hostname not in scan database")
594 def stageIswaitforever(self):
595 if 'waitforever' in self.data['stage']:
601 category = self.data['category']
602 prev_category = self.data['prev_category']
603 print "IMPROVED: ", category, prev_category
604 val = cmpCategoryVal(category, prev_category)
608 return self.severity() > 0
610 def end_record(self):
611 return node_end_record(self.hostname)
613 def reset_stage(self):
614 self.data['stage'] = 'findbad'
617 def open_tickets(self):
618 if self.ticket and self.ticket.status['status'] == 'open':
621 def setIntrospect(self):
624 def email_notice(self):
625 message = self._get_message_for_condition()
626 message.send(self._get_contacts_for_condition())
628 def close_ticket(self):
630 self.ticket.closeTicket()
632 def exempt_from_penalties(self):
633 bl = database.dbLoad("l_blacklist")
634 return self.hostname in bl
638 def escellate_penalty(self):
640 def reduce_penalty(self):
645 return self.target.verify(self.data)
647 def _get_condition(self):
648 return self.data['category'].lower()
650 def _get_stage(self):
653 "secondnotice_noslicecreation"
654 "thirdnotice_disableslices"
656 delta = current_time - self.data['time']
658 def _get_message_for_condition(self):
660 def _get_contacts_for_condition(self):
663 class Action(MonRecord):
664 def __init__(self, host, data):
666 MonRecord.__init__(self, data)
669 def deltaDays(self, delta):
670 t = datetime.fromtimestamp(self.__dict__['time'])
671 d = t + timedelta(delta)
672 self.__dict__['time'] = time.mktime(d.timetuple())
674 def node_end_record(node):
675 act_all = database.dbLoad("act_all")
676 if node not in act_all:
680 if len(act_all[node]) == 0:
684 pm = database.dbLoad("monitor_persistmessages")
689 print "deleting node record"
691 database.dbDump("monitor_persistmessages", pm)
693 a = Action(node, act_all[node][0])
695 a.delField('found_rt_ticket')
696 a.delField('second-mail-at-oneweek')
697 a.delField('second-mail-at-twoweeks')
698 a.delField('first-found')
700 rec['action'] = ["close_rt"]
701 rec['category'] = "ALPHA" # assume that it's up...
702 rec['stage'] = "monitor-end-record"
703 rec['ticket_id'] = None
704 rec['time'] = time.time() - 7*60*60*24
705 act_all[node].insert(0,rec)
706 database.dbDump("act_all", act_all)
710 if __name__ == "__main__":
712 #r.email("test", "body of test message", ['database@cs.princeton.edu'])
713 #from emailTxt import mailtxt
715 #database.dbDump("persistmessages", {});
716 #args = {'url_list': 'http://www.planet-lab.org/bootcds/planet1.usb\n','hostname': 'planet1','hostname_list': ' blahblah - days down\n'}
717 #m = PersistMessage("blue", "test 1", mailtxt.newdown_one[1] % args, True)
718 #m.send(['soltesz@cs.utk.edu'])
719 #m = PersistMessage("blue", "test 1 - part 2", mailtxt.newalphacd_one[1] % args, True)
720 # TRICK timer to thinking some time has passed.
721 #m.actiontracker.time = time.time() - 6*60*60*24
722 #m.send(['soltesz@cs.utk.edu'])