add a link to both types of node history, status and data
[monitor.git] / web / MonitorWeb / monitorweb / controllers.py
1 import turbogears as tg
2 from turbogears import controllers, expose, flash, exception_handler
3 from turbogears import widgets
4 from cherrypy import request, response
5 import cherrypy
6 # from monitorweb import model
7 # import logging
8 # log = logging.getLogger("monitorweb.controllers")
9 import re
10 from monitor.database.info.model import *
11 #from monitor.database.zabbixapi.model import *
12 #from monitor.database.dborm import zab_session as session
13 #from monitor.database.dborm import zab_metadata as metadata
14 from monitor_xmlrpc import MonitorXmlrpcServer
15
16 from monitor import reboot
17 from monitor import scanapi
18 import time
19
20 from monitor.wrapper.plccache import plcdb_hn2lb as site_hn2lb
21
22 from monitorweb.templates.links import *
23
24
25 # make it easier group objects without invoking the elixir auto-write feature.
26 class aggregate: pass
27
28
29 def query_to_dict(query):
30         """ take a url query string and chop it up """
31         val = {}
32         query_fields = query.split('&')
33         for f in query_fields:
34                 (k,v) = urllib.splitvalue(f)
35                 val[k] = v
36
37         return val
38
39 def format_ports(data, pcumodel=None):
40         retval = []
41         filtered_length=0
42
43         if pcumodel:
44                 supported_ports=reboot.model_to_object(pcumodel).supported_ports
45         else:
46                 # ports of a production node
47                 supported_ports=[22,80,806]
48
49         if data and len(data.keys()) > 0 :
50                 for port in supported_ports:
51                         try:
52                                 state = data[str(port)]
53                         except:
54                                 state = "unknown"
55
56                         if state == "filtered":
57                                 filtered_length += 1
58                                 
59                         retval.append( (port, state) )
60
61         if retval == []: 
62                 retval = [( "Closed/Filtered", "state" )]
63
64         if filtered_length == len(supported_ports):
65                 retval = [( "All Filtered", "state" )]
66
67         return retval
68
69 def format_pcu_shortstatus(pcu):
70         status = "error"
71         if pcu:
72                 if pcu.reboot_trial_status == str(0):
73                         status = "Ok"
74                 elif pcu.reboot_trial_status == "NetDown" or pcu.reboot_trial_status == "Not_Run":
75                         status = pcu.reboot_trial_status
76                 else:
77                         status = "error"
78
79         return status
80
81 def prep_pcu_for_display(pcu):
82         agg = aggregate()
83         agg.pcu = pcu 
84                 
85         try:
86                 agg.loginbase = PlcSite.query.get(pcu.plc_pcu_stats['site_id']).plc_site_stats['login_base']
87         except:
88                 agg.loginbase = "unknown"
89
90         agg.ports = format_ports(pcu.port_status, pcu.plc_pcu_stats['model'])
91         agg.status = format_pcu_shortstatus(pcu)
92
93         #print pcu.entry_complete
94         agg.entry_complete_str = pcu.entry_complete
95         #pcu.entry_complete_str += "".join([ f[0] for f in pcu.entry_complete.split() ])
96         if pcu.dns_status == "NOHOSTNAME":
97                 agg.dns_short_status = 'NoHost'
98         elif pcu.dns_status == "DNS-OK":
99                 agg.dns_short_status = 'Ok'
100         elif pcu.dns_status == "DNS-NOENTRY":
101                 agg.dns_short_status = 'NoEntry'
102         elif pcu.dns_status == "NO-DNS-OR-IP":
103                 agg.dns_short_status = 'NoHostOrIP'
104         elif pcu.dns_status == "DNS-MISMATCH":
105                 agg.dns_short_status = 'Mismatch'
106         return agg
107
108 class NodeWidget(widgets.Widget):
109         pass
110
111 def prep_node_for_display(node, pcuhash=None, preppcu=True, asofdate=None):
112         agg = aggregate()
113         agg.node = node
114
115         if node.plc_pcuid and preppcu:
116                 if pcuhash:
117                         pcu = pcuhash[node.plc_pcuid]
118                 else:
119                         pcu = FindbadPCURecord.get_latest_by(plc_pcuid=node.plc_pcuid)
120
121                 if pcu:
122                         agg.pcu_status = pcu.reboot_trial_status
123                         agg.pcu_short_status = format_pcu_shortstatus(pcu)
124                         agg.pcu = prep_pcu_for_display(pcu)
125                 else:
126                         agg.pcu_short_status = "none"
127                         agg.pcu_status = "nodata"
128                         agg.pcu = None
129
130         else:
131                 agg.pcu_status = "nopcu"
132                 agg.pcu_short_status = "none"
133                 agg.pcu = None
134
135
136         if node.kernel_version:
137                 agg.kernel = node.kernel_version.split()[2]
138         else:
139                 agg.kernel = ""
140
141         try:
142                 agg.loginbase = PlcSite.query.get(node.plc_node_stats['site_id']).plc_site_stats['login_base']
143         except:
144                 agg.loginbase = "unknown"
145
146         if agg.loginbase:
147                 agg.site = HistorySiteRecord.by_loginbase(agg.loginbase)
148
149                 if asofdate:
150                         agg.site = agg.site.get_as_of(asofdate)
151
152                 if agg.site is None:
153                         # TODO: need a cleaner fix for this...
154                         agg.site = HistorySiteRecord.by_loginbase("pl")
155                         if not agg.site:
156                                 agg.site = HistorySiteRecord.by_loginbase("ple")
157
158         agg.history = HistoryNodeRecord.by_hostname(node.hostname)
159         if asofdate:
160                 agg.history = agg.history.get_as_of(asofdate)
161
162         agg.ports = format_ports(node.port_status)
163
164         try:
165                 exists = node.plc_node_stats['last_contact']
166         except:
167                 # TODO: this should not assign to the fb object!
168                 node.plc_node_stats = {'last_contact' : None}
169         
170         return agg
171
172
173 class Root(controllers.RootController, MonitorXmlrpcServer):
174         @expose(template="monitorweb.templates.welcome")
175         def index(self):
176                 # log.debug("Happy TurboGears Controller Responding For Duty")
177                 flash("Welcome To MyOps!")
178                 return dict(now=time.ctime())
179
180         @expose(template="monitorweb.templates.nodelist")
181         def node2(self, filter=None):
182                 nhquery = HistoryNodeRecord.query.all()
183                 query = []
184                 for nh in nhquery:
185                         if filter:
186                                 if nh.status == filter:
187                                         query.append(nh)
188                         else:
189                                 query.append(nh)
190
191                 rquery=[]
192                 for q in query:
193                         fb = FindbadNodeRecord.get_latest_by(hostname=q.hostname)
194                         agg = prep_node_for_display(fb)
195                         rquery.append(agg)
196
197                 #fbquery = FindbadNodeRecord.get_all_latest()
198                 #fbpcus = FindbadPCURecord.get_all_latest()
199                 #def fbtohash(fbpculist):
200                 #       h = {}
201                 #       for p in fbpculist:
202                 #               h[p.plc_pcuid] = p
203 #
204 #               pcuhash = fbtohash(fbpcus)
205
206 #               query = []
207 #               for node in fbquery:
208 #                       # NOTE: reformat some fields.
209 #                       agg = prep_node_for_display(node, pcuhash)
210 #                       if not agg.history:
211 #                               continue
212 #
213 #                       if filter:
214 #                               if agg.history.status == filter:
215 #                                       query.append(agg)
216 #                       else:
217 #                               query.append(agg)
218                                 
219                 widget = NodeWidget(template='monitorweb.templates.node_template')
220                 return dict(now=time.ctime(), query=rquery, nodewidget=widget)
221
222         @expose(template="monitorweb.templates.nodelist")
223         def node(self, filter='boot'):
224                 print "NODE------------------"
225                 print "befor-len: ", len( [ i for i in session] )
226                 session.flush(); session.clear()
227                 print "after-len: ", len( [ i for i in session] )
228                 fbquery = FindbadNodeRecord.get_all_latest()
229                 query = []
230                 filtercount = {'down' : 0, 'boot': 0, 'debug' : 0, 'diagnose' : 0, 'disabled': 0, 
231                                                 'neverboot' : 0, 'pending' : 0, 'all' : 0, None : 0}
232                 for node in fbquery:
233                         # NOTE: reformat some fields.
234                         agg = prep_node_for_display(node)
235
236                         if not agg.history:
237                                 continue
238
239                         if agg.history.status in ['down', 'offline']:
240                                 if node.plc_node_stats and node.plc_node_stats['last_contact'] != None:
241                                         filtercount['down'] += 1
242                                 else:
243                                         filtercount['neverboot'] += 1
244                         elif agg.history.status in ['good', 'online']:
245                                 filtercount['boot'] += 1
246                         elif agg.history.status in ['debug', 'monitordebug']:
247                                 filtercount['debug'] += 1
248                         else:
249                                 if filtercount.has_key(agg.history.status):
250                                         filtercount[agg.history.status] += 1
251                                 
252
253                         # NOTE: apply filter
254                         if filter == "neverboot":
255                                 if not node.plc_node_stats or node.plc_node_stats['last_contact'] == None:
256                                         query.append(agg)
257                         elif filter == "all":
258                                 query.append(agg)
259                         elif filter == agg.history.status:
260                                 query.append(agg)
261                         elif filter == 'boot':
262                                 query.append(agg)
263
264                                 
265                 widget = NodeWidget(template='monitorweb.templates.node_template')
266                 return dict(now=time.ctime(), query=query, fc=filtercount, nodewidget=widget)
267         
268         def nodeaction_handler(self, tg_exceptions=None):
269                 """Handle any kind of error."""
270                 print "NODEACTION_HANDLER------------------"
271
272                 if 'pcuid' in request.params:
273                         pcuid = request.params['pcuid']
274                 else:
275                         refurl = request.headers.get("Referer",link("pcu"))
276                         print refurl
277
278                         # TODO: do this more intelligently...
279                         uri_fields = urllib.splitquery(refurl)
280                         if uri_fields[1] is not None:
281                                 val = query_to_dict(uri_fields[1])
282                                 if 'pcuid' in val:
283                                         pcuid = val['pcuid']
284                                 elif 'hostname' in val:
285                                         pcuid = FindbadNodeRecord.get_latest_by(hostname=val['hostname']).plc_pcuid
286                                 else:
287                                         pcuid=None
288                         else:
289                                 pcuid=None
290
291                 cherry_trail = cherrypy._cputil.get_object_trail()
292                 for i in cherry_trail:
293                         print "trail: ", i
294
295                 print pcuid
296                 return self.pcuview(None, pcuid, **dict(exceptions=tg_exceptions))
297
298         def nodeaction(self, **data):
299                 print "NODEACTION------------------"
300                 for item in data.keys():
301                         print "%s %s" % ( item, data[item] )
302
303                 if 'hostname' in data:
304                         hostname = data['hostname']
305                 else:
306                         flash("No hostname given in submitted data")
307                         return
308
309                 if 'submit' in data or 'type' in data:
310                         try:
311                                 action = data['submit']
312                         except:
313                                 action = data['type']
314                 else:
315                         flash("No submit action given in submitted data")
316                         return
317
318                 if action == "Reboot":
319                         print "REBOOT: %s" % hostname
320                         ret = reboot.reboot_str(str(hostname))
321                         print ret
322                         if ret: raise RuntimeError("Error using PCU: " + str(ret))
323                         flash("Reboot appeared to work.  Allow at most 5 minutes.  Then run ExternalScan to check current status.")
324
325                 elif action == "ExternalScan":
326                         scanapi.externalprobe(str(hostname))
327                         flash("External Scan Successful!")
328                 elif action == "InternalScan":
329                         scanapi.internalprobe(str(hostname))
330                         flash("Internal Scan Successful!")
331                 else:
332                         # unknown action
333                         raise RuntimeError("Unknown action given")
334                 return
335
336         # TODO: add form validation
337         @expose(template="monitorweb.templates.pcuview")
338         @exception_handler(nodeaction_handler,"isinstance(tg_exceptions,RuntimeError)")
339         def pcuview(self, loginbase=None, pcuid=None, hostname=None, since=20, **data):
340                 session.flush(); session.clear()
341                 sitequery=[]
342                 pcuquery=[]
343                 nodequery=[]
344                 actions=[]
345                 exceptions = None
346
347                 try: since = int(since)
348                 except: since = 7
349
350                 for key in data:
351                         print key, data[key]
352
353                 if 'submit' in data.keys() or 'type' in data.keys():
354                         if hostname: data['hostname'] = hostname
355                         self.nodeaction(**data)
356                 if 'exceptions' in data:
357                         exceptions = data['exceptions']
358
359                 if 'query' in data:
360                         obj = data['query']
361                         if len(obj.split(".")) > 1: hostname = obj
362                         else: loginbase=obj
363
364                 if pcuid:
365                         print "pcuid: %s" % pcuid
366                         pcu = FindbadPCURecord.get_latest_by(plc_pcuid=pcuid)
367                         loginbase = PlcSite.query.get(pcu.plc_pcu_stats['site_id']).plc_site_stats['login_base']
368
369                 if hostname:
370                         node = FindbadNodeRecord.get_latest_by(hostname=hostname)
371                         loginbase = PlcSite.query.get(node.plc_node_stats['site_id']).plc_site_stats['login_base']
372
373                 if loginbase:
374                         actions = ActionRecord.query.filter_by(loginbase=loginbase
375                                                         ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
376                                                         ).order_by(ActionRecord.date_created.desc())
377                         actions = [ a for a in actions ]
378                         sitequery = [HistorySiteRecord.by_loginbase(loginbase)]
379                         pcus = {}
380                         for node in FindbadNodeRecord.query.filter_by(loginbase=loginbase):
381                                         # NOTE: reformat some fields.
382                                         agg = prep_node_for_display(node)
383                                         nodequery += [agg]
384                                         if agg.pcu: #.pcu.plc_pcuid:    # not None
385                                                 #pcu = FindbadPCURecord.get_latest_by(plc_pcuid=agg.plc_pcuid)
386                                                 #prep_pcu_for_display(pcu)
387                                                 pcus[agg.pcu.pcu.plc_pcuid] = agg.pcu
388
389                         for pcuid_key in pcus:
390                                 pcuquery += [pcus[pcuid_key]]
391
392                 return dict(sitequery=sitequery, pcuquery=pcuquery, nodequery=nodequery, actions=actions, since=since, exceptions=exceptions)
393
394         @expose(template="monitorweb.templates.pcuhistory")
395         def pcuhistory(self, pcu_id=None):
396                 query = []
397                 if pcu_id:
398                         fbnode = HistoryPCURecord.get_by(plc_pcuid=pcu_id)
399                         l = fbnode.versions[-100:]
400                         l.reverse()
401                         for pcu in l:
402                                 #prep_node_for_display(node)
403                                 query.append(pcu)
404
405                 return dict(query=query, pcu_id=pcu_id)
406
407         @expose(template="monitorweb.templates.nodescanhistory")
408         def nodescanhistory(self, hostname=None, length=10):
409                 try: length = int(length)
410                 except: length = 10
411
412                 fbnode = FindbadNodeRecord.get_by(hostname=hostname)
413                 # TODO: add links for earlier history if desired.
414                 l = fbnode.versions[-length:]
415                 l.reverse()
416                 query=[]
417                 for node in l:
418                         agg = prep_node_for_display(node, pcuhash=None, preppcu=False, asofdate=node.timestamp)
419                         query.append(agg)
420
421                 return dict(query=query, hostname=hostname)
422
423         @expose(template="monitorweb.templates.nodehistory")
424         def nodehistory(self, hostname=None):
425                 query = []
426                 if hostname:
427                         fbnode = HistoryNodeRecord.get_by(hostname=hostname)
428                         l = fbnode.versions[-100:]
429                         l.reverse()
430                         for node in l:
431                                 #prep_node_for_display(node)
432                                 query.append(node)
433
434                 return dict(query=query, hostname=hostname)
435
436         @expose(template="monitorweb.templates.sitehistory")
437         def sitehistory(self, loginbase=None):
438                 query = []
439                 if loginbase:
440                         fbsite = HistorySiteRecord.get_by(loginbase=loginbase)
441                         # TODO: add links for earlier history if desired.
442                         l = fbsite.versions[-100:]
443                         l.reverse()
444                         for site in l:
445                                 query.append(site)
446                 return dict(query=query, loginbase=loginbase)
447
448
449         @expose(template="monitorweb.templates.pculist")
450         def pcu(self, filter='all'):
451                 print "PCUVIEW------------------"
452                 print "befor-len: ", len( [ i for i in session] )
453                 session.flush(); session.clear()
454                 print "after-len: ", len( [ i for i in session] )
455                 fbquery = FindbadPCURecord.get_all_latest()
456                 query = []
457                 filtercount = {'ok' : 0, 'NetDown': 0, 'Not_Run' : 0, 'pending' : 0, 'all' : 0}
458                 for node in fbquery:
459
460                         # NOTE: count filter
461                         if node.reboot_trial_status == str(0):
462                                 filtercount['ok'] += 1
463                         elif node.reboot_trial_status == 'NetDown' or node.reboot_trial_status == 'Not_Run':
464                                 filtercount[node.reboot_trial_status] += 1
465                         else:
466                                 filtercount['pending'] += 1
467
468                         pcuagg = prep_pcu_for_display(node)
469
470                         # NOTE: apply filter
471                         if filter == "all":
472                                 query.append(pcuagg)
473                         elif filter == "ok" and node.reboot_trial_status == str(0):
474                                 query.append(pcuagg)
475                         elif filter == node.reboot_trial_status:
476                                 query.append(pcuagg)
477                         elif filter == "pending":
478                                 # TODO: look in message logs...
479                                 if node.reboot_trial_status != str(0) and \
480                                         node.reboot_trial_status != 'NetDown' and \
481                                         node.reboot_trial_status != 'Not_Run':
482
483                                         query.append(pcuagg)
484                                 
485                 return dict(query=query, fc=filtercount)
486
487         @expose(template="monitorweb.templates.sitelist")
488         def site(self, filter='all'):
489                 print "SITE------------------"
490                 print "befor-len: ", len( [ i for i in session] )
491                 session.flush(); session.clear()
492                 print "after-len: ", len( [ i for i in session] )
493                 filtercount = {'good' : 0, 'down': 0, 'online':0, 'offline' : 0, 'new' : 0, 'pending' : 0, 'all' : 0}
494                 fbquery = HistorySiteRecord.query.all()
495                 query = []
496                 for site in fbquery:
497                         # count filter
498                         filtercount['all'] += 1
499                         if site.new and site.slices_used == 0 and not site.enabled:
500                                 filtercount['new'] += 1
501                         elif not site.enabled:
502                                 filtercount['pending'] += 1
503                         elif site.status in ['good', 'online']:
504                                 filtercount['good'] += 1
505                         elif site.status in ['down', 'offline']:
506                                 filtercount['down'] += 1
507
508                         # apply filter
509                         if filter == "all":
510                                 query.append(site)
511                         elif filter == 'new' and site.new and site.slices_used == 0 and not site.enabled:
512                                 query.append(site)
513                         elif filter == "pending" and not site.enabled:
514                                 query.append(site)
515                         elif filter == 'good' and site.status in ['good', 'online']:
516                                 query.append(site)
517                         elif filter == 'down' and site.status in ['down', 'offline']:
518                                 query.append(site)
519                                 
520                 return dict(query=query, fc=filtercount)
521         @expose(template="monitorweb.templates.sitesummary")
522         def sitesummary(self, loginbase="princeton"):
523                 nodequery = []
524                 for node in FindbadNodeRecord.query.filter_by(loginbase=loginbase):
525                         agg = prep_node_for_display(node)
526                         nodequery += [agg]
527                 
528                 return dict(nodequery=nodequery, loginbase=loginbase)
529
530         @expose(template="monitorweb.templates.summary")
531         def summary(self, since=7):
532                 sumdata = {}
533                 sumdata['nodes'] = {}
534                 sumdata['sites'] = {}
535                 sumdata['pcus'] = {}
536
537                 def summarize(query, type):
538                         for o in query:
539                                 if o.status not in sumdata[type]:
540                                         sumdata[type][o.status] = 0
541                                 sumdata[type][o.status] += 1
542
543                 fbquery = HistorySiteRecord.query.all()
544                 summarize(fbquery, 'sites')
545                 fbquery = HistoryPCURecord.query.all()
546                 summarize(fbquery, 'pcus')
547                 fbquery = HistoryNodeRecord.query.all()
548                 summarize(fbquery, 'nodes')
549
550                 if 'monitordebug' in sumdata['nodes']:
551                         d = sumdata['nodes']['monitordebug']
552                         del sumdata['nodes']['monitordebug']
553                         sumdata['nodes']['failboot'] = d
554                 
555                 return dict(sumdata=sumdata, setorder=['good', 'offline', 'down', 'online']) 
556
557         @expose(template="monitorweb.templates.actionsummary")
558         def actionsummary(self, since=7):
559                 from monitor.wrapper.emailTxt import mailtxt
560
561                 types = filter(lambda x: 'notice' in x, dir(mailtxt))
562                 results = {}
563
564                 try: since = int(since)
565                 except: since = 7
566
567                 for  t in types:
568                         acts = ActionRecord.query.filter(ActionRecord.action_type==t
569                                         ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since))
570                         results[t] = acts.count()
571                 return dict(results=results)
572
573         @expose(template="monitorweb.templates.actionlist")
574         def actionlist(self, action_type='down_notice', since=7, loginbase=None):
575
576                 try: since = int(since)
577                 except: since = 7
578
579                 if loginbase:
580                         acts = ActionRecord.query.filter_by(loginbase=loginbase
581                                 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
582                                 ).order_by(ActionRecord.date_created.desc())
583                 else:
584                         acts = ActionRecord.query.filter(ActionRecord.action_type==action_type
585                                 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
586                                 ).order_by(ActionRecord.date_created.desc())
587                 query = [ a for a in acts ]
588                 
589                 return dict(actions=query, action_type=action_type, since=since)