1 import turbogears as tg
2 from turbogears import controllers, expose, flash, exception_handler, redirect
3 from turbogears import widgets
4 from cherrypy import request, response
6 # from monitorweb import model
8 # log = logging.getLogger("monitorweb.controllers")
11 from monitor.database.info.model import *
12 #from monitor.database.zabbixapi.model import *
13 from monitor_xmlrpc import MonitorXmlrpcServer
15 from monitor import reboot
16 from monitor import scanapi
19 from monitor.wrapper.plccache import plcdb_hn2lb as site_hn2lb
21 from monitorweb.templates.links import *
23 class ObjectQueryFields(widgets.WidgetsList):
24 """The WidgetsList defines the fields of the form."""
29 class NodeQueryFields(widgets.WidgetsList):
30 """The WidgetsList defines the fields of the form."""
32 object = widgets.RadioButtonList(label="Query Type", options=[('nodes', 'All Nodes'),
33 ('nodehistory', 'Single Node History'),
34 #('sites', 'All Sites'),
35 #('sitehistory', 'Single Site History'),
37 nodehistory_hostname = widgets.TextField(label="Hostname Node History", attrs={'size':30})
39 hostname = widgets.CheckBox(label="Hostname")
40 firewall = widgets.CheckBox(label="Firewall?")
41 ssh_status = widgets.CheckBox(label="SSH Status")
42 ssh_error = widgets.CheckBox(label="SSH Errors")
43 dns_status = widgets.CheckBox(label="DNS Status")
44 nm_status = widgets.CheckBox(label="NM Status")
45 princeton_comon_dir = widgets.CheckBox(label="CoMon Dir")
46 princeton_comon_running = widgets.CheckBox(label="CoMon Running")
47 princeton_comon_procs = widgets.CheckBox(label="CoMon Processes")
48 external_dns_status = widgets.CheckBox(label="Hostname Resolves?")
49 kernel_version = widgets.CheckBox(label="Kernel")
50 bootcd_version = widgets.CheckBox(label="BootCD")
51 observed_status = widgets.CheckBox(label="Observed Status")
52 port_status = widgets.CheckBox(label="Port Status")
53 rpms = widgets.CheckBox(label="RPM")
54 rpmvalue = widgets.TextField(label="RPM Pattern")
56 class QueryForm(widgets.TableForm):
58 <form xmlns:py="http://purl.org/kid/ns#"
65 <div py:for="field in hidden_fields"
66 py:replace="field.display(value_for(field), **params_for(field))"
68 <table border="0" cellspacing="0" cellpadding="2" py:attrs="table_attrs">
69 <tr py:for="i, field in enumerate(fields)"
70 class="${i%2 and 'odd' or 'even'}"
73 <label class="fieldlabel" for="${field.field_id}" py:content="field.label" />
76 <span py:replace="field.display(value_for(field), **params_for(field))" />
77 <span py:if="error_for(field)" class="fielderror" py:content="error_for(field)" />
78 <span py:if="field.help_text" class="fieldhelp" py:content="field.help_text" />
83 <td py:content="submit.display(submit_text)" />
89 def getNodeQueryForm():
90 return QueryForm(fields=NodeQueryFields(), action="query")
92 # make it easier group objects without invoking the elixir auto-write feature.
96 def query_to_dict(query):
97 """ take a url query string and chop it up """
99 query_fields = query.split('&')
100 for f in query_fields:
101 (k,v) = urllib.splitvalue(f)
106 def format_ports(data, pcumodel=None):
111 supported_ports=reboot.model_to_object(pcumodel).supported_ports
113 # ports of a production node
114 supported_ports=[22,80,806]
116 if data and len(data.keys()) > 0 :
117 for port in supported_ports:
119 state = data[str(port)]
123 if state == "filtered":
126 retval.append( (port, state) )
129 retval = [( "Closed/Filtered", "state" )]
131 if filtered_length == len(supported_ports):
132 retval = [( "All Filtered", "state" )]
136 def format_pcu_shortstatus(pcu):
139 if pcu.reboot_trial_status == str(0):
141 elif pcu.reboot_trial_status == "NetDown" or pcu.reboot_trial_status == "Not_Run":
142 status = pcu.reboot_trial_status
148 def prep_pcu_for_display(pcu):
153 agg.loginbase = PlcSite.query.get(pcu.plc_pcu_stats['site_id']).plc_site_stats['login_base']
155 agg.loginbase = "unknown"
157 agg.pcuhist = HistoryPCURecord.query.get(pcu.plc_pcuid)
159 agg.ports = format_ports(pcu.port_status, pcu.plc_pcu_stats['model'])
160 agg.status = format_pcu_shortstatus(pcu)
162 #print pcu.entry_complete
163 agg.entry_complete_str = pcu.entry_complete
164 #pcu.entry_complete_str += "".join([ f[0] for f in pcu.entry_complete.split() ])
165 if pcu.dns_status == "NOHOSTNAME":
166 agg.dns_short_status = 'NoHost'
167 elif pcu.dns_status == "DNS-OK":
168 agg.dns_short_status = 'Ok'
169 elif pcu.dns_status == "DNS-NOENTRY":
170 agg.dns_short_status = 'NoEntry'
171 elif pcu.dns_status == "NO-DNS-OR-IP":
172 agg.dns_short_status = 'NoHostOrIP'
173 elif pcu.dns_status == "DNS-MISMATCH":
174 agg.dns_short_status = 'Mismatch'
177 class NodeWidget(widgets.Widget):
180 def prep_nodehist(node):
183 agg.loginbase = "unknown"
185 agg.loginbase = PlcSite.query.get(node.plc_siteid).plc_site_stats['login_base']
187 agg.loginbase = "exception"
192 def prep_node_for_display(node, pcuhash=None, preppcu=True, asofdate=None):
196 if node.plc_pcuid and preppcu:
198 pcu = pcuhash[node.plc_pcuid]
200 pcu = FindbadPCURecord.get_latest_by(plc_pcuid=node.plc_pcuid)
203 agg.pcu_status = pcu.reboot_trial_status
204 agg.pcu_short_status = format_pcu_shortstatus(pcu)
205 agg.pcu = prep_pcu_for_display(pcu)
207 agg.pcu_short_status = "none"
208 agg.pcu_status = "nodata"
212 agg.pcu_status = "nopcu"
213 agg.pcu_short_status = "none"
217 if node.kernel_version:
218 agg.kernel = node.kernel_version.split()[2]
223 agg.loginbase = PlcSite.query.get(node.plc_node_stats['site_id']).plc_site_stats['login_base']
225 agg.loginbase = "unknown"
228 agg.site = HistorySiteRecord.by_loginbase(agg.loginbase)
231 agg.site = agg.site.get_as_of(asofdate)
234 # TODO: need a cleaner fix for this...
235 agg.site = HistorySiteRecord.by_loginbase("pl")
237 agg.site = HistorySiteRecord.by_loginbase("ple")
239 agg.history = HistoryNodeRecord.by_hostname(node.hostname)
241 agg.history = agg.history.get_as_of(asofdate)
243 agg.ports = format_ports(node.port_status)
246 exists = node.plc_node_stats['last_contact']
248 # TODO: this should not assign to the fb object!
249 node.plc_node_stats = {'last_contact' : None}
254 class Root(controllers.RootController, MonitorXmlrpcServer):
255 @expose(template="monitorweb.templates.welcome")
257 # log.debug("Happy TurboGears Controller Responding For Duty")
258 flash("Welcome To MyOps!")
259 return dict(now=time.ctime())
261 @expose(template="monitorweb.templates.nodelist", allow_json=True)
262 def node3(self, filter=None):
263 nhquery = HistoryNodeRecord.query.all()
267 if nh.status == filter:
274 fb = FindbadNodeRecord.get_latest_by(hostname=q.hostname)
277 return dict(now=time.ctime(), query=rquery)
279 def node_query(self, filter):
280 nhquery = HistoryNodeRecord.query.all()
284 if nh.status == filter:
291 fb = FindbadNodeRecord.get_latest_by(hostname=q.hostname)
292 agg = prep_node_for_display(fb)
296 @expose("cheetah:monitorweb.templates.nodelist_plain", as_format="plain",
297 accept_format="text/plain", content_type="text/plain")
298 @expose(template="monitorweb.templates.nodelist", allow_json=True)
299 def node2(self, filter=None):
300 rquery=self.node_query(filter)
301 widget = NodeWidget(template='monitorweb.templates.node_template')
302 return dict(now=time.ctime(), query=rquery, nodewidget=widget)
304 @expose("cheetah:monitorweb.templates.query_plain", as_format="plain",
305 accept_format="text/plain", content_type="text/plain")
306 @expose(template="monitorweb.templates.query", allow_json=True)
307 def query(self, **data):
315 if 'object' in data and data['object'] == "nodes":
316 fbquery = FindbadNodeRecord.get_all_latest()
317 elif 'object' in data and data['object'] == "nodehistory":
318 hostname = data['nodehistory_hostname']
319 data['date_checked'] = 'date_checked'
320 fbrecord = FindbadNodeRecord.get_by(hostname=hostname)
321 fbquery = fbrecord.versions[-500:]
325 # NOTE: reformat some fields.
326 if type(node) is not type(FindbadNodeRecord):
327 agg = node.__dict__.copy()
330 agg.update(agg['plc_node_stats'])
331 if agg['kernel_version']:
332 agg['kernel_version'] = agg['kernel_version'].split()[2]
333 if 'rpmvalue' in data and 'rpms' in data:
335 rpm_list = agg['rpms'].split()
336 rpm_list = filter(lambda x: data['rpmvalue'] in x, rpm_list)
337 agg['rpms'] = " ".join(rpm_list)
345 del fields['rpmvalue']
346 del fields['nodehistory_hostname']
348 return dict(now=time.ctime(), query=query, fields=fields, data=data, queryform=getNodeQueryForm())
350 @expose(template="monitorweb.templates.nodefast", allow_json=True)
351 def node(self, filter=None):
352 nhquery = HistoryNodeRecord.query.all()
356 if nh.status == filter:
357 agg = prep_nodehist(nh)
360 agg = prep_nodehist(nh)
363 return dict(now=time.ctime(), query=query)
365 @expose(template="monitorweb.templates.nodelist")
366 def nodeslow(self, filter='boot'):
367 print "NODE------------------"
368 print "befor-len: ", len( [ i for i in session] )
369 session.flush(); session.clear()
370 print "after-len: ", len( [ i for i in session] )
371 fbquery = FindbadNodeRecord.get_all_latest()
373 filtercount = {'down' : 0, 'boot': 0, 'debug' : 0, 'diagnose' : 0, 'disabled': 0,
374 'neverboot' : 0, 'pending' : 0, 'all' : 0, None : 0}
376 # NOTE: reformat some fields.
377 agg = prep_node_for_display(node)
382 if agg.history.status in ['down', 'offline']:
383 if node.plc_node_stats and node.plc_node_stats['last_contact'] != None:
384 filtercount['down'] += 1
386 filtercount['neverboot'] += 1
387 elif agg.history.status in ['good', 'online']:
388 filtercount['boot'] += 1
389 elif agg.history.status in ['debug', 'monitordebug']:
390 filtercount['debug'] += 1
392 if filtercount.has_key(agg.history.status):
393 filtercount[agg.history.status] += 1
397 if filter == "neverboot":
398 if not node.plc_node_stats or node.plc_node_stats['last_contact'] == None:
400 elif filter == "all":
402 elif filter == agg.history.status:
404 elif filter == 'boot':
408 widget = NodeWidget(template='monitorweb.templates.node_template')
409 return dict(now=time.ctime(), query=query, fc=filtercount, nodewidget=widget)
411 def nodeaction_handler(self, tg_exceptions=None):
412 """Handle any kind of error."""
413 print "NODEACTION_HANDLER------------------"
415 if 'pcuid' in request.params:
416 pcuid = request.params['pcuid']
418 refurl = request.headers.get("Referer",link("pcu"))
421 # TODO: do this more intelligently...
422 uri_fields = urllib.splitquery(refurl)
423 if uri_fields[1] is not None:
424 val = query_to_dict(uri_fields[1])
427 elif 'hostname' in val:
428 pcuid = FindbadNodeRecord.get_latest_by(hostname=val['hostname']).plc_pcuid
434 cherry_trail = cherrypy._cputil.get_object_trail()
435 for i in cherry_trail:
439 return self.pcuview(None, pcuid, **dict(exceptions=tg_exceptions))
441 def nodeaction(self, **data):
442 print "NODEACTION------------------"
443 for item in data.keys():
444 print "%s %s" % ( item, data[item] )
446 if 'hostname' in data:
447 hostname = data['hostname']
449 flash("No hostname given in submitted data")
452 if 'submit' in data or 'type' in data:
454 action = data['submit']
456 action = data['type']
458 flash("No submit action given in submitted data")
461 if action == "Reboot":
462 print "REBOOT: %s" % hostname
463 ret = reboot.reboot_str(str(hostname))
465 if ret: raise RuntimeError("Error using PCU: " + str(ret))
466 flash("Reboot appeared to work. Allow at most 5 minutes. Then run ExternalScan to check current status.")
468 elif action == "ExternalScan":
469 scanapi.externalprobe(str(hostname))
470 flash("External Scan Successful!")
471 elif action == "InternalScan":
472 scanapi.internalprobe(str(hostname))
473 flash("Internal Scan Successful!")
476 raise RuntimeError("Unknown action given")
479 @expose(template="monitorweb.templates.simpleview")
480 def simpleview(self, **data):
481 return self.pre_view(**data)
483 @expose(template="monitorweb.templates.detailview")
484 def detailview(self, **data):
485 return self.pre_view(**data)
487 def pre_view(self, **data):
488 session.flush(); session.clear()
495 # if objtype is not None, then treat 'hostname' or 'loginbase' as a search pattern
510 fields = obj.split(":")
513 obj = fields[1].replace("*", "%")
516 if len(obj.split(".")) > 1 or objtype == "node":
521 if 'loginbase' in data:
522 loginbase = data['loginbase']
524 if 'hostname' in data:
525 hostname = data['hostname']
528 try: pcuid = int(data['pcuid'])
532 try: since = int(since)
536 print "pcuid: %s" % pcuid
537 pcu = FindbadPCURecord.get_latest_by(plc_pcuid=pcuid)
538 loginbase_list += [ PlcSite.query.get(pcu.plc_pcu_stats['site_id']).plc_site_stats['login_base'] ]
542 nodes = [ FindbadNodeRecord.get_latest_by(hostname=hostname) ]
544 nodes = FindbadNodeRecord.query.filter(FindbadNodeRecord.hostname.like(hostname))
547 lb = PlcSite.query.get(node.plc_node_stats['site_id']).plc_site_stats['login_base']
548 if lb not in loginbase_list:
549 loginbase_list += [ lb ]
553 loginbase_list = [ loginbase ]
555 loginbase_list = HistorySiteRecord.query.filter(HistorySiteRecord.loginbase.like(loginbase))
556 loginbase_list = [ l.loginbase for l in loginbase_list ]
560 for loginbase in loginbase_list:
561 actions = ActionRecord.query.filter_by(loginbase=loginbase
562 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
563 ).order_by(ActionRecord.date_created.desc())
564 actions_list += [ a for a in actions ]
565 site = HistorySiteRecord.by_loginbase(loginbase)
567 sitequery.append(site)
568 # NOTE: because a single pcu may be assigned to multiple hosts,
569 # track unique pcus by their plc_pcuid, then turn dict into list
571 for node in FindbadNodeRecord.query.filter_by(loginbase=loginbase):
572 # NOTE: reformat some fields.
573 agg = prep_node_for_display(node)
576 pcus[agg.pcu.pcu.plc_pcuid] = agg.pcu
578 for pcuid_key in pcus:
579 pcuquery += [pcus[pcuid_key]]
581 return dict(sitequery=sitequery, pcuquery=pcuquery, nodequery=nodequery, actions=actions_list, since=since, exceptions=exceptions)
584 # TODO: add form validation
585 @expose(template="monitorweb.templates.pcuview")
586 @exception_handler(nodeaction_handler,"isinstance(tg_exceptions,RuntimeError)")
587 def pcuview(self, loginbase=None, pcuid=None, hostname=None, since=20, **data):
588 session.flush(); session.clear()
595 try: since = int(since)
601 if 'submit' in data.keys() or 'type' in data.keys():
602 if hostname: data['hostname'] = hostname
603 self.nodeaction(**data)
604 if 'exceptions' in data:
605 exceptions = data['exceptions']
609 if len(obj.split(".")) > 1: hostname = obj
613 print "pcuid: %s" % pcuid
614 pcu = FindbadPCURecord.get_latest_by(plc_pcuid=pcuid)
615 loginbase = PlcSite.query.get(pcu.plc_pcu_stats['site_id']).plc_site_stats['login_base']
618 node = FindbadNodeRecord.get_latest_by(hostname=hostname)
619 loginbase = PlcSite.query.get(node.plc_node_stats['site_id']).plc_site_stats['login_base']
622 actions = ActionRecord.query.filter_by(loginbase=loginbase
623 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
624 ).order_by(ActionRecord.date_created.desc())
625 actions = [ a for a in actions ]
626 sitequery = [HistorySiteRecord.by_loginbase(loginbase)]
628 for node in FindbadNodeRecord.query.filter_by(loginbase=loginbase):
629 # NOTE: reformat some fields.
630 agg = prep_node_for_display(node)
632 if agg.pcu: #.pcu.plc_pcuid: # not None
633 #pcu = FindbadPCURecord.get_latest_by(plc_pcuid=agg.plc_pcuid)
634 #prep_pcu_for_display(pcu)
635 pcus[agg.pcu.pcu.plc_pcuid] = agg.pcu
637 for pcuid_key in pcus:
638 pcuquery += [pcus[pcuid_key]]
640 return dict(sitequery=sitequery, pcuquery=pcuquery, nodequery=nodequery, actions=actions, since=since, exceptions=exceptions)
642 @expose(template="monitorweb.templates.pcuhistory")
643 def pcuhistory(self, pcu_id=None):
646 fbnode = HistoryPCURecord.get_by(plc_pcuid=pcu_id)
647 l = fbnode.versions[-100:]
650 #prep_node_for_display(node)
653 return dict(query=query, pcu_id=pcu_id)
655 @expose(template="monitorweb.templates.nodescanhistory")
656 def nodescanhistory(self, hostname=None, length=10):
657 try: length = int(length)
660 fbnode = FindbadNodeRecord.get_by(hostname=hostname)
661 # TODO: add links for earlier history if desired.
662 l = fbnode.versions[-length:]
666 agg = prep_node_for_display(node, pcuhash=None, preppcu=False, asofdate=node.timestamp)
669 if 'length' in request.params:
670 del request.params['length']
671 return dict(query=query, hostname=hostname, params=request.params)
673 @expose(template="monitorweb.templates.nodehistory")
674 def nodehistory(self, hostname=None):
677 fbnode = HistoryNodeRecord.get_by(hostname=hostname)
678 l = fbnode.versions[-100:]
681 #prep_node_for_display(node)
684 return dict(query=query, hostname=hostname)
686 @expose(template="monitorweb.templates.sitehistory")
687 def sitehistory(self, loginbase=None):
690 fbsite = HistorySiteRecord.get_by(loginbase=loginbase)
691 # TODO: add links for earlier history if desired.
692 l = fbsite.versions[-100:]
696 return dict(query=query, loginbase=loginbase)
699 @expose(template="monitorweb.templates.pculist")
700 def pcu(self, filter='all'):
701 print "PCUVIEW------------------"
702 print "befor-len: ", len( [ i for i in session] )
703 session.flush(); session.clear()
704 print "after-len: ", len( [ i for i in session] )
705 fbquery = FindbadPCURecord.get_all_latest()
707 filtercount = {'ok' : 0, 'NetDown': 0, 'Not_Run' : 0, 'pending' : 0, 'all' : 0}
711 if node.reboot_trial_status == str(0):
712 filtercount['ok'] += 1
713 elif node.reboot_trial_status == 'NetDown' or node.reboot_trial_status == 'Not_Run':
714 filtercount[node.reboot_trial_status] += 1
716 filtercount['pending'] += 1
718 pcuagg = prep_pcu_for_display(node)
723 elif filter == "ok" and node.reboot_trial_status == str(0):
725 elif filter == node.reboot_trial_status:
727 elif filter == "pending":
728 # TODO: look in message logs...
729 if node.reboot_trial_status != str(0) and \
730 node.reboot_trial_status != 'NetDown' and \
731 node.reboot_trial_status != 'Not_Run':
735 return dict(query=query, fc=filtercount)
737 @expose(template="monitorweb.templates.sitelist")
738 def site(self, filter='all'):
739 print "SITE------------------"
740 print "befor-len: ", len( [ i for i in session] )
741 session.flush(); session.clear()
742 print "after-len: ", len( [ i for i in session] )
743 filtercount = {'good' : 0, 'down': 0, 'online':0, 'offline' : 0, 'new' : 0, 'pending' : 0, 'all' : 0}
744 fbquery = HistorySiteRecord.query.all()
748 filtercount['all'] += 1
749 if site.new and site.slices_used == 0 and not site.enabled:
750 filtercount['new'] += 1
751 elif not site.enabled:
752 filtercount['pending'] += 1
753 elif site.status in ['good', 'online']:
754 filtercount['good'] += 1
755 elif site.status in ['down', 'offline']:
756 filtercount['down'] += 1
761 elif filter == 'new' and site.new and site.slices_used == 0 and not site.enabled:
763 elif filter == "pending" and not site.enabled:
765 elif filter == 'good' and site.status in ['good', 'online']:
767 elif filter == 'down' and site.status in ['down', 'offline']:
770 return dict(query=query, fc=filtercount)
771 @expose(template="monitorweb.templates.sitesummary")
772 def sitesummary(self, loginbase="princeton"):
774 for node in FindbadNodeRecord.query.filter_by(loginbase=loginbase):
775 agg = prep_node_for_display(node)
778 return dict(nodequery=nodequery, loginbase=loginbase)
780 @expose(template="monitorweb.templates.summary")
781 def summary(self, since=7):
783 sumdata['nodes'] = {}
784 sumdata['sites'] = {}
787 def summarize(query, type):
789 if o.status not in sumdata[type]:
790 sumdata[type][o.status] = 0
791 sumdata[type][o.status] += 1
793 fbquery = HistorySiteRecord.query.all()
794 summarize(fbquery, 'sites')
795 fbquery = HistoryPCURecord.query.all()
796 summarize(fbquery, 'pcus')
797 fbquery = HistoryNodeRecord.query.all()
798 summarize(fbquery, 'nodes')
800 if 'monitordebug' in sumdata['nodes']:
801 d = sumdata['nodes']['monitordebug']
802 del sumdata['nodes']['monitordebug']
803 sumdata['nodes']['failboot'] = d
805 return dict(sumdata=sumdata, setorder=['good', 'offline', 'down', 'online'])
807 @expose(template="monitorweb.templates.actionsummary")
808 def actionsummary(self, since=7):
809 from monitor.wrapper.emailTxt import mailtxt
811 types = filter(lambda x: 'notice' in x, dir(mailtxt))
814 print mon_metadata.bind
815 if session.bind is None:
816 #TODO: figure out why this value gets cleared out...
817 session.bind = mon_metadata.bind
818 result = session.execute("select distinct(action_type) from actionrecord;")
820 types = [r[0] for r in result]
822 try: since = int(since)
826 acts = ActionRecord.query.filter(ActionRecord.action_type==t
827 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since))
828 results[t] = acts.count()
829 return dict(results=results)
831 @expose(template="monitorweb.templates.actionlist")
832 def actionlist(self, action_type='down_notice', since=7, loginbase=None):
834 try: since = int(since)
838 acts = ActionRecord.query.filter_by(loginbase=loginbase
839 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
840 ).order_by(ActionRecord.date_created.desc())
842 acts = ActionRecord.query.filter(ActionRecord.action_type==action_type
843 ).filter(ActionRecord.date_created >= datetime.now() - timedelta(since)
844 ).order_by(ActionRecord.date_created.desc())
845 query = [ a for a in acts ]
847 return dict(actions=query, action_type=action_type, since=since)
850 def upload(self, log, **keywords):
852 data = log.file.read()
853 target_file_name = os.path.join(os.getcwd(), log.filename)
854 # open file in binary mode for writing
856 f = open(target_file_name, 'wb')
861 #flash("File uploaded successfully: %s saved as: %s" \
862 # % (upload_file.filename, target_file_name))
863 #u = UploadedFile(filename=upload_file.filename,
864 # abspath=target_file_name, size=0)