Various bugfies to the main Observer loop
[plstackapi.git] / planetstack / openstack_observer / event_loop.py
1 import os
2 import imp
3 import inspect
4 import time
5 import sys
6 import traceback
7 import commands
8 import threading
9 import json
10 import pdb
11 import pprint
12
13
14 from datetime import datetime
15 from collections import defaultdict
16 from core.models import *
17 from django.db.models import F, Q
18 from django.db import connection
19 #from openstack.manager import OpenStackManager
20 from openstack.driver import OpenStackDriver
21 from util.logger import Logger, logging, logger
22 #from timeout import timeout
23 from planetstack.config import Config
24 from observer.steps import *
25 from syncstep import SyncStep
26 from toposort import toposort
27 from observer.error_mapper import *
28 from openstack_observer.openstacksyncstep import OpenStackSyncStep
29
30
31 debug_mode = False
32
33 class bcolors:
34     HEADER = '\033[95m'
35     OKBLUE = '\033[94m'
36     OKGREEN = '\033[92m'
37     WARNING = '\033[93m'
38     FAIL = '\033[91m'
39     ENDC = '\033[0m'
40     BOLD = '\033[1m'
41     UNDERLINE = '\033[4m'
42
43 logger = Logger(level=logging.INFO)
44
45 class StepNotReady(Exception):
46         pass
47
48 class NoOpDriver:
49         def __init__(self):
50                  self.enabled = True
51                  self.dependency_graph = None
52
53 STEP_STATUS_WORKING=1
54 STEP_STATUS_OK=2
55 STEP_STATUS_KO=3
56
57 def invert_graph(g):
58         ig = {}
59         for k,v in g.items():
60                 for v0 in v:
61                         try:
62                                 ig[v0].append(k)
63                         except:
64                                 ig=[k]
65         return ig
66
67 class PlanetStackObserver:
68         #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivilege,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
69         sync_steps = []
70
71         
72         def __init__(self):
73                 # The Condition object that gets signalled by Feefie events
74                 self.step_lookup = {}
75                 self.load_sync_step_modules()
76                 self.load_sync_steps()
77                 self.event_cond = threading.Condition()
78
79                 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
80                 if self.driver_kind=="openstack":
81                         self.driver = OpenStackDriver()
82                 else:
83                         self.driver = NoOpDriver()
84
85         def wait_for_event(self, timeout):
86                 self.event_cond.acquire()
87                 self.event_cond.wait(timeout)
88                 self.event_cond.release()
89
90         def wake_up(self):
91                 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
92                 self.event_cond.acquire()
93                 self.event_cond.notify()
94                 self.event_cond.release()
95
96         def load_sync_step_modules(self, step_dir=None):
97                 if step_dir is None:
98                         if hasattr(Config(), "observer_steps_dir"):
99                                 step_dir = Config().observer_steps_dir
100                         else:
101                                 step_dir = "/opt/planetstack/observer/steps"
102
103                 for fn in os.listdir(step_dir):
104                         pathname = os.path.join(step_dir,fn)
105                         if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
106                                 module = imp.load_source(fn[:-3],pathname)
107                                 for classname in dir(module):
108                                         c = getattr(module, classname, None)
109
110                                         # make sure 'c' is a descendent of SyncStep and has a
111                                         # provides field (this eliminates the abstract base classes
112                                         # since they don't have a provides)
113
114                                         if inspect.isclass(c) and (issubclass(c, SyncStep) or issubclass(c,OpenStackSyncStep)) and hasattr(c,"provides") and (c not in self.sync_steps):
115                                                 self.sync_steps.append(c)
116                 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
117                 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
118
119         def load_sync_steps(self):
120                 dep_path = Config().observer_dependency_graph
121                 logger.info('Loading model dependency graph from %s' % dep_path)
122                 try:
123                         # This contains dependencies between records, not sync steps
124                         self.model_dependency_graph = json.loads(open(dep_path).read())
125                         for lst in self.model_dependency_graph.values():
126                                 for k in lst:
127                                         try:
128                                                 deps = self.model_dependency_graph[k]
129                                         except:
130                                                 self.model_dependency_graph[k] = []
131                 except Exception,e:
132                         raise e
133
134                 try:
135                         backend_path = Config().observer_pl_dependency_graph
136                         logger.info('Loading backend dependency graph from %s' % backend_path)
137                         # This contains dependencies between backend records
138                         self.backend_dependency_graph = json.loads(open(backend_path).read())
139                 except Exception,e:
140                         logger.info('Backend dependency graph not loaded')
141                         # We can work without a backend graph
142                         self.backend_dependency_graph = {}
143
144                 provides_dict = {}
145                 for s in self.sync_steps:
146                         self.step_lookup[s.__name__] = s 
147                         for m in s.provides:
148                                 try:
149                                         provides_dict[m.__name__].append(s.__name__)
150                                 except KeyError:
151                                         provides_dict[m.__name__]=[s.__name__]
152
153                 step_graph = {}
154                 for k,v in self.model_dependency_graph.iteritems():
155                         try:
156                                 for source in provides_dict[k]:
157                                         if (not v):
158                                                 step_graph[source] = []
159                 
160                                         for m in v:
161                                                 try:
162                                                         for dest in provides_dict[m]:
163                                                                 # no deps, pass
164                                                                 try:
165                                                                         if (dest not in step_graph[source]):
166                                                                                 step_graph[source].append(dest)
167                                                                 except:
168                                                                         step_graph[source]=[dest]
169                                                 except KeyError:
170                                                         pass
171                                         
172                         except KeyError:
173                                 pass
174                                 # no dependencies, pass
175                 
176                 #import pdb
177                 #pdb.set_trace()
178                 if (self.backend_dependency_graph):
179                         backend_dict = {}
180                         for s in self.sync_steps:
181                                 for m in s.serves:
182                                         backend_dict[m]=s.__name__
183                                         
184                         for k,v in backend_dependency_graph.iteritems():
185                                 try:
186                                         source = backend_dict[k]
187                                         for m in v:
188                                                 try:
189                                                         dest = backend_dict[m]
190                                                 except KeyError:
191                                                         # no deps, pass
192                                                         pass
193                                                 step_graph[source]=dest
194                                                 
195                                 except KeyError:
196                                         pass
197                                         # no dependencies, pass
198
199                 self.dependency_graph = step_graph
200                 self.deletion_dependency_graph = invert_graph(step_graph)
201
202                 pp = pprint.PrettyPrinter(indent=4)
203                 pp.pprint(step_graph)
204                 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
205                 #self.ordered_steps = ['SyncRoles', 'SyncControllerSites', 'SyncControllerSitePrivileges','SyncImages', 'SyncControllerImages','SyncControllerUsers','SyncControllerUserSitePrivileges','SyncControllerSlices', 'SyncControllerSlicePrivileges', 'SyncControllerUserSlicePrivileges', 'SyncControllerNetworks','SyncSlivers']
206                 #self.ordered_steps = ['SyncControllerSites']
207
208                 print "Order of steps=",self.ordered_steps
209
210                 self.load_run_times()
211                 
212
213         def check_duration(self, step, duration):
214                 try:
215                         if (duration > step.deadline):
216                                 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
217                 except AttributeError:
218                         # S doesn't have a deadline
219                         pass
220
221         def update_run_time(self, step, deletion):
222                 if (not deletion):
223                         self.last_run_times[step.__name__]=time.time()
224                 else:
225                         self.last_deletion_run_times[step.__name__]=time.time()
226
227
228         def check_schedule(self, step, deletion):
229                 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
230
231                 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
232                 try:
233                         if (time_since_last_run < step.requested_interval):
234                                 raise StepNotReady
235                 except AttributeError:
236                         logger.info('Step %s does not have requested_interval set'%step.__name__)
237                         raise StepNotReady
238         
239         def load_run_times(self):
240                 try:
241                         jrun_times = open('/tmp/observer_run_times').read()
242                         self.last_run_times = json.loads(jrun_times)
243                 except:
244                         self.last_run_times={}
245                         for e in self.ordered_steps:
246                                 self.last_run_times[e]=0
247                 try:
248                         jrun_times = open('/tmp/observer_deletion_run_times').read()
249                         self.last_deletion_run_times = json.loads(jrun_times)
250                 except:
251                         self.last_deletion_run_times={}
252                         for e in self.ordered_steps:
253                                 self.last_deletion_run_times[e]=0
254
255
256         def save_run_times(self):
257                 run_times = json.dumps(self.last_run_times)
258                 open('/tmp/observer_run_times','w').write(run_times)
259
260                 deletion_run_times = json.dumps(self.last_deletion_run_times)
261                 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
262
263         def check_class_dependency(self, step, failed_steps):
264                 step.dependenices = []
265                 for obj in step.provides:
266                         step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
267                 for failed_step in failed_steps:
268                         if (failed_step in step.dependencies):
269                                 raise StepNotReady
270
271         def sync(self, S, deletion):
272             try:
273                 step = self.step_lookup[S]
274                 start_time=time.time()
275
276                 logger.info("Starting to work on step %s" % step.__name__)
277                 
278                 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
279
280                 # Wait for step dependencies to be met
281                 try:
282                         deps = self.dependency_graph[S]
283                         has_deps = True
284                 except KeyError:
285                         has_deps = False
286
287                 go = True
288
289                 failed_dep = None
290                 if (has_deps):
291                         for d in deps:
292                                 if d==step.__name__:
293                                     logger.info("   step %s self-wait skipped" % step.__name__)
294                                     go = True
295                                     continue
296
297                                 cond = self.step_conditions[d]
298                                 cond.acquire()
299                                 if (self.step_status[d] is STEP_STATUS_WORKING):
300                                         logger.info("  step %s wait on dep %s" % (step.__name__, d))
301                                         cond.wait()
302                                 elif self.step_status[d] == STEP_STATUS_OK:
303                                         go = True
304                                 else:
305                                         go = False
306                                         failed_dep = d
307                                 cond.release()
308                                 if (not go):
309                                         break
310                 else:
311                         go = True
312
313                 if (not go):
314                         print bcolors.FAIL + "Step %r skipped on %r" % (step,failed_dep) + bcolors.ENDC
315                         # SMBAKER: sync_step was not defined here, so I changed
316                         #    this from 'sync_step' to 'step'. Verify.
317                         self.failed_steps.append(step)
318                         my_status = STEP_STATUS_KO
319                 else:
320                         sync_step = step(driver=self.driver,error_map=self.error_mapper)
321                         sync_step.__name__ = step.__name__
322                         sync_step.dependencies = []
323                         try:
324                                 mlist = sync_step.provides
325
326                                 for m in mlist:
327                                         sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
328                         except KeyError:
329                                 pass
330                         sync_step.debug_mode = debug_mode
331
332                         should_run = False
333                         try:
334                                 # Various checks that decide whether
335                                 # this step runs or not
336                                 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
337                                 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
338                                 should_run = True
339                         except StepNotReady:
340                                 logger.info('Step not ready: %s'%sync_step.__name__)
341                                 self.failed_steps.append(sync_step)
342                                 my_status = STEP_STATUS_KO
343                         except Exception,e:
344                                 logger.error('%r' % e)
345                                 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
346                                 self.failed_steps.append(sync_step)
347                                 my_status = STEP_STATUS_KO
348
349                         if (should_run):
350                                 try:
351                                         duration=time.time() - start_time
352
353                                         logger.info('Executing step %s' % sync_step.__name__)
354
355                                         print bcolors.OKBLUE + "Executing step %s" % sync_step.__name__ + bcolors.ENDC
356                                         failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
357
358                                         self.check_duration(sync_step, duration)
359
360                                         if failed_objects:
361                                                 self.failed_step_objects.update(failed_objects)
362
363                                         logger.info("Step %r succeeded" % step)
364                                         print bcolors.OKGREEN + "Step %r succeeded" % step + bcolors.ENDC
365                                         my_status = STEP_STATUS_OK
366                                         self.update_run_time(sync_step,deletion)
367                                 except Exception,e:
368                                         print bcolors.FAIL + "Model step %r failed" % (step) + bcolors.ENDC
369                                         logger.error('Model step %r failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % (step, e))
370                                         logger.log_exc(e)
371                                         self.failed_steps.append(S)
372                                         my_status = STEP_STATUS_KO
373                         else:
374                                 logger.info("Step %r succeeded due to non-run" % step)
375                                 my_status = STEP_STATUS_OK
376
377                 try:
378                         my_cond = self.step_conditions[S]
379                         my_cond.acquire()
380                         self.step_status[S]=my_status
381                         my_cond.notify_all()
382                         my_cond.release()
383                 except KeyError,e:
384                         logger.info('Step %r is a leaf' % step)
385                         pass
386             finally:
387                 connection.close()
388
389         def run(self):
390                 if not self.driver.enabled:
391                         return
392
393                 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
394                         return
395
396                 while True:
397                         try:
398                                 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
399                                 self.error_mapper = ErrorMapper(error_map_file)
400
401                                 # Set of whole steps that failed
402                                 self.failed_steps = []
403
404                                 # Set of individual objects within steps that failed
405                                 self.failed_step_objects = set()
406
407                                 # Set up conditions and step status
408                                 # This is needed for steps to run in parallel
409                                 # while obeying dependencies.
410
411                                 providers = set()
412                                 for v in self.dependency_graph.values():
413                                         if (v):
414                                                 providers.update(v)
415
416                                 self.step_conditions = {}
417                                 self.step_status = {}
418                                 for p in list(providers):
419                                         self.step_conditions[p] = threading.Condition()
420                                         self.step_status[p] = STEP_STATUS_WORKING
421
422
423                                 logger.info('Waiting for event')
424                                 tBeforeWait = time.time()
425                                 self.wait_for_event(timeout=30)
426                                 logger.info('Observer woke up')
427
428                                 # Two passes. One for sync, the other for deletion.
429                                 for deletion in [False,True]:
430                                         threads = []
431                                         logger.info('Deletion=%r...'%deletion)
432                                         schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
433
434                                         for S in schedule:
435                                                 thread = threading.Thread(target=self.sync, args=(S, deletion))
436
437                                                 logger.info('Deletion=%r...'%deletion)
438                                                 threads.append(thread)
439
440                                         # Start threads 
441                                         for t in threads:
442                                                 t.start()
443
444                                         # Wait for all threads to finish before continuing with the run loop
445                                         for t in threads:
446                                                 t.join()
447
448                                 self.save_run_times()
449                         except Exception, e:
450                                 logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e)
451                                 logger.log_exc("Exception in observer run loop")
452                                 traceback.print_exc()