Bug fixes to parallelization
[plstackapi.git] / planetstack / ec2_observer / event_loop.py
1 import os
2 import imp
3 import inspect
4 import time
5 import traceback
6 import commands
7 import threading
8 import json
9
10 from datetime import datetime
11 from collections import defaultdict
12 from core.models import *
13 from django.db.models import F, Q
14 #from openstack.manager import OpenStackManager
15 from openstack.driver import OpenStackDriver
16 from util.logger import Logger, logging, logger
17 #from timeout import timeout
18 from planetstack.config import Config
19 from observer.steps import *
20 from syncstep import SyncStep
21 from toposort import toposort
22 from observer.error_mapper import *
23
24 debug_mode = False
25
26 logger = Logger(level=logging.INFO)
27
28 class StepNotReady(Exception):
29         pass
30
31 class NoOpDriver:
32         def __init__(self):
33                  self.enabled = True
34                  self.dependency_graph = None
35
36 STEP_STATUS_WORKING=1
37 STEP_STATUS_OK=2
38 STEP_STATUS_KO=3
39
40 def invert_graph(g):
41         ig = {}
42         for k,v in g.items():
43                 for v0 in v:
44                         try:
45                                 ig[v0].append(k)
46                         except:
47                                 ig=[k]
48         return ig
49
50 class PlanetStackObserver:
51         #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
52         sync_steps = []
53
54         
55         def __init__(self):
56                 # The Condition object that gets signalled by Feefie events
57                 self.step_lookup = {}
58                 self.load_sync_step_modules()
59                 self.load_sync_steps()
60                 self.event_cond = threading.Condition()
61
62                 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
63                 if self.driver_kind=="openstack":
64                         self.driver = OpenStackDriver()
65                 else:
66                         self.driver = NoOpDriver()
67
68         def wait_for_event(self, timeout):
69                 self.event_cond.acquire()
70                 self.event_cond.wait(timeout)
71                 self.event_cond.release()
72
73         def wake_up(self):
74                 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
75                 self.event_cond.acquire()
76                 self.event_cond.notify()
77                 self.event_cond.release()
78
79         def load_sync_step_modules(self, step_dir=None):
80                 if step_dir is None:
81                         if hasattr(Config(), "observer_steps_dir"):
82                                 step_dir = Config().observer_steps_dir
83                         else:
84                                 step_dir = "/opt/planetstack/observer/steps"
85
86                 for fn in os.listdir(step_dir):
87                         pathname = os.path.join(step_dir,fn)
88                         if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
89                                 module = imp.load_source(fn[:-3],pathname)
90                                 for classname in dir(module):
91                                         c = getattr(module, classname, None)
92
93                                         # make sure 'c' is a descendent of SyncStep and has a
94                                         # provides field (this eliminates the abstract base classes
95                                         # since they don't have a provides)
96
97                                         if inspect.isclass(c) and issubclass(c, SyncStep) and hasattr(c,"provides") and (c not in self.sync_steps):
98                                                 self.sync_steps.append(c)
99                 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
100                 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
101
102         def load_sync_steps(self):
103                 dep_path = Config().observer_dependency_graph
104                 logger.info('Loading model dependency graph from %s' % dep_path)
105                 try:
106                         # This contains dependencies between records, not sync steps
107                         self.model_dependency_graph = json.loads(open(dep_path).read())
108                 except Exception,e:
109                         raise e
110
111                 try:
112                         backend_path = Config().observer_pl_dependency_graph
113                         logger.info('Loading backend dependency graph from %s' % backend_path)
114                         # This contains dependencies between backend records
115                         self.backend_dependency_graph = json.loads(open(backend_path).read())
116                 except Exception,e:
117                         logger.info('Backend dependency graph not loaded')
118                         # We can work without a backend graph
119                         self.backend_dependency_graph = {}
120
121                 provides_dict = {}
122                 for s in self.sync_steps:
123                         self.step_lookup[s.__name__] = s 
124                         for m in s.provides:
125                                 try:
126                                         provides_dict[m.__name__].append(s.__name__)
127                                 except KeyError:
128                                         provides_dict[m.__name__]=[s.__name__]
129
130                 step_graph = {}
131                 for k,v in self.model_dependency_graph.iteritems():
132                         try:
133                                 for source in provides_dict[k]:
134                                         for m in v:
135                                                 try:
136                                                         for dest in provides_dict[m]:
137                                                                 # no deps, pass
138                                                                 try:
139                                                                         if (dest not in step_graph[source]):
140                                                                                 step_graph[source].append(dest)
141                                                                 except:
142                                                                         step_graph[source]=[dest]
143                                                 except KeyError:
144                                                         pass
145                                         
146                         except KeyError:
147                                 pass
148                                 # no dependencies, pass
149                 
150                 #import pdb
151                 #pdb.set_trace()
152                 if (self.backend_dependency_graph):
153                         backend_dict = {}
154                         for s in self.sync_steps:
155                                 for m in s.serves:
156                                         backend_dict[m]=s.__name__
157                                         
158                         for k,v in backend_dependency_graph.iteritems():
159                                 try:
160                                         source = backend_dict[k]
161                                         for m in v:
162                                                 try:
163                                                         dest = backend_dict[m]
164                                                 except KeyError:
165                                                         # no deps, pass
166                                                         pass
167                                                 step_graph[source]=dest
168                                                 
169                                 except KeyError:
170                                         pass
171                                         # no dependencies, pass
172
173                 self.dependency_graph = step_graph
174                 self.deletion_dependency_graph = invert_graph(step_graph)
175
176                 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
177                 print "Order of steps=",self.ordered_steps
178                 self.load_run_times()
179                 
180
181         def check_duration(self, step, duration):
182                 try:
183                         if (duration > step.deadline):
184                                 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
185                 except AttributeError:
186                         # S doesn't have a deadline
187                         pass
188
189         def update_run_time(self, step, deletion):
190                 if (not deletion):
191                         self.last_run_times[step.__name__]=time.time()
192                 else:
193                         self.last_deletion_run_times[step.__name__]=time.time()
194
195
196         def check_schedule(self, step, deletion):
197                 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
198
199                 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
200                 try:
201                         if (time_since_last_run < step.requested_interval):
202                                 raise StepNotReady
203                 except AttributeError:
204                         logger.info('Step %s does not have requested_interval set'%step.__name__)
205                         raise StepNotReady
206         
207         def load_run_times(self):
208                 try:
209                         jrun_times = open('/tmp/observer_run_times').read()
210                         self.last_run_times = json.loads(jrun_times)
211                 except:
212                         self.last_run_times={}
213                         for e in self.ordered_steps:
214                                 self.last_run_times[e]=0
215                 try:
216                         jrun_times = open('/tmp/observer_deletion_run_times').read()
217                         self.last_deletion_run_times = json.loads(jrun_times)
218                 except:
219                         self.last_deletion_run_times={}
220                         for e in self.ordered_steps:
221                                 self.last_deletion_run_times[e]=0
222
223
224
225         def save_run_times(self):
226                 run_times = json.dumps(self.last_run_times)
227                 open('/tmp/observer_run_times','w').write(run_times)
228
229                 deletion_run_times = json.dumps(self.last_deletion_run_times)
230                 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
231
232         def check_class_dependency(self, step, failed_steps):
233                 step.dependenices = []
234                 for obj in step.provides:
235                         step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
236                 for failed_step in failed_steps:
237                         if (failed_step in step.dependencies):
238                                 raise StepNotReady
239
240         def sync(self, S, deletion):
241                 step = self.step_lookup[S]
242                 start_time=time.time()
243                 
244                 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
245
246                 # Wait for step dependencies to be met
247                 try:
248                         deps = self.dependency_graph[S]
249                         has_deps = True
250                 except KeyError:
251                         has_deps = False
252
253                 if (has_deps):
254                         for d in deps:
255                                 cond = self.step_conditions[d]
256                                 cond.acquire()
257                                 if (self.step_status[d] is STEP_STATUS_WORKING):
258                                         cond.wait()
259                                 cond.release()
260                         go = self.step_status[d] == STEP_STATUS_OK
261                 else:
262                         go = True
263
264                 if (not go):
265                         self.failed_steps.append(sync_step)
266                         my_status = STEP_STATUS_KO
267                 else:
268                         sync_step = step(driver=self.driver,error_map=self.error_mapper)
269                         sync_step.__name__ = step.__name__
270                         sync_step.dependencies = []
271                         try:
272                                 mlist = sync_step.provides
273                                 
274                                 for m in mlist:
275                                         sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
276                         except KeyError:
277                                 pass
278                         sync_step.debug_mode = debug_mode
279
280                         should_run = False
281                         try:
282                                 # Various checks that decide whether
283                                 # this step runs or not
284                                 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
285                                 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
286                                 should_run = True
287                         except StepNotReady:
288                                 logging.info('Step not ready: %s'%sync_step.__name__)
289                                 self.failed_steps.append(sync_step)
290                                 my_status = STEP_STATUS_KO
291                         except Exception,e:
292                                 logging.error('%r',e)
293                                 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
294                                 self.failed_steps.append(sync_step)
295                                 my_status = STEP_STATUS_KO
296
297                         if (should_run):
298                                 try:
299                                         duration=time.time() - start_time
300
301                                         logger.info('Executing step %s' % sync_step.__name__)
302
303                                         failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
304
305                                         self.check_duration(sync_step, duration)
306
307                                         if failed_objects:
308                                                 self.failed_step_objects.update(failed_objects)
309
310                                         my_status = STEP_STATUS_OK
311                                         self.update_run_time(sync_step,deletion)
312                                 except Exception,e:
313                                         logging.error('Model step failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
314                                         logger.log_exc(e)
315                                         self.failed_steps.append(S)
316                                         my_status = STEP_STATUS_KO
317                         else:
318                                 my_status = STEP_STATUS_OK
319                 
320                 try:
321                         my_cond = self.step_conditions[S]
322                         my_cond.acquire()
323                         self.step_status[S]=my_status
324                         my_cond.notify_all()
325                         my_cond.release()
326                 except KeyError,e:
327                         logging.info('Step %r is a leaf')
328                         pass
329
330         def run(self):
331                 if not self.driver.enabled:
332                         return
333
334                 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
335                         return
336
337                 while True:
338                         try:
339                                 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
340                                 self.error_mapper = ErrorMapper(error_map_file)
341
342                                 # Set of whole steps that failed
343                                 self.failed_steps = []
344
345                                 # Set of individual objects within steps that failed
346                                 self.failed_step_objects = set()
347
348                                 # Set up conditions and step status
349                                 # This is needed for steps to run in parallel
350                                 # while obeying dependencies.
351
352                                 providers = set()
353                                 for v in self.dependency_graph.values():
354                                         if (v):
355                                                 providers.update(v)
356
357                                 self.step_conditions = {}
358                                 self.step_status = {}
359                                 for p in list(providers):
360                                         self.step_conditions[p] = threading.Condition()
361                                         self.step_status[p] = STEP_STATUS_IDLE
362
363
364                                 logger.info('Waiting for event')
365                                 tBeforeWait = time.time()
366                                 self.wait_for_event(timeout=30)
367                                 logger.info('Observer woke up')
368
369                                 # Two passes. One for sync, the other for deletion.
370                                 for deletion in [False,True]:
371                                         threads = []
372                                         logger.info('Deletion=%r...'%deletion)
373                                         schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
374
375                                         for S in schedule:
376                                                 thread = threading.Thread(target=self.sync, args=(S, deletion))
377
378                                                 logger.info('Deletion=%r...'%deletion)
379                                                 threads.append(thread)
380
381                                         # Start threads 
382                                         for t in threads:
383                                                 t.start()
384
385                                         # Wait for all threads to finish before continuing with the run loop
386                                         for t in threads:
387                                                 t.join()
388
389                                 self.save_run_times()
390                         except Exception, e:
391                                 logging.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
392                                 logger.log_exc("Exception in observer run loop")
393                                 traceback.print_exc()