11 from datetime import datetime
12 from collections import defaultdict
13 from core.models import *
14 from django.db.models import F, Q
15 #from openstack.manager import OpenStackManager
16 from openstack.driver import OpenStackDriver
17 from util.logger import Logger, logging, logger
18 #from timeout import timeout
19 from planetstack.config import Config
20 from observer.steps import *
21 from syncstep import SyncStep
22 from toposort import toposort
23 from observer.error_mapper import *
27 logger = Logger(level=logging.INFO)
29 class StepNotReady(Exception):
35 self.dependency_graph = None
51 class PlanetStackObserver:
52 #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
57 # The Condition object that gets signalled by Feefie events
59 self.load_sync_step_modules()
60 self.load_sync_steps()
61 self.event_cond = threading.Condition()
63 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
64 if self.driver_kind=="openstack":
65 self.driver = OpenStackDriver()
67 self.driver = NoOpDriver()
69 def wait_for_event(self, timeout):
70 self.event_cond.acquire()
71 self.event_cond.wait(timeout)
72 self.event_cond.release()
75 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
76 self.event_cond.acquire()
77 self.event_cond.notify()
78 self.event_cond.release()
80 def load_sync_step_modules(self, step_dir=None):
82 if hasattr(Config(), "observer_steps_dir"):
83 step_dir = Config().observer_steps_dir
85 step_dir = "/opt/planetstack/observer/steps"
87 for fn in os.listdir(step_dir):
88 pathname = os.path.join(step_dir,fn)
89 if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
90 module = imp.load_source(fn[:-3],pathname)
91 for classname in dir(module):
92 c = getattr(module, classname, None)
94 # make sure 'c' is a descendent of SyncStep and has a
95 # provides field (this eliminates the abstract base classes
96 # since they don't have a provides)
98 if inspect.isclass(c) and issubclass(c, SyncStep) and hasattr(c,"provides") and (c not in self.sync_steps):
99 self.sync_steps.append(c)
100 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
101 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
103 def load_sync_steps(self):
104 dep_path = Config().observer_dependency_graph
105 logger.info('Loading model dependency graph from %s' % dep_path)
107 # This contains dependencies between records, not sync steps
108 self.model_dependency_graph = json.loads(open(dep_path).read())
113 backend_path = Config().observer_pl_dependency_graph
114 logger.info('Loading backend dependency graph from %s' % backend_path)
115 # This contains dependencies between backend records
116 self.backend_dependency_graph = json.loads(open(backend_path).read())
118 logger.info('Backend dependency graph not loaded')
119 # We can work without a backend graph
120 self.backend_dependency_graph = {}
123 for s in self.sync_steps:
124 self.step_lookup[s.__name__] = s
127 provides_dict[m.__name__].append(s.__name__)
129 provides_dict[m.__name__]=[s.__name__]
132 for k,v in self.model_dependency_graph.iteritems():
134 for source in provides_dict[k]:
137 for dest in provides_dict[m]:
140 if (dest not in step_graph[source]):
141 step_graph[source].append(dest)
143 step_graph[source]=[dest]
149 # no dependencies, pass
153 if (self.backend_dependency_graph):
155 for s in self.sync_steps:
157 backend_dict[m]=s.__name__
159 for k,v in backend_dependency_graph.iteritems():
161 source = backend_dict[k]
164 dest = backend_dict[m]
168 step_graph[source]=dest
172 # no dependencies, pass
174 self.dependency_graph = step_graph
175 self.deletion_dependency_graph = invert_graph(step_graph)
177 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
178 print "Order of steps=",self.ordered_steps
179 self.load_run_times()
182 def check_duration(self, step, duration):
184 if (duration > step.deadline):
185 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
186 except AttributeError:
187 # S doesn't have a deadline
190 def update_run_time(self, step, deletion):
192 self.last_run_times[step.__name__]=time.time()
194 self.last_deletion_run_times[step.__name__]=time.time()
197 def check_schedule(self, step, deletion):
198 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
200 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
202 if (time_since_last_run < step.requested_interval):
204 except AttributeError:
205 logger.info('Step %s does not have requested_interval set'%step.__name__)
208 def load_run_times(self):
210 jrun_times = open('/tmp/observer_run_times').read()
211 self.last_run_times = json.loads(jrun_times)
213 self.last_run_times={}
214 for e in self.ordered_steps:
215 self.last_run_times[e]=0
217 jrun_times = open('/tmp/observer_deletion_run_times').read()
218 self.last_deletion_run_times = json.loads(jrun_times)
220 self.last_deletion_run_times={}
221 for e in self.ordered_steps:
222 self.last_deletion_run_times[e]=0
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)
229 deletion_run_times = json.dumps(self.last_deletion_run_times)
230 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
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):
240 def sync(self, S, deletion):
241 step = self.step_lookup[S]
242 start_time=time.time()
244 logger.info("Starting to work on step %s" % step.__name__)
246 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
248 # Wait for step dependencies to be met
250 deps = self.dependency_graph[S]
258 logger.info(" step %s self-wait skipped" % step.__name__)
261 cond = self.step_conditions[d]
263 if (self.step_status[d] is STEP_STATUS_WORKING):
264 logger.info(" step %s wait on dep %s" % (step.__name__, d))
267 go = self.step_status[d] == STEP_STATUS_OK
272 # SMBAKER: sync_step was not defined here, so I changed
273 # this from 'sync_step' to 'step'. Verify.
274 self.failed_steps.append(step)
275 my_status = STEP_STATUS_KO
277 sync_step = step(driver=self.driver,error_map=self.error_mapper)
278 sync_step.__name__ = step.__name__
279 sync_step.dependencies = []
281 mlist = sync_step.provides
284 sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
287 sync_step.debug_mode = debug_mode
291 # Various checks that decide whether
292 # this step runs or not
293 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
294 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
297 logger.info('Step not ready: %s'%sync_step.__name__)
298 self.failed_steps.append(sync_step)
299 my_status = STEP_STATUS_KO
301 logger.error('%r' % e)
302 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
303 self.failed_steps.append(sync_step)
304 my_status = STEP_STATUS_KO
308 duration=time.time() - start_time
310 logger.info('Executing step %s' % sync_step.__name__)
312 failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
314 self.check_duration(sync_step, duration)
317 self.failed_step_objects.update(failed_objects)
319 logger.info("Step %r succeeded" % step)
320 my_status = STEP_STATUS_OK
321 self.update_run_time(sync_step,deletion)
323 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))
325 self.failed_steps.append(S)
326 my_status = STEP_STATUS_KO
328 logger.info("Step %r succeeded due to non-run" % step)
329 my_status = STEP_STATUS_OK
332 my_cond = self.step_conditions[S]
334 self.step_status[S]=my_status
338 logger.info('Step %r is a leaf' % step)
342 if not self.driver.enabled:
345 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
350 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
351 self.error_mapper = ErrorMapper(error_map_file)
353 # Set of whole steps that failed
354 self.failed_steps = []
356 # Set of individual objects within steps that failed
357 self.failed_step_objects = set()
359 # Set up conditions and step status
360 # This is needed for steps to run in parallel
361 # while obeying dependencies.
364 for v in self.dependency_graph.values():
368 self.step_conditions = {}
369 self.step_status = {}
370 for p in list(providers):
371 self.step_conditions[p] = threading.Condition()
372 self.step_status[p] = STEP_STATUS_WORKING
375 logger.info('Waiting for event')
376 tBeforeWait = time.time()
377 self.wait_for_event(timeout=30)
378 logger.info('Observer woke up')
380 # Two passes. One for sync, the other for deletion.
381 for deletion in [False,True]:
383 logger.info('Deletion=%r...'%deletion)
384 schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
387 thread = threading.Thread(target=self.sync, args=(S, deletion))
389 logger.info('Deletion=%r...'%deletion)
390 threads.append(thread)
396 # Wait for all threads to finish before continuing with the run loop
400 self.save_run_times()
402 logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e)
403 logger.log_exc("Exception in observer run loop")
404 traceback.print_exc()