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 django.db import connection
16 #from openstack.manager import OpenStackManager
17 from openstack.driver import OpenStackDriver
18 from util.logger import Logger, logging, logger
19 #from timeout import timeout
20 from planetstack.config import Config
21 from observer.steps import *
22 from syncstep import SyncStep
23 from toposort import toposort
24 from observer.error_mapper import *
28 logger = Logger(level=logging.INFO)
30 class StepNotReady(Exception):
36 self.dependency_graph = None
52 class PlanetStackObserver:
53 #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
58 # The Condition object that gets signalled by Feefie events
60 self.load_sync_step_modules()
61 self.load_sync_steps()
62 self.event_cond = threading.Condition()
64 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
65 if self.driver_kind=="openstack":
66 self.driver = OpenStackDriver()
68 self.driver = NoOpDriver()
70 def wait_for_event(self, timeout):
71 self.event_cond.acquire()
72 self.event_cond.wait(timeout)
73 self.event_cond.release()
76 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
77 self.event_cond.acquire()
78 self.event_cond.notify()
79 self.event_cond.release()
81 def load_sync_step_modules(self, step_dir=None):
83 if hasattr(Config(), "observer_steps_dir"):
84 step_dir = Config().observer_steps_dir
86 step_dir = "/opt/planetstack/observer/steps"
88 for fn in os.listdir(step_dir):
89 pathname = os.path.join(step_dir,fn)
90 if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
91 module = imp.load_source(fn[:-3],pathname)
92 for classname in dir(module):
93 c = getattr(module, classname, None)
95 # make sure 'c' is a descendent of SyncStep and has a
96 # provides field (this eliminates the abstract base classes
97 # since they don't have a provides)
99 if inspect.isclass(c) and issubclass(c, SyncStep) and hasattr(c,"provides") and (c not in self.sync_steps):
100 self.sync_steps.append(c)
101 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
102 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
104 def load_sync_steps(self):
105 dep_path = Config().observer_dependency_graph
106 logger.info('Loading model dependency graph from %s' % dep_path)
108 # This contains dependencies between records, not sync steps
109 self.model_dependency_graph = json.loads(open(dep_path).read())
114 backend_path = Config().observer_pl_dependency_graph
115 logger.info('Loading backend dependency graph from %s' % backend_path)
116 # This contains dependencies between backend records
117 self.backend_dependency_graph = json.loads(open(backend_path).read())
119 logger.info('Backend dependency graph not loaded')
120 # We can work without a backend graph
121 self.backend_dependency_graph = {}
124 for s in self.sync_steps:
125 self.step_lookup[s.__name__] = s
128 provides_dict[m.__name__].append(s.__name__)
130 provides_dict[m.__name__]=[s.__name__]
133 for k,v in self.model_dependency_graph.iteritems():
135 for source in provides_dict[k]:
138 for dest in provides_dict[m]:
141 if (dest not in step_graph[source]):
142 step_graph[source].append(dest)
144 step_graph[source]=[dest]
150 # no dependencies, pass
154 if (self.backend_dependency_graph):
156 for s in self.sync_steps:
158 backend_dict[m]=s.__name__
160 for k,v in backend_dependency_graph.iteritems():
162 source = backend_dict[k]
165 dest = backend_dict[m]
169 step_graph[source]=dest
173 # no dependencies, pass
175 self.dependency_graph = step_graph
176 self.deletion_dependency_graph = invert_graph(step_graph)
178 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
179 print "Order of steps=",self.ordered_steps
180 self.load_run_times()
183 def check_duration(self, step, duration):
185 if (duration > step.deadline):
186 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
187 except AttributeError:
188 # S doesn't have a deadline
191 def update_run_time(self, step, deletion):
193 self.last_run_times[step.__name__]=time.time()
195 self.last_deletion_run_times[step.__name__]=time.time()
198 def check_schedule(self, step, deletion):
199 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
201 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
203 if (time_since_last_run < step.requested_interval):
205 except AttributeError:
206 logger.info('Step %s does not have requested_interval set'%step.__name__)
209 def load_run_times(self):
211 jrun_times = open('/tmp/observer_run_times').read()
212 self.last_run_times = json.loads(jrun_times)
214 self.last_run_times={}
215 for e in self.ordered_steps:
216 self.last_run_times[e]=0
218 jrun_times = open('/tmp/observer_deletion_run_times').read()
219 self.last_deletion_run_times = json.loads(jrun_times)
221 self.last_deletion_run_times={}
222 for e in self.ordered_steps:
223 self.last_deletion_run_times[e]=0
226 def save_run_times(self):
227 run_times = json.dumps(self.last_run_times)
228 open('/tmp/observer_run_times','w').write(run_times)
230 deletion_run_times = json.dumps(self.last_deletion_run_times)
231 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
233 def check_class_dependency(self, step, failed_steps):
234 step.dependenices = []
235 for obj in step.provides:
236 step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
237 for failed_step in failed_steps:
238 if (failed_step in step.dependencies):
241 def sync(self, S, deletion):
243 step = self.step_lookup[S]
244 start_time=time.time()
246 logger.info("Starting to work on step %s" % step.__name__)
248 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
250 # Wait for step dependencies to be met
252 deps = self.dependency_graph[S]
260 logger.info(" step %s self-wait skipped" % step.__name__)
263 cond = self.step_conditions[d]
265 if (self.step_status[d] is STEP_STATUS_WORKING):
266 logger.info(" step %s wait on dep %s" % (step.__name__, d))
269 go = self.step_status[d] == STEP_STATUS_OK
274 # SMBAKER: sync_step was not defined here, so I changed
275 # this from 'sync_step' to 'step'. Verify.
276 self.failed_steps.append(step)
277 my_status = STEP_STATUS_KO
279 sync_step = step(driver=self.driver,error_map=self.error_mapper)
280 sync_step.__name__ = step.__name__
281 sync_step.dependencies = []
283 mlist = sync_step.provides
286 sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
289 sync_step.debug_mode = debug_mode
293 # Various checks that decide whether
294 # this step runs or not
295 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
296 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
299 logger.info('Step not ready: %s'%sync_step.__name__)
300 self.failed_steps.append(sync_step)
301 my_status = STEP_STATUS_KO
303 logger.error('%r' % e)
304 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
305 self.failed_steps.append(sync_step)
306 my_status = STEP_STATUS_KO
310 duration=time.time() - start_time
312 logger.info('Executing step %s' % sync_step.__name__)
314 failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
316 self.check_duration(sync_step, duration)
319 self.failed_step_objects.update(failed_objects)
321 logger.info("Step %r succeeded" % step)
322 my_status = STEP_STATUS_OK
323 self.update_run_time(sync_step,deletion)
325 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))
327 self.failed_steps.append(S)
328 my_status = STEP_STATUS_KO
330 logger.info("Step %r succeeded due to non-run" % step)
331 my_status = STEP_STATUS_OK
334 my_cond = self.step_conditions[S]
336 self.step_status[S]=my_status
340 logger.info('Step %r is a leaf' % step)
346 if not self.driver.enabled:
349 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
354 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
355 self.error_mapper = ErrorMapper(error_map_file)
357 # Set of whole steps that failed
358 self.failed_steps = []
360 # Set of individual objects within steps that failed
361 self.failed_step_objects = set()
363 # Set up conditions and step status
364 # This is needed for steps to run in parallel
365 # while obeying dependencies.
368 for v in self.dependency_graph.values():
372 self.step_conditions = {}
373 self.step_status = {}
374 for p in list(providers):
375 self.step_conditions[p] = threading.Condition()
376 self.step_status[p] = STEP_STATUS_WORKING
379 logger.info('Waiting for event')
380 tBeforeWait = time.time()
381 self.wait_for_event(timeout=30)
382 logger.info('Observer woke up')
384 # Two passes. One for sync, the other for deletion.
385 for deletion in [False,True]:
387 logger.info('Deletion=%r...'%deletion)
388 schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
391 thread = threading.Thread(target=self.sync, args=(S, deletion))
393 logger.info('Deletion=%r...'%deletion)
394 threads.append(thread)
400 # Wait for all threads to finish before continuing with the run loop
404 self.save_run_times()
406 logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e)
407 logger.log_exc("Exception in observer run loop")
408 traceback.print_exc()