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 *
25 from openstack_observer.openstacksyncstep import OpenStackSyncStep
40 logger = Logger(level=logging.INFO)
42 class StepNotReady(Exception):
48 self.dependency_graph = None
64 class PlanetStackObserver:
65 #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivilege,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
70 # The Condition object that gets signalled by Feefie events
72 self.load_sync_step_modules()
73 self.load_sync_steps()
74 self.event_cond = threading.Condition()
76 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
77 if self.driver_kind=="openstack":
78 self.driver = OpenStackDriver()
80 self.driver = NoOpDriver()
82 def wait_for_event(self, timeout):
83 self.event_cond.acquire()
84 self.event_cond.wait(timeout)
85 self.event_cond.release()
88 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
89 self.event_cond.acquire()
90 self.event_cond.notify()
91 self.event_cond.release()
93 def load_sync_step_modules(self, step_dir=None):
95 if hasattr(Config(), "observer_steps_dir"):
96 step_dir = Config().observer_steps_dir
98 step_dir = "/opt/planetstack/observer/steps"
100 for fn in os.listdir(step_dir):
101 pathname = os.path.join(step_dir,fn)
102 if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
103 module = imp.load_source(fn[:-3],pathname)
104 for classname in dir(module):
105 c = getattr(module, classname, None)
107 # make sure 'c' is a descendent of SyncStep and has a
108 # provides field (this eliminates the abstract base classes
109 # since they don't have a provides)
111 if inspect.isclass(c) and (issubclass(c, SyncStep) or issubclass(c,OpenStackSyncStep)) and hasattr(c,"provides") and (c not in self.sync_steps):
112 self.sync_steps.append(c)
113 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
114 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
116 def load_sync_steps(self):
117 dep_path = Config().observer_dependency_graph
118 logger.info('Loading model dependency graph from %s' % dep_path)
120 # This contains dependencies between records, not sync steps
121 self.model_dependency_graph = json.loads(open(dep_path).read())
126 backend_path = Config().observer_pl_dependency_graph
127 logger.info('Loading backend dependency graph from %s' % backend_path)
128 # This contains dependencies between backend records
129 self.backend_dependency_graph = json.loads(open(backend_path).read())
131 logger.info('Backend dependency graph not loaded')
132 # We can work without a backend graph
133 self.backend_dependency_graph = {}
136 for s in self.sync_steps:
137 self.step_lookup[s.__name__] = s
140 provides_dict[m.__name__].append(s.__name__)
142 provides_dict[m.__name__]=[s.__name__]
145 for k,v in self.model_dependency_graph.iteritems():
147 for source in provides_dict[k]:
150 for dest in provides_dict[m]:
153 if (dest not in step_graph[source]):
154 step_graph[source].append(dest)
156 step_graph[source]=[dest]
162 # no dependencies, pass
166 if (self.backend_dependency_graph):
168 for s in self.sync_steps:
170 backend_dict[m]=s.__name__
172 for k,v in backend_dependency_graph.iteritems():
174 source = backend_dict[k]
177 dest = backend_dict[m]
181 step_graph[source]=dest
185 # no dependencies, pass
187 self.dependency_graph = step_graph
188 self.deletion_dependency_graph = invert_graph(step_graph)
190 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
191 print "Order of steps=",self.ordered_steps
192 self.load_run_times()
195 def check_duration(self, step, duration):
197 if (duration > step.deadline):
198 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
199 except AttributeError:
200 # S doesn't have a deadline
203 def update_run_time(self, step, deletion):
205 self.last_run_times[step.__name__]=time.time()
207 self.last_deletion_run_times[step.__name__]=time.time()
210 def check_schedule(self, step, deletion):
211 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
213 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
215 if (time_since_last_run < step.requested_interval):
217 except AttributeError:
218 logger.info('Step %s does not have requested_interval set'%step.__name__)
221 def load_run_times(self):
223 jrun_times = open('/tmp/observer_run_times').read()
224 self.last_run_times = json.loads(jrun_times)
226 self.last_run_times={}
227 for e in self.ordered_steps:
228 self.last_run_times[e]=0
230 jrun_times = open('/tmp/observer_deletion_run_times').read()
231 self.last_deletion_run_times = json.loads(jrun_times)
233 self.last_deletion_run_times={}
234 for e in self.ordered_steps:
235 self.last_deletion_run_times[e]=0
238 def save_run_times(self):
239 run_times = json.dumps(self.last_run_times)
240 open('/tmp/observer_run_times','w').write(run_times)
242 deletion_run_times = json.dumps(self.last_deletion_run_times)
243 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
245 def check_class_dependency(self, step, failed_steps):
246 step.dependenices = []
247 for obj in step.provides:
248 step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
249 for failed_step in failed_steps:
250 if (failed_step in step.dependencies):
253 def sync(self, S, deletion):
255 step = self.step_lookup[S]
256 start_time=time.time()
258 logger.info("Starting to work on step %s" % step.__name__)
260 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
262 # Wait for step dependencies to be met
264 deps = self.dependency_graph[S]
274 logger.info(" step %s self-wait skipped" % step.__name__)
278 cond = self.step_conditions[d]
280 if (self.step_status[d] is STEP_STATUS_WORKING):
281 logger.info(" step %s wait on dep %s" % (step.__name__, d))
284 go = go or self.step_status[d] == STEP_STATUS_OK
289 # SMBAKER: sync_step was not defined here, so I changed
290 # this from 'sync_step' to 'step'. Verify.
291 self.failed_steps.append(step)
292 my_status = STEP_STATUS_KO
294 sync_step = step(driver=self.driver,error_map=self.error_mapper)
295 sync_step.__name__ = step.__name__
296 sync_step.dependencies = []
298 mlist = sync_step.provides
301 sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
304 sync_step.debug_mode = debug_mode
308 # Various checks that decide whether
309 # this step runs or not
310 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
311 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
314 logger.info('Step not ready: %s'%sync_step.__name__)
315 self.failed_steps.append(sync_step)
316 my_status = STEP_STATUS_KO
318 logger.error('%r' % e)
319 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
320 self.failed_steps.append(sync_step)
321 my_status = STEP_STATUS_KO
325 duration=time.time() - start_time
327 logger.info('Executing step %s' % sync_step.__name__)
329 failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
331 self.check_duration(sync_step, duration)
334 self.failed_step_objects.update(failed_objects)
336 logger.info("Step %r succeeded" % step)
337 my_status = STEP_STATUS_OK
338 self.update_run_time(sync_step,deletion)
340 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))
342 self.failed_steps.append(S)
343 my_status = STEP_STATUS_KO
345 logger.info("Step %r succeeded due to non-run" % step)
346 my_status = STEP_STATUS_OK
349 my_cond = self.step_conditions[S]
351 self.step_status[S]=my_status
355 logger.info('Step %r is a leaf' % step)
361 if not self.driver.enabled:
364 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
369 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
370 self.error_mapper = ErrorMapper(error_map_file)
372 # Set of whole steps that failed
373 self.failed_steps = []
375 # Set of individual objects within steps that failed
376 self.failed_step_objects = set()
378 # Set up conditions and step status
379 # This is needed for steps to run in parallel
380 # while obeying dependencies.
383 for v in self.dependency_graph.values():
387 self.step_conditions = {}
388 self.step_status = {}
389 for p in list(providers):
390 self.step_conditions[p] = threading.Condition()
391 self.step_status[p] = STEP_STATUS_WORKING
394 logger.info('Waiting for event')
395 tBeforeWait = time.time()
396 self.wait_for_event(timeout=30)
397 logger.info('Observer woke up')
399 # Two passes. One for sync, the other for deletion.
400 for deletion in [False,True]:
402 logger.info('Deletion=%r...'%deletion)
403 schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
406 thread = threading.Thread(target=self.sync, args=(S, deletion))
408 logger.info('Deletion=%r...'%deletion)
409 threads.append(thread)
415 # Wait for all threads to finish before continuing with the run loop
419 self.save_run_times()
421 logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e)
422 logger.log_exc("Exception in observer run loop")
423 traceback.print_exc()