blob: 9e8c7b4d5b7c8454fa443fad9f1cc874c79c9b5b [file] [log] [blame]
Scott Baker45fb7a12013-12-31 00:56:19 -08001import os
2import imp
3import inspect
Sapan Bhatia24836f12013-08-27 10:16:05 -04004import time
5import traceback
6import commands
7import threading
8import json
Sapan Bhatiaab202a62014-09-03 11:30:21 -04009import pdb
Sapan Bhatia24836f12013-08-27 10:16:05 -040010
11from datetime import datetime
12from collections import defaultdict
13from core.models import *
14from django.db.models import F, Q
Tony Mack387a73f2013-09-18 07:59:14 -040015#from openstack.manager import OpenStackManager
16from openstack.driver import OpenStackDriver
Sapan Bhatia24836f12013-08-27 10:16:05 -040017from util.logger import Logger, logging, logger
18#from timeout import timeout
Sapan Bhatia757e0b62013-09-02 16:55:00 -040019from planetstack.config import Config
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040020from observer.steps import *
Scott Baker45fb7a12013-12-31 00:56:19 -080021from syncstep import SyncStep
Sapan Bhatia45cbbc32014-03-11 17:48:30 -040022from toposort import toposort
Sapan Bhatia13d89152014-07-23 10:35:33 -040023from observer.error_mapper import *
Sapan Bhatia24836f12013-08-27 10:16:05 -040024
Sapan Bhatia13c7f112013-09-02 14:19:35 -040025debug_mode = False
Sapan Bhatia24836f12013-08-27 10:16:05 -040026
Andy Bavier04111b72013-10-22 16:47:10 -040027logger = Logger(level=logging.INFO)
Sapan Bhatia24836f12013-08-27 10:16:05 -040028
Sapan Bhatia13c7f112013-09-02 14:19:35 -040029class StepNotReady(Exception):
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040030 pass
Sapan Bhatia24836f12013-08-27 10:16:05 -040031
Scott Baker7771f412014-01-02 16:36:41 -080032class NoOpDriver:
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040033 def __init__(self):
34 self.enabled = True
Sapan Bhatiaab202a62014-09-03 11:30:21 -040035 self.dependency_graph = None
36
37STEP_STATUS_WORKING=1
38STEP_STATUS_OK=2
39STEP_STATUS_KO=3
40
41def invert_graph(g):
42 ig = {}
43 for k,v in g.items():
44 for v0 in v:
45 try:
46 ig[v0].append(k)
47 except:
48 ig=[k]
49 return ig
Scott Baker7771f412014-01-02 16:36:41 -080050
Sapan Bhatia24836f12013-08-27 10:16:05 -040051class PlanetStackObserver:
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040052 #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
53 sync_steps = []
Sapan Bhatia24836f12013-08-27 10:16:05 -040054
Sapan Bhatiaab202a62014-09-03 11:30:21 -040055
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040056 def __init__(self):
57 # The Condition object that gets signalled by Feefie events
58 self.step_lookup = {}
59 self.load_sync_step_modules()
60 self.load_sync_steps()
61 self.event_cond = threading.Condition()
Scott Baker7771f412014-01-02 16:36:41 -080062
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040063 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
64 if self.driver_kind=="openstack":
65 self.driver = OpenStackDriver()
66 else:
67 self.driver = NoOpDriver()
Sapan Bhatia24836f12013-08-27 10:16:05 -040068
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040069 def wait_for_event(self, timeout):
70 self.event_cond.acquire()
71 self.event_cond.wait(timeout)
72 self.event_cond.release()
Scott Baker45fb7a12013-12-31 00:56:19 -080073
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040074 def wake_up(self):
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()
Sapan Bhatia24836f12013-08-27 10:16:05 -040079
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040080 def load_sync_step_modules(self, step_dir=None):
81 if step_dir is None:
82 if hasattr(Config(), "observer_steps_dir"):
83 step_dir = Config().observer_steps_dir
84 else:
85 step_dir = "/opt/planetstack/observer/steps"
Scott Baker45fb7a12013-12-31 00:56:19 -080086
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040087 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)
Scott Baker45fb7a12013-12-31 00:56:19 -080093
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040094 # 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)
Scott Baker45fb7a12013-12-31 00:56:19 -080097
Sapan Bhatiaf73664b2014-04-28 13:07:18 -040098 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])
Scott Baker45fb7a12013-12-31 00:56:19 -0800102
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400103 def load_sync_steps(self):
104 dep_path = Config().observer_dependency_graph
105 logger.info('Loading model dependency graph from %s' % dep_path)
106 try:
107 # This contains dependencies between records, not sync steps
108 self.model_dependency_graph = json.loads(open(dep_path).read())
109 except Exception,e:
110 raise e
Sapan Bhatia24836f12013-08-27 10:16:05 -0400111
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400112 try:
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())
117 except Exception,e:
118 logger.info('Backend dependency graph not loaded')
119 # We can work without a backend graph
120 self.backend_dependency_graph = {}
Sapan Bhatia24836f12013-08-27 10:16:05 -0400121
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400122 provides_dict = {}
123 for s in self.sync_steps:
124 self.step_lookup[s.__name__] = s
125 for m in s.provides:
126 try:
127 provides_dict[m.__name__].append(s.__name__)
128 except KeyError:
129 provides_dict[m.__name__]=[s.__name__]
Sapan Bhatia04c94ad2013-09-02 18:00:28 -0400130
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400131 step_graph = {}
132 for k,v in self.model_dependency_graph.iteritems():
133 try:
134 for source in provides_dict[k]:
135 for m in v:
136 try:
137 for dest in provides_dict[m]:
138 # no deps, pass
139 try:
140 if (dest not in step_graph[source]):
141 step_graph[source].append(dest)
142 except:
143 step_graph[source]=[dest]
144 except KeyError:
145 pass
146
147 except KeyError:
148 pass
149 # no dependencies, pass
150
151 #import pdb
152 #pdb.set_trace()
153 if (self.backend_dependency_graph):
154 backend_dict = {}
155 for s in self.sync_steps:
156 for m in s.serves:
157 backend_dict[m]=s.__name__
158
159 for k,v in backend_dependency_graph.iteritems():
160 try:
161 source = backend_dict[k]
162 for m in v:
163 try:
164 dest = backend_dict[m]
165 except KeyError:
166 # no deps, pass
167 pass
168 step_graph[source]=dest
169
170 except KeyError:
171 pass
172 # no dependencies, pass
Sapan Bhatia24836f12013-08-27 10:16:05 -0400173
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400174 self.dependency_graph = step_graph
175 self.deletion_dependency_graph = invert_graph(step_graph)
Sapan Bhatia24836f12013-08-27 10:16:05 -0400176
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400177 self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps))
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400178 print "Order of steps=",self.ordered_steps
179 self.load_run_times()
180
Sapan Bhatia24836f12013-08-27 10:16:05 -0400181
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400182 def check_duration(self, step, duration):
183 try:
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
188 pass
Sapan Bhatia24836f12013-08-27 10:16:05 -0400189
Sapan Bhatia285decb2014-04-30 00:31:44 -0400190 def update_run_time(self, step, deletion):
191 if (not deletion):
192 self.last_run_times[step.__name__]=time.time()
193 else:
194 self.last_deletion_run_times[step.__name__]=time.time()
Sapan Bhatia13c7f112013-09-02 14:19:35 -0400195
Sapan Bhatia285decb2014-04-30 00:31:44 -0400196
197 def check_schedule(self, step, deletion):
198 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
199
200 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400201 try:
202 if (time_since_last_run < step.requested_interval):
203 raise StepNotReady
204 except AttributeError:
205 logger.info('Step %s does not have requested_interval set'%step.__name__)
206 raise StepNotReady
207
208 def load_run_times(self):
209 try:
210 jrun_times = open('/tmp/observer_run_times').read()
211 self.last_run_times = json.loads(jrun_times)
212 except:
213 self.last_run_times={}
214 for e in self.ordered_steps:
215 self.last_run_times[e]=0
Sapan Bhatia285decb2014-04-30 00:31:44 -0400216 try:
217 jrun_times = open('/tmp/observer_deletion_run_times').read()
218 self.last_deletion_run_times = json.loads(jrun_times)
219 except:
220 self.last_deletion_run_times={}
221 for e in self.ordered_steps:
222 self.last_deletion_run_times[e]=0
223
Sapan Bhatia36938ca2013-09-02 14:35:24 -0400224
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400225 def save_run_times(self):
226 run_times = json.dumps(self.last_run_times)
227 open('/tmp/observer_run_times','w').write(run_times)
Sapan Bhatia36938ca2013-09-02 14:35:24 -0400228
Sapan Bhatia285decb2014-04-30 00:31:44 -0400229 deletion_run_times = json.dumps(self.last_deletion_run_times)
230 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
231
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400232 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
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400240 def sync(self, S, deletion):
241 step = self.step_lookup[S]
242 start_time=time.time()
Scott Bakeradc73172014-09-04 10:36:51 -0700243
244 logger.info("Starting to work on step %s" % step.__name__)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400245
246 dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph
Sapan Bhatia51f48932014-08-25 04:17:12 -0400247
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400248 # Wait for step dependencies to be met
249 try:
250 deps = self.dependency_graph[S]
251 has_deps = True
252 except KeyError:
253 has_deps = False
Sapan Bhatia51f48932014-08-25 04:17:12 -0400254
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400255 if (has_deps):
256 for d in deps:
Scott Bakeradc73172014-09-04 10:36:51 -0700257 if d==step.__name__:
258 logger.info(" step %s self-wait skipped" % step.__name__)
259 continue
260
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400261 cond = self.step_conditions[d]
262 cond.acquire()
263 if (self.step_status[d] is STEP_STATUS_WORKING):
Scott Bakeradc73172014-09-04 10:36:51 -0700264 logger.info(" step %s wait on dep %s" % (step.__name__, d))
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400265 cond.wait()
266 cond.release()
267 go = self.step_status[d] == STEP_STATUS_OK
268 else:
269 go = True
270
271 if (not go):
Scott Bakeradc73172014-09-04 10:36:51 -0700272 # SMBAKER: sync_step was not defined here, so I changed
273 # this from 'sync_step' to 'step'. Verify.
274 self.failed_steps.append(step)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400275 my_status = STEP_STATUS_KO
276 else:
277 sync_step = step(driver=self.driver,error_map=self.error_mapper)
Sapan Bhatia51f48932014-08-25 04:17:12 -0400278 sync_step.__name__ = step.__name__
279 sync_step.dependencies = []
280 try:
281 mlist = sync_step.provides
Scott Bakeradc73172014-09-04 10:36:51 -0700282
Sapan Bhatia51f48932014-08-25 04:17:12 -0400283 for m in mlist:
284 sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
285 except KeyError:
286 pass
287 sync_step.debug_mode = debug_mode
288
289 should_run = False
290 try:
291 # Various checks that decide whether
292 # this step runs or not
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400293 self.check_class_dependency(sync_step, self.failed_steps) # dont run Slices if Sites failed
Sapan Bhatia51f48932014-08-25 04:17:12 -0400294 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
295 should_run = True
296 except StepNotReady:
Scott Bakeradc73172014-09-04 10:36:51 -0700297 logger.info('Step not ready: %s'%sync_step.__name__)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400298 self.failed_steps.append(sync_step)
299 my_status = STEP_STATUS_KO
Sapan Bhatia51f48932014-08-25 04:17:12 -0400300 except Exception,e:
Scott Bakeradc73172014-09-04 10:36:51 -0700301 logger.error('%r' % e)
Sapan Bhatia51f48932014-08-25 04:17:12 -0400302 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400303 self.failed_steps.append(sync_step)
304 my_status = STEP_STATUS_KO
Sapan Bhatia51f48932014-08-25 04:17:12 -0400305
306 if (should_run):
307 try:
308 duration=time.time() - start_time
309
310 logger.info('Executing step %s' % sync_step.__name__)
311
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400312 failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion)
Sapan Bhatia51f48932014-08-25 04:17:12 -0400313
314 self.check_duration(sync_step, duration)
Sapan Bhatia51f48932014-08-25 04:17:12 -0400315
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400316 if failed_objects:
317 self.failed_step_objects.update(failed_objects)
318
Scott Bakeradc73172014-09-04 10:36:51 -0700319 logger.info("Step %r succeeded" % step)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400320 my_status = STEP_STATUS_OK
Sapan Bhatia51f48932014-08-25 04:17:12 -0400321 self.update_run_time(sync_step,deletion)
322 except Exception,e:
Scott Bakeradc73172014-09-04 10:36:51 -0700323 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))
Sapan Bhatia51f48932014-08-25 04:17:12 -0400324 logger.log_exc(e)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400325 self.failed_steps.append(S)
326 my_status = STEP_STATUS_KO
327 else:
Scott Bakeradc73172014-09-04 10:36:51 -0700328 logger.info("Step %r succeeded due to non-run" % step)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400329 my_status = STEP_STATUS_OK
Scott Bakeradc73172014-09-04 10:36:51 -0700330
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400331 try:
332 my_cond = self.step_conditions[S]
333 my_cond.acquire()
334 self.step_status[S]=my_status
335 my_cond.notify_all()
336 my_cond.release()
337 except KeyError,e:
Scott Bakeradc73172014-09-04 10:36:51 -0700338 logger.info('Step %r is a leaf' % step)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400339 pass
340
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400341 def run(self):
342 if not self.driver.enabled:
343 return
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400344
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400345 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
346 return
347
348 while True:
349 try:
Sapan Bhatia31ebe5c2014-04-29 00:24:09 -0400350 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400351 self.error_mapper = ErrorMapper(error_map_file)
352
353 # Set of whole steps that failed
354 self.failed_steps = []
355
356 # Set of individual objects within steps that failed
357 self.failed_step_objects = set()
358
359 # Set up conditions and step status
360 # This is needed for steps to run in parallel
361 # while obeying dependencies.
362
363 providers = set()
364 for v in self.dependency_graph.values():
365 if (v):
366 providers.update(v)
367
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
373
Sapan Bhatia31ebe5c2014-04-29 00:24:09 -0400374
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400375 logger.info('Waiting for event')
376 tBeforeWait = time.time()
Sapan Bhatia13d89152014-07-23 10:35:33 -0400377 self.wait_for_event(timeout=30)
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400378 logger.info('Observer woke up')
379
Sapan Bhatia285decb2014-04-30 00:31:44 -0400380 # Two passes. One for sync, the other for deletion.
Sapan Bhatia0f727b82014-08-18 02:44:20 -0400381 for deletion in [False,True]:
Sapan Bhatia51f48932014-08-25 04:17:12 -0400382 threads = []
Sapan Bhatiae82f5e52014-07-23 10:02:45 -0400383 logger.info('Deletion=%r...'%deletion)
Sapan Bhatiaab202a62014-09-03 11:30:21 -0400384 schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps)
385
386 for S in schedule:
387 thread = threading.Thread(target=self.sync, args=(S, deletion))
388
389 logger.info('Deletion=%r...'%deletion)
390 threads.append(thread)
Sapan Bhatia285decb2014-04-30 00:31:44 -0400391
Sapan Bhatia51f48932014-08-25 04:17:12 -0400392 # Start threads
393 for t in threads:
394 t.start()
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400395
Sapan Bhatia51f48932014-08-25 04:17:12 -0400396 # Wait for all threads to finish before continuing with the run loop
397 for t in threads:
398 t.join()
Sapan Bhatia285decb2014-04-30 00:31:44 -0400399
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400400 self.save_run_times()
401 except Exception, e:
Scott Bakeradc73172014-09-04 10:36:51 -0700402 logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e)
Sapan Bhatiaf73664b2014-04-28 13:07:18 -0400403 logger.log_exc("Exception in observer run loop")
404 traceback.print_exc()