blob: e703869a4c369bd63edb5eb6de76970ede3b41b0 [file] [log] [blame]
Sapan Bhatia26d40bc2014-05-12 15:28:02 -04001import os
2import imp
3import inspect
4import time
5import traceback
6import commands
7import threading
8import json
9
10from datetime import datetime
11from collections import defaultdict
12from core.models import *
13from django.db.models import F, Q
14#from openstack.manager import OpenStackManager
15from openstack.driver import OpenStackDriver
16from util.logger import Logger, logging, logger
17#from timeout import timeout
18from planetstack.config import Config
19from ec2_observer.steps import *
20from syncstep import SyncStep
21from toposort import toposort
Sapan Bhatiab7262192014-07-22 00:30:16 -040022from ec2_observer.error_mapper import *
Sapan Bhatia26d40bc2014-05-12 15:28:02 -040023
24debug_mode = False
25
26logger = Logger(level=logging.INFO)
27
28class StepNotReady(Exception):
29 pass
30
31class NoOpDriver:
32 def __init__(self):
33 self.enabled = True
34
35class PlanetStackObserver:
36 #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
37 sync_steps = []
38
39 def __init__(self):
40 # The Condition object that gets signalled by Feefie events
41 self.step_lookup = {}
42 self.load_sync_step_modules()
43 self.load_sync_steps()
44 self.event_cond = threading.Condition()
45
46 self.driver_kind = getattr(Config(), "observer_driver", "openstack")
47 if self.driver_kind=="openstack":
48 self.driver = OpenStackDriver()
49 else:
50 self.driver = NoOpDriver()
51
52 def wait_for_event(self, timeout):
53 self.event_cond.acquire()
54 self.event_cond.wait(timeout)
55 self.event_cond.release()
56
57 def wake_up(self):
58 logger.info('Wake up routine called. Event cond %r'%self.event_cond)
59 self.event_cond.acquire()
60 self.event_cond.notify()
61 self.event_cond.release()
62
63 def load_sync_step_modules(self, step_dir=None):
64 if step_dir is None:
65 if hasattr(Config(), "observer_steps_dir"):
66 step_dir = Config().observer_steps_dir
67 else:
68 step_dir = "/opt/planetstack/observer/steps"
69
70 for fn in os.listdir(step_dir):
71 pathname = os.path.join(step_dir,fn)
72 if os.path.isfile(pathname) and fn.endswith(".py") and (fn!="__init__.py"):
73 module = imp.load_source(fn[:-3],pathname)
74 for classname in dir(module):
75 c = getattr(module, classname, None)
76
77 # make sure 'c' is a descendent of SyncStep and has a
78 # provides field (this eliminates the abstract base classes
79 # since they don't have a provides)
80
81 if inspect.isclass(c) and issubclass(c, SyncStep) and hasattr(c,"provides") and (c not in self.sync_steps):
82 self.sync_steps.append(c)
83 logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]))
84 # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])
85
86 def load_sync_steps(self):
87 dep_path = Config().observer_dependency_graph
88 logger.info('Loading model dependency graph from %s' % dep_path)
89 try:
90 # This contains dependencies between records, not sync steps
91 self.model_dependency_graph = json.loads(open(dep_path).read())
92 except Exception,e:
93 raise e
94
95 try:
96 backend_path = Config().observer_pl_dependency_graph
97 logger.info('Loading backend dependency graph from %s' % backend_path)
98 # This contains dependencies between backend records
99 self.backend_dependency_graph = json.loads(open(backend_path).read())
100 except Exception,e:
101 logger.info('Backend dependency graph not loaded')
102 # We can work without a backend graph
103 self.backend_dependency_graph = {}
104
105 provides_dict = {}
106 for s in self.sync_steps:
107 self.step_lookup[s.__name__] = s
108 for m in s.provides:
109 try:
110 provides_dict[m.__name__].append(s.__name__)
111 except KeyError:
112 provides_dict[m.__name__]=[s.__name__]
113
114
115 step_graph = {}
116 for k,v in self.model_dependency_graph.iteritems():
117 try:
118 for source in provides_dict[k]:
119 for m in v:
120 try:
121 for dest in provides_dict[m]:
122 # no deps, pass
123 try:
124 if (dest not in step_graph[source]):
125 step_graph[source].append(dest)
126 except:
127 step_graph[source]=[dest]
128 except KeyError:
129 pass
130
131 except KeyError:
132 pass
133 # no dependencies, pass
134
135 #import pdb
136 #pdb.set_trace()
137 if (self.backend_dependency_graph):
138 backend_dict = {}
139 for s in self.sync_steps:
140 for m in s.serves:
141 backend_dict[m]=s.__name__
142
143 for k,v in backend_dependency_graph.iteritems():
144 try:
145 source = backend_dict[k]
146 for m in v:
147 try:
148 dest = backend_dict[m]
149 except KeyError:
150 # no deps, pass
151 pass
152 step_graph[source]=dest
153
154 except KeyError:
155 pass
156 # no dependencies, pass
157
158 dependency_graph = step_graph
159
160 self.ordered_steps = toposort(dependency_graph, map(lambda s:s.__name__,self.sync_steps))
161 print "Order of steps=",self.ordered_steps
162 self.load_run_times()
163
164
165 def check_duration(self, step, duration):
166 try:
167 if (duration > step.deadline):
168 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
169 except AttributeError:
170 # S doesn't have a deadline
171 pass
172
173 def update_run_time(self, step, deletion):
174 if (not deletion):
175 self.last_run_times[step.__name__]=time.time()
176 else:
177 self.last_deletion_run_times[step.__name__]=time.time()
178
179
180 def check_schedule(self, step, deletion):
181 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
182
183 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
184 try:
185 if (time_since_last_run < step.requested_interval):
186 raise StepNotReady
187 except AttributeError:
188 logger.info('Step %s does not have requested_interval set'%step.__name__)
189 raise StepNotReady
190
191 def load_run_times(self):
192 try:
193 jrun_times = open('/tmp/observer_run_times').read()
194 self.last_run_times = json.loads(jrun_times)
195 except:
196 self.last_run_times={}
197 for e in self.ordered_steps:
198 self.last_run_times[e]=0
199 try:
200 jrun_times = open('/tmp/observer_deletion_run_times').read()
201 self.last_deletion_run_times = json.loads(jrun_times)
202 except:
203 self.last_deletion_run_times={}
204 for e in self.ordered_steps:
205 self.last_deletion_run_times[e]=0
206
207
208
209 def save_run_times(self):
210 run_times = json.dumps(self.last_run_times)
211 open('/tmp/observer_run_times','w').write(run_times)
212
213 deletion_run_times = json.dumps(self.last_deletion_run_times)
214 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
215
216 def check_class_dependency(self, step, failed_steps):
217 step.dependenices = []
218 for obj in step.provides:
219 step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
220 for failed_step in failed_steps:
221 if (failed_step in step.dependencies):
222 raise StepNotReady
223
224 def run(self):
225 if not self.driver.enabled:
226 return
227 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
228 return
229
230 while True:
231 try:
232 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
233 error_mapper = ErrorMapper(error_map_file)
234
235 logger.info('Waiting for event')
236 tBeforeWait = time.time()
Sapan Bhatiae7e4ca12014-07-22 01:27:02 -0400237 self.wait_for_event(timeout=5)
Sapan Bhatia26d40bc2014-05-12 15:28:02 -0400238 logger.info('Observer woke up')
239
240 # Two passes. One for sync, the other for deletion.
241 for deletion in (False,True):
Sapan Bhatiabab33762014-07-22 01:21:36 -0400242 logger.info('Deletion=%r...'%deletion)
Sapan Bhatia26d40bc2014-05-12 15:28:02 -0400243 # Set of whole steps that failed
244 failed_steps = []
245
246 # Set of individual objects within steps that failed
247 failed_step_objects = set()
248
249 ordered_steps = self.ordered_steps if not deletion else reversed(self.ordered_steps)
250
251 for S in ordered_steps:
252 step = self.step_lookup[S]
253 start_time=time.time()
254
255 sync_step = step(driver=self.driver,error_map=error_mapper)
256 sync_step.__name__ = step.__name__
257 sync_step.dependencies = []
258 try:
259 mlist = sync_step.provides
260
261 for m in mlist:
262 sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
263 except KeyError:
264 pass
265 sync_step.debug_mode = debug_mode
266
267 should_run = False
268 try:
269 # Various checks that decide whether
270 # this step runs or not
271 self.check_class_dependency(sync_step, failed_steps) # dont run Slices if Sites failed
Sapan Bhatia35ecc412014-07-22 00:27:35 -0400272 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
Sapan Bhatia26d40bc2014-05-12 15:28:02 -0400273 should_run = True
274 except StepNotReady:
275 logging.info('Step not ready: %s'%sync_step.__name__)
276 failed_steps.append(sync_step)
277 except Exception,e:
278 logging.error('%r',e)
279 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
280 failed_steps.append(sync_step)
281
282 if (should_run):
283 try:
284 duration=time.time() - start_time
285
286 logger.info('Executing step %s' % sync_step.__name__)
287
288 # ********* This is the actual sync step
289 #import pdb
290 #pdb.set_trace()
291 failed_objects = sync_step(failed=list(failed_step_objects), deletion=deletion)
292
293
294 self.check_duration(sync_step, duration)
295 if failed_objects:
296 failed_step_objects.update(failed_objects)
297
Sapan Bhatia06ca5082014-07-22 01:22:00 -0400298 self.update_run_time(sync_step,deletion)
Sapan Bhatia26d40bc2014-05-12 15:28:02 -0400299 except Exception,e:
300 logging.error('Model step failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
301 logger.log_exc(e)
302 failed_steps.append(S)
303 self.save_run_times()
304 except Exception, e:
305 logging.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
306 logger.log_exc("Exception in observer run loop")
307 traceback.print_exc()