Map errors and feed them into the database.
[plstackapi.git] / planetstack / observer / event_loop.py
1 import os
2 import imp
3 import inspect
4 import time
5 import traceback
6 import commands
7 import threading
8 import json
9
10 from datetime import datetime
11 from collections import defaultdict
12 from core.models import *
13 from django.db.models import F, Q
14 #from openstack.manager import OpenStackManager
15 from openstack.driver import OpenStackDriver
16 from util.logger import Logger, logging, logger
17 #from timeout import timeout
18 from planetstack.config import Config
19 from observer.steps import *
20 from syncstep import SyncStep
21 from toposort import toposort
22 from observer.error_mapper import error_mapper
23
24 debug_mode = False
25
26 logger = Logger(level=logging.INFO)
27
28 class StepNotReady(Exception):
29         pass
30
31 class NoOpDriver:
32         def __init__(self):
33                  self.enabled = True
34
35 class 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):
174                 self.last_run_times[step.__name__]=time.time()
175
176         def check_schedule(self, step):
177                 time_since_last_run = time.time() - self.last_run_times.get(step.__name__, 0)
178                 try:
179                         if (time_since_last_run < step.requested_interval):
180                                 raise StepNotReady
181                 except AttributeError:
182                         logger.info('Step %s does not have requested_interval set'%step.__name__)
183                         raise StepNotReady
184         
185         def load_run_times(self):
186                 try:
187                         jrun_times = open('/tmp/observer_run_times').read()
188                         self.last_run_times = json.loads(jrun_times)
189                 except:
190                         self.last_run_times={}
191                         for e in self.ordered_steps:
192                                 self.last_run_times[e]=0
193
194
195         def save_run_times(self):
196                 run_times = json.dumps(self.last_run_times)
197                 open('/tmp/observer_run_times','w').write(run_times)
198
199         def check_class_dependency(self, step, failed_steps):
200                 step.dependenices = []
201                 for obj in step.provides:
202                         step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
203                 for failed_step in failed_steps:
204                         if (failed_step in step.dependencies):
205                                 raise StepNotReady
206
207         def run(self):
208                 if not self.driver.enabled:
209                         return
210                 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
211                         return
212
213                 while True:
214                         try:
215                                 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
216                                 error_mapper = ErrorMapper(error_map_file)
217
218                                 logger.info('Waiting for event')
219                                 tBeforeWait = time.time()
220                                 self.wait_for_event(timeout=30)
221                                 logger.info('Observer woke up')
222
223                                 # Set of whole steps that failed
224                                 failed_steps = []
225
226                                 # Set of individual objects within steps that failed
227                                 failed_step_objects = set()
228
229                                 for S in self.ordered_steps:
230                                         step = self.step_lookup[S]
231                                         start_time=time.time()
232                                         
233                                         sync_step = step(driver=self.driver,error_map=error_mapper)
234                                         sync_step.__name__ = step.__name__
235                                         sync_step.dependencies = []
236                                         try:
237                                                 mlist = sync_step.provides
238                                                 
239                                                 for m in mlist:
240                                                         sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
241                                         except KeyError:
242                                                 pass
243                                         sync_step.debug_mode = debug_mode
244
245                                         should_run = False
246                                         try:
247                                                 # Various checks that decide whether
248                                                 # this step runs or not
249                                                 self.check_class_dependency(sync_step, failed_steps) # dont run Slices if Sites failed
250                                                 self.check_schedule(sync_step) # dont run sync_network_routes if time since last run < 1 hour
251                                                 should_run = True
252                                         except StepNotReady:
253                                                 logging.info('Step not ready: %s'%sync_step.__name__)
254                                                 failed_steps.append(sync_step)
255                                         except Exception,e:
256                                                 logging.error('%r',e)
257                                                 logger.log_exc("sync step failed: %r!"%sync_step)
258                                                 failed_steps.append(sync_step)
259
260                                         if (should_run):
261                                                 try:
262                                                         duration=time.time() - start_time
263
264                                                         logger.info('Executing step %s' % sync_step.__name__)
265
266                                                         # ********* This is the actual sync step
267                                                         #import pdb
268                                                         #pdb.set_trace()
269                                                         failed_objects = sync_step(failed=list(failed_step_objects))
270
271
272                                                         self.check_duration(sync_step, duration)
273                                                         if failed_objects:
274                                                                 failed_step_objects.update(failed_objects)
275                                                         self.update_run_time(sync_step)
276                                                 except Exception,e:
277                                                         logging.error('Model step failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
278                                                         logger.log_exc(e)
279                                                         failed_steps.append(S)
280                                 self.save_run_times()
281                         except Exception, e:
282                                 logging.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
283                                 logger.log_exc("Exception in observer run loop")
284                                 traceback.print_exc()