planetstack/openstack_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 *
  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                 step_graph = {}
 115                 for k,v in self.model_dependency_graph.iteritems():
 116                         try:
 117                                 for source in provides_dict[k]:
 118                                         for m in v:
 119                                                 try:
 120                                                         for dest in provides_dict[m]:
 121                                                                 # no deps, pass
 122                                                                 try:
 123                                                                         if (dest not in step_graph[source]):
 124                                                                                 step_graph[source].append(dest)
 125                                                                 except:
 126                                                                         step_graph[source]=[dest]
 127                                                 except KeyError:
 128                                                         pass
 129
 130                         except KeyError:
 131                                 pass
 132                                 # no dependencies, pass
 133
 134                 #import pdb
 135                 #pdb.set_trace()
 136                 if (self.backend_dependency_graph):
 137                         backend_dict = {}
 138                         for s in self.sync_steps:
 139                                 for m in s.serves:
 140                                         backend_dict[m]=s.__name__
 141
 142                         for k,v in backend_dependency_graph.iteritems():
 143                                 try:
 144                                         source = backend_dict[k]
 145                                         for m in v:
 146                                                 try:
 147                                                         dest = backend_dict[m]
 148                                                 except KeyError:
 149                                                         # no deps, pass
 150                                                         pass
 151                                                 step_graph[source]=dest
 152
 153                                 except KeyError:
 154                                         pass
 155                                         # no dependencies, pass
 156
 157                 dependency_graph = step_graph
 158
 159                 self.ordered_steps = toposort(dependency_graph, map(lambda s:s.__name__,self.sync_steps))
 160                 print "Order of steps=",self.ordered_steps
 161                 self.load_run_times()
 162
 163
 164         def check_duration(self, step, duration):
 165                 try:
 166                         if (duration > step.deadline):
 167                                 logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
 168                 except AttributeError:
 169                         # S doesn't have a deadline
 170                         pass
 171
 172         def update_run_time(self, step, deletion):
 173                 if (not deletion):
 174                         self.last_run_times[step.__name__]=time.time()
 175                 else:
 176                         self.last_deletion_run_times[step.__name__]=time.time()
 177
 178
 179         def check_schedule(self, step, deletion):
 180                 last_run_times = self.last_run_times if not deletion else self.last_deletion_run_times
 181
 182                 time_since_last_run = time.time() - last_run_times.get(step.__name__, 0)
 183                 try:
 184                         if (time_since_last_run < step.requested_interval):
 185                                 raise StepNotReady
 186                 except AttributeError:
 187                         logger.info('Step %s does not have requested_interval set'%step.__name__)
 188                         raise StepNotReady
 189
 190         def load_run_times(self):
 191                 try:
 192                         jrun_times = open('/tmp/observer_run_times').read()
 193                         self.last_run_times = json.loads(jrun_times)
 194                 except:
 195                         self.last_run_times={}
 196                         for e in self.ordered_steps:
 197                                 self.last_run_times[e]=0
 198                 try:
 199                         jrun_times = open('/tmp/observer_deletion_run_times').read()
 200                         self.last_deletion_run_times = json.loads(jrun_times)
 201                 except:
 202                         self.last_deletion_run_times={}
 203                         for e in self.ordered_steps:
 204                                 self.last_deletion_run_times[e]=0
 205
 206
 207
 208         def save_run_times(self):
 209                 run_times = json.dumps(self.last_run_times)
 210                 open('/tmp/observer_run_times','w').write(run_times)
 211
 212                 deletion_run_times = json.dumps(self.last_deletion_run_times)
 213                 open('/tmp/observer_deletion_run_times','w').write(deletion_run_times)
 214
 215         def check_class_dependency(self, step, failed_steps):
 216                 step.dependenices = []
 217                 for obj in step.provides:
 218                         step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
 219                 for failed_step in failed_steps:
 220                         if (failed_step in step.dependencies):
 221                                 raise StepNotReady
 222
 223         def sync(self, ordered_steps, error_mapper, deletion):
 224                 # Set of whole steps that failed
 225                 failed_steps = []
 226
 227                 # Set of individual objects within steps that failed
 228                 failed_step_objects = set()
 229
 230                 for S in ordered_steps:
 231                         step = self.step_lookup[S]
 232                         start_time=time.time()
 233
 234                         sync_step = step(driver=self.driver,error_map=error_mapper)
 235                         sync_step.__name__ = step.__name__
 236                         sync_step.dependencies = []
 237                         try:
 238                                 mlist = sync_step.provides
 239
 240                                 for m in mlist:
 241                                         sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
 242                         except KeyError:
 243                                 pass
 244                         sync_step.debug_mode = debug_mode
 245
 246                         should_run = False
 247                         try:
 248                                 # Various checks that decide whether
 249                                 # this step runs or not
 250                                 self.check_class_dependency(sync_step, failed_steps) # dont run Slices if Sites failed
 251                                 self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour
 252                                 should_run = True
 253                         except StepNotReady:
 254                                 logging.info('Step not ready: %s'%sync_step.__name__)
 255                                 failed_steps.append(sync_step)
 256                         except Exception,e:
 257                                 logging.error('%r',e)
 258                                 logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion))
 259                                 failed_steps.append(sync_step)
 260
 261                         if (should_run):
 262                                 try:
 263                                         duration=time.time() - start_time
 264
 265                                         logger.info('Executing step %s' % sync_step.__name__)
 266
 267                                         # ********* This is the actual sync step
 268                                         #import pdb
 269                                         #pdb.set_trace()
 270                                         failed_objects = sync_step(failed=list(failed_step_objects), deletion=deletion)
 271
 272
 273                                         self.check_duration(sync_step, duration)
 274                                         if failed_objects:
 275                                                 failed_step_objects.update(failed_objects)
 276
 277                                         self.update_run_time(sync_step,deletion)
 278                                 except Exception,e:
 279                                         logging.error('Model step failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
 280                                         logger.log_exc(e)
 281                                         failed_steps.append(S)
 282         def run(self):
 283                 if not self.driver.enabled:
 284                         return
 285                 if (self.driver_kind=="openstack") and (not self.driver.has_openstack):
 286                         return
 287
 288                 while True:
 289                         try:
 290                                 error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt")
 291                                 error_mapper = ErrorMapper(error_map_file)
 292
 293                                 logger.info('Waiting for event')
 294                                 tBeforeWait = time.time()
 295                                 self.wait_for_event(timeout=30)
 296                                 logger.info('Observer woke up')
 297
 298                                 # Two passes. One for sync, the other for deletion.
 299                                 for deletion in [False,True]:
 300                                         threads = []
 301                                         logger.info('Deletion=%r...'%deletion)
 302                                         ordered_steps = self.ordered_steps if not deletion else reversed(self.ordered_steps)
 303                                         thread = threading.Thread(target=self.sync, args=(ordered_steps,error_mapper,deletion))
 304                                         logger.info('Deletion=%r...'%deletion)
 305                                         threads.append(thread)
 306
 307                                         # Start threads
 308                                         for t in threads:
 309                                                 t.start()
 310
 311                                         # Wait for all threads to finish before continuing with the run loop
 312                                         for t in threads:
 313                                                 t.join()
 314
 315                                 self.save_run_times()
 316                         except Exception, e:
 317                                 logging.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e)
 318                                 logger.log_exc("Exception in observer run loop")
 319                                 traceback.print_exc()