from collections import defaultdict
from core.models import *
from django.db.models import F, Q
-from openstack.manager import OpenStackManager
+#from openstack.manager import OpenStackManager
+from openstack.driver import OpenStackDriver
from util.logger import Logger, logging, logger
#from timeout import timeout
+from planetstack.config import Config
+from observer.steps import *
+debug_mode = False
logger = Logger(logfile='observer.log', level=logging.INFO)
-def toposort(g, steps):
- reverse = {}
-
- for k,v in g.items():
- for rk in v:
- try:
- reverse[rk].append(k)
- except:
- reverse[rk]=k
-
- sources = []
- for k,v in g.items():
- if not reverse.has_key(k):
- sources.append(k)
-
-
- for k,v in reverse.iteritems():
- if (not v):
- sources.append(k)
-
- order = []
- marked = []
- while sources:
- n = sources.pop()
- try:
- for m in g[n]:
- if m not in marked:
- sources.append(m)
- marked.append(m)
- except KeyError:
- pass
- if (n in steps):
- order.append(n)
-
- return order
+class StepNotReady(Exception):
+ pass
+
+def toposort(g, steps=None):
+ if (not steps):
+ keys = set(g.keys())
+ values = set({})
+ for v in g.values():
+ values=values | set(v)
+
+ steps=list(keys|values)
+ reverse = {}
+
+ for k,v in g.items():
+ for rk in v:
+ try:
+ reverse[rk].append(k)
+ except:
+ reverse[rk]=k
+
+ sources = []
+ for k,v in g.items():
+ if not reverse.has_key(k):
+ sources.append(k)
+
+
+ for k,v in reverse.iteritems():
+ if (not v):
+ sources.append(k)
+
+ order = []
+ marked = []
+
+ while sources:
+ n = sources.pop()
+ try:
+ for m in g[n]:
+ if m not in marked:
+ sources.append(m)
+ marked.append(m)
+ except KeyError:
+ pass
+ if (n in steps):
+ order.append(n)
+
+ order.reverse()
+ order.extend(set(steps)-set(order))
+ return order
class PlanetStackObserver:
- sync_steps = ['SyncNetworks','SyncNetworkSlivers','SyncSites','SyncSitePrivileges','SyncSlices','SyncSliceMemberships','SyncSlivers','SyncSliverIps']
+ sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector]
def __init__(self):
- self.manager = OpenStackManager()
# The Condition object that gets signalled by Feefie events
- self.load_sync_steps()
+ self.step_lookup = {}
+ self.load_sync_steps()
self.event_cond = threading.Condition()
- self.load_enacted()
+ self.driver = OpenStackDriver()
def wait_for_event(self, timeout):
self.event_cond.acquire()
self.event_cond.notify()
self.event_cond.release()
- def load_sync_steps(self):
- dep_path = Config().pl_dependency_path
- try:
- # This contains dependencies between records, not sync steps
- self.model_dependency_graph = json.loads(open(dep_path).read())
- except Exception,e:
- raise e
-
- backend_path = Config().backend_dependency_path
- try:
- # This contains dependencies between backend records
- self.backend_dependency_graph = json.loads(open(backend_path).read())
- except Exception,e:
- raise e
-
- provides_dict = {}
- for s in sync_steps:
- for m in s.provides:
- provides_dict[m]=s.__name__
-
- step_graph = {}
- for k,v in model_dependency_graph.iteritems():
- try:
- source = provides_dict[k]
- for m in v:
- try:
- dest = provides_dict[m]
- except KeyError:
- # no deps, pass
- step_graph[source]=dest
-
- except KeyError:
- pass
- # no dependencies, pass
-
- if (backend_dependency_graph):
- backend_dict = {}
- for s in sync_steps:
- for m in s.serves:
- backend_dict[m]=s.__name__
-
- for k,v in backend_dependency_graph.iteritems():
- try:
- source = backend_dict[k]
- for m in v:
- try:
- dest = backend_dict[m]
- except KeyError:
- # no deps, pass
- step_graph[source]=dest
-
- except KeyError:
- pass
- # no dependencies, pass
-
- dependency_graph = step_graph
-
- self.ordered_steps = toposort(dependency_graph, steps)
-
+ def load_sync_steps(self):
+ dep_path = Config().observer_dependency_graph
+ try:
+ # This contains dependencies between records, not sync steps
+ self.model_dependency_graph = json.loads(open(dep_path).read())
+ except Exception,e:
+ raise e
+
+ try:
+ backend_path = Config().observer_pl_dependency_graph
+ # This contains dependencies between backend records
+ self.backend_dependency_graph = json.loads(open(backend_path).read())
+ except Exception,e:
+ # We can work without a backend graph
+ self.backend_dependency_graph = {}
+
+ provides_dict = {}
+ for s in self.sync_steps:
+ self.step_lookup[s.__name__] = s
+ for m in s.provides:
+ try:
+ provides_dict[m.__name__].append(s.__name__)
+ except KeyError:
+ provides_dict[m.__name__]=[s.__name__]
+
+
+ step_graph = {}
+ for k,v in self.model_dependency_graph.iteritems():
+ try:
+ for source in provides_dict[k]:
+ for m in v:
+ try:
+ for dest in provides_dict[m]:
+ # no deps, pass
+ try:
+ step_graph[source].append(dest)
+ except:
+ step_graph[source]=[dest]
+ except KeyError:
+ pass
+
+ except KeyError:
+ pass
+ # no dependencies, pass
+
+ #import pdb
+ #pdb.set_trace()
+ if (self.backend_dependency_graph):
+ backend_dict = {}
+ for s in self.sync_steps:
+ for m in s.serves:
+ backend_dict[m]=s.__name__
+
+ for k,v in backend_dependency_graph.iteritems():
+ try:
+ source = backend_dict[k]
+ for m in v:
+ try:
+ dest = backend_dict[m]
+ except KeyError:
+ # no deps, pass
+ pass
+ step_graph[source]=dest
+
+ except KeyError:
+ pass
+ # no dependencies, pass
+
+ dependency_graph = step_graph
+
+ self.ordered_steps = toposort(dependency_graph, map(lambda s:s.__name__,self.sync_steps))
+ print "Order of steps=",self.ordered_steps
+ self.load_run_times()
+
+
+ def check_duration(self, step, duration):
+ try:
+ if (duration > step.deadline):
+ logger.info('Sync step %s missed deadline, took %.2f seconds'%(step.name,duration))
+ except AttributeError:
+ # S doesn't have a deadline
+ pass
+
+ def update_run_time(self, step):
+ self.last_run_times[step.__name__]=time.time()
+
+ def check_schedule(self, step):
+ time_since_last_run = time.time() - self.last_run_times.get(step.__name__, 0)
+ try:
+ if (time_since_last_run < step.requested_interval):
+ raise StepNotReady
+ except AttributeError:
+ logger.info('Step %s does not have requested_interval set'%step.__name__)
+ raise StepNotReady
+
+ def load_run_times(self):
+ try:
+ jrun_times = open('/tmp/observer_run_times').read()
+ self.last_run_times = json.loads(jrun_times)
+ except:
+ self.last_run_times={}
+ for e in self.ordered_steps:
+ self.last_run_times[e]=0
+
+
+ def save_run_times(self):
+ run_times = json.dumps(self.last_run_times)
+ open('/tmp/observer_run_times','w').write(run_times)
+
+ def check_class_dependency(self, step, failed_steps):
+ step.dependenices = []
+ for obj in step.provides:
+ step.dependenices.extend(self.model_dependency_graph.get(obj.__name__, []))
+ for failed_step in failed_steps:
+ if (failed_step in step.dependencies):
+ raise StepNotReady
def run(self):
- if not self.manager.enabled or not self.manager.has_openstack:
+ if not self.driver.enabled or not self.driver.has_openstack:
return
-
-
while True:
try:
- start_time=time.time()
-
logger.info('Waiting for event')
tBeforeWait = time.time()
- self.wait_for_event(timeout=300)
+ self.wait_for_event(timeout=30)
+ logger.info('Observer woke up')
- for S in self.ordered_steps:
- sync_step = S()
- sync_step()
+ # Set of whole steps that failed
+ failed_steps = []
- # Enforce 5 minutes between wakeups
- tSleep = 300 - (time.time() - tBeforeWait)
- if tSleep > 0:
- logger.info('Sleeping for %d seconds' % tSleep)
- time.sleep(tSleep)
+ # Set of individual objects within steps that failed
+ failed_step_objects = []
- logger.info('Observer woke up')
+ for S in self.ordered_steps:
+ step = self.step_lookup[S]
+ start_time=time.time()
+
+ sync_step = step(driver=self.driver)
+ sync_step.__name__ = step.__name__
+ sync_step.dependencies = []
+ try:
+ mlist = sync_step.provides
+
+ for m in mlist:
+ sync_step.dependencies.extend(self.model_dependency_graph[m.__name__])
+ except KeyError:
+ pass
+ sync_step.debug_mode = debug_mode
+
+ should_run = False
+ try:
+ # Various checks that decide whether
+ # this step runs or not
+ self.check_class_dependency(sync_step, failed_steps) # dont run Slices if Sites failed
+ self.check_schedule(sync_step) # dont run sync_network_routes if time since last run < 1 hour
+ should_run = True
+ except StepNotReady:
+ logging.info('Step not ready: %s'%sync_step.__name__)
+ failed_steps.append(sync_step)
+ except:
+ failed_steps.append(sync_step)
+
+ if (should_run):
+ try:
+ duration=time.time() - start_time
+
+ # ********* This is the actual sync step
+ #import pdb
+ #pdb.set_trace()
+ failed_objects = sync_step(failed=failed_step_objects)
+
+
+ self.check_duration(sync_step, duration)
+ if failed_objects:
+ failed_step_objects.extend(failed_objects)
+ self.update_run_time(sync_step)
+ except:
+ raise
+ failed_steps.append(S)
+ self.save_run_times()
except:
logger.log_exc("Exception in observer run loop")
traceback.print_exc()