X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=planetstack%2Fopenstack_observer%2Fevent_loop.py;h=51be5d3d4ae8f37bf6ac5a3a042f8cfa594dab5c;hb=05af1ce7d9e949e0c25646090c4bf48edcb29536;hp=0e54bd3eed4980ebb7a2ada38db59c302feb13a5;hpb=e82f5e5d223182820127c6ddd26e0c568d5bded3;p=plstackapi.git diff --git a/planetstack/openstack_observer/event_loop.py b/planetstack/openstack_observer/event_loop.py index 0e54bd3..51be5d3 100644 --- a/planetstack/openstack_observer/event_loop.py +++ b/planetstack/openstack_observer/event_loop.py @@ -2,15 +2,20 @@ import os import imp import inspect import time +import sys import traceback import commands import threading import json +import pdb +import pprint + from datetime import datetime from collections import defaultdict from core.models import * from django.db.models import F, Q +from django.db import connection #from openstack.manager import OpenStackManager from openstack.driver import OpenStackDriver from util.logger import Logger, logging, logger @@ -19,10 +24,22 @@ from planetstack.config import Config from observer.steps import * from syncstep import SyncStep from toposort import toposort -from ec2_observer.error_mapper import * +from observer.error_mapper import * +from openstack_observer.openstacksyncstep import OpenStackSyncStep + debug_mode = False +class bcolors: + HEADER = '\033[95m' + OKBLUE = '\033[94m' + OKGREEN = '\033[92m' + WARNING = '\033[93m' + FAIL = '\033[91m' + ENDC = '\033[0m' + BOLD = '\033[1m' + UNDERLINE = '\033[4m' + logger = Logger(level=logging.INFO) class StepNotReady(Exception): @@ -31,11 +48,27 @@ class StepNotReady(Exception): class NoOpDriver: def __init__(self): self.enabled = True + self.dependency_graph = None + +STEP_STATUS_WORKING=1 +STEP_STATUS_OK=2 +STEP_STATUS_KO=3 + +def invert_graph(g): + ig = {} + for k,v in g.items(): + for v0 in v: + try: + ig[v0].append(k) + except: + ig=[k] + return ig class PlanetStackObserver: - #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivileges,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector] + #sync_steps = [SyncNetworks,SyncNetworkSlivers,SyncSites,SyncSitePrivilege,SyncSlices,SyncSliceMemberships,SyncSlivers,SyncSliverIps,SyncExternalRoutes,SyncUsers,SyncRoles,SyncNodes,SyncImages,GarbageCollector] sync_steps = [] + def __init__(self): # The Condition object that gets signalled by Feefie events self.step_lookup = {} @@ -78,7 +111,7 @@ class PlanetStackObserver: # provides field (this eliminates the abstract base classes # since they don't have a provides) - if inspect.isclass(c) and issubclass(c, SyncStep) and hasattr(c,"provides") and (c not in self.sync_steps): + if inspect.isclass(c) and (issubclass(c, SyncStep) or issubclass(c,OpenStackSyncStep)) and hasattr(c,"provides") and (c not in self.sync_steps): self.sync_steps.append(c) logger.info('loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps])) # print 'loaded sync steps: %s' % ",".join([x.__name__ for x in self.sync_steps]) @@ -89,6 +122,12 @@ class PlanetStackObserver: try: # This contains dependencies between records, not sync steps self.model_dependency_graph = json.loads(open(dep_path).read()) + for lst in self.model_dependency_graph.values(): + for k in lst: + try: + deps = self.model_dependency_graph[k] + except: + self.model_dependency_graph[k] = [] except Exception,e: raise e @@ -111,11 +150,13 @@ class PlanetStackObserver: 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]: + if (not v): + step_graph[source] = [] + for m in v: try: for dest in provides_dict[m]: @@ -155,10 +196,17 @@ class PlanetStackObserver: pass # no dependencies, pass - dependency_graph = step_graph + self.dependency_graph = step_graph + self.deletion_dependency_graph = invert_graph(step_graph) + + pp = pprint.PrettyPrinter(indent=4) + pp.pprint(step_graph) + self.ordered_steps = toposort(self.dependency_graph, map(lambda s:s.__name__,self.sync_steps)) + #self.ordered_steps = ['SyncRoles', 'SyncControllerSites', 'SyncControllerSitePrivileges','SyncImages', 'SyncControllerImages','SyncControllerUsers','SyncControllerUserSitePrivileges','SyncControllerSlices', 'SyncControllerSlicePrivileges', 'SyncControllerUserSlicePrivileges', 'SyncControllerNetworks','SyncSlivers'] + #self.ordered_steps = ['SyncControllerSites'] - 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() @@ -205,7 +253,6 @@ class PlanetStackObserver: self.last_deletion_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) @@ -221,87 +268,185 @@ class PlanetStackObserver: if (failed_step in step.dependencies): raise StepNotReady + def sync(self, S, deletion): + try: + step = self.step_lookup[S] + start_time=time.time() + + logger.info("Starting to work on step %s" % step.__name__) + + dependency_graph = self.dependency_graph if not deletion else self.deletion_dependency_graph + + # Wait for step dependencies to be met + try: + deps = self.dependency_graph[S] + has_deps = True + except KeyError: + has_deps = False + + go = True + + failed_dep = None + if (has_deps): + for d in deps: + if d==step.__name__: + logger.info(" step %s self-wait skipped" % step.__name__) + go = True + continue + + cond = self.step_conditions[d] + cond.acquire() + if (self.step_status[d] is STEP_STATUS_WORKING): + logger.info(" step %s wait on dep %s" % (step.__name__, d)) + cond.wait() + elif self.step_status[d] == STEP_STATUS_OK: + go = True + else: + go = False + failed_dep = d + cond.release() + if (not go): + break + else: + go = True + + if (not go): + print bcolors.FAIL + "Step %r skipped on %r" % (step,failed_dep) + bcolors.ENDC + # SMBAKER: sync_step was not defined here, so I changed + # this from 'sync_step' to 'step'. Verify. + self.failed_steps.append(step) + my_status = STEP_STATUS_KO + else: + sync_step = step(driver=self.driver,error_map=self.error_mapper) + 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, self.failed_steps) # dont run Slices if Sites failed + self.check_schedule(sync_step, deletion) # dont run sync_network_routes if time since last run < 1 hour + should_run = True + except StepNotReady: + logger.info('Step not ready: %s'%sync_step.__name__) + self.failed_steps.append(sync_step) + my_status = STEP_STATUS_KO + except Exception,e: + logger.error('%r' % e) + logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion)) + self.failed_steps.append(sync_step) + my_status = STEP_STATUS_KO + + if (should_run): + try: + duration=time.time() - start_time + + logger.info('Executing step %s' % sync_step.__name__) + + print bcolors.OKBLUE + "Executing step %s" % sync_step.__name__ + bcolors.ENDC + failed_objects = sync_step(failed=list(self.failed_step_objects), deletion=deletion) + + self.check_duration(sync_step, duration) + + if failed_objects: + self.failed_step_objects.update(failed_objects) + + logger.info("Step %r succeeded" % step) + print bcolors.OKGREEN + "Step %r succeeded" % step + bcolors.ENDC + my_status = STEP_STATUS_OK + self.update_run_time(sync_step,deletion) + except Exception,e: + print bcolors.FAIL + "Model step %r failed" % (step) + bcolors.ENDC + 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)) + logger.log_exc(e) + self.failed_steps.append(S) + my_status = STEP_STATUS_KO + else: + logger.info("Step %r succeeded due to non-run" % step) + my_status = STEP_STATUS_OK + + try: + my_cond = self.step_conditions[S] + my_cond.acquire() + self.step_status[S]=my_status + my_cond.notify_all() + my_cond.release() + except KeyError,e: + logger.info('Step %r is a leaf' % step) + pass + finally: + connection.close() + def run(self): if not self.driver.enabled: return + if (self.driver_kind=="openstack") and (not self.driver.has_openstack): return while True: try: error_map_file = getattr(Config(), "error_map_path", "/opt/planetstack/error_map.txt") - error_mapper = ErrorMapper(error_map_file) + self.error_mapper = ErrorMapper(error_map_file) + + # Set of whole steps that failed + self.failed_steps = [] + + # Set of individual objects within steps that failed + self.failed_step_objects = set() + + # Set up conditions and step status + # This is needed for steps to run in parallel + # while obeying dependencies. + + providers = set() + for v in self.dependency_graph.values(): + if (v): + providers.update(v) + + self.step_conditions = {} + self.step_status = {} + for p in list(providers): + self.step_conditions[p] = threading.Condition() + self.step_status[p] = STEP_STATUS_WORKING + logger.info('Waiting for event') tBeforeWait = time.time() - self.wait_for_event(timeout=5) + self.wait_for_event(timeout=30) logger.info('Observer woke up') # Two passes. One for sync, the other for deletion. - for deletion in (False,True): + for deletion in [False,True]: + threads = [] logger.info('Deletion=%r...'%deletion) - # Set of whole steps that failed - failed_steps = [] + schedule = self.ordered_steps if not deletion else reversed(self.ordered_steps) - # Set of individual objects within steps that failed - failed_step_objects = set() + for S in schedule: + thread = threading.Thread(target=self.sync, args=(S, deletion)) - ordered_steps = self.ordered_steps if not deletion else reversed(self.ordered_steps) + logger.info('Deletion=%r...'%deletion) + threads.append(thread) - for S in ordered_steps: - step = self.step_lookup[S] - start_time=time.time() - - sync_step = step(driver=self.driver,error_map=error_mapper) - 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 + # Start threads + for t in threads: + t.start() + + # Wait for all threads to finish before continuing with the run loop + for t in threads: + t.join() - 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, deletion) # 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 Exception,e: - logging.error('%r',e) - logger.log_exc("sync step failed: %r. Deletion: %r"%(sync_step,deletion)) - failed_steps.append(sync_step) - - if (should_run): - try: - duration=time.time() - start_time - - logger.info('Executing step %s' % sync_step.__name__) - - # ********* This is the actual sync step - #import pdb - #pdb.set_trace() - failed_objects = sync_step(failed=list(failed_step_objects), deletion=deletion) - - - self.check_duration(sync_step, duration) - if failed_objects: - failed_step_objects.update(failed_objects) - - self.update_run_time(sync_step,deletion) - except Exception,e: - logging.error('Model step failed. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e) - logger.log_exc(e) - failed_steps.append(S) self.save_run_times() except Exception, e: - logging.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!',e) + logger.error('Core error. This seems like a misconfiguration or bug: %r. This error will not be relayed to the user!' % e) logger.log_exc("Exception in observer run loop") traceback.print_exc()