def setUp(self):
django.setup()
Scheduler.objects.initialize_scheduler()
self._driver = MagicMock()
self._job_exe_manager = RunningJobExecutionManager()
self._job_type_manager = JobTypeManager()
self._node_manager = NodeManager()
self._offer_manager = OfferManager()
self._scheduler_manager = SchedulerManager()
self._workspace_manager = WorkspaceManager()
self._scheduler_manager.sync_with_database()
self.node_agent_1 = 'agent_1'
self.node_agent_2 = 'agent_2'
self.node_1 = node_test_utils.create_node(hostname='host_1', slave_id=self.node_agent_1)
self.node_2 = node_test_utils.create_node(hostname='host_2', slave_id=self.node_agent_2)
self.slave_infos = [SlaveInfo('host_1', slave_id=self.node_agent_1),
SlaveInfo('host_2', slave_id=self.node_agent_2)]
self._node_manager.add_agent_ids([self.node_agent_1, self.node_agent_2])
with patch('scheduler.sync.node_manager.api.get_slaves') as mock_get_slaves:
mock_get_slaves.return_value = self.slave_infos
self._node_manager.sync_with_database('master_host', 5050)
self.queue_1 = queue_test_utils.create_queue(cpus_required=4.0, mem_required=1024.0, disk_in_required=100.0,
disk_out_required=200.0, disk_total_required=300.0)
self.queue_2 = queue_test_utils.create_queue(cpus_required=8.0, mem_required=512.0, disk_in_required=400.0,
disk_out_required=45.0, disk_total_required=445.0)
self._job_type_manager.sync_with_database()
self._scheduling_thread = SchedulingThread(self._driver, '123', self._job_exe_manager, self._job_type_manager,
self._node_manager, self._offer_manager, self._scheduler_manager,
self._workspace_manager)
def registered(self, driver, frameworkId, masterInfo):
"""
Invoked when the scheduler successfully registers with a Mesos master.
It is called with the frameworkId, a unique ID generated by the
master, and the masterInfo which is information about the master
itself.
See documentation for :meth:`mesos_api.mesos.Scheduler.registered`.
"""
self._driver = driver
self._framework_id = frameworkId.value
self._master_hostname = masterInfo.hostname
self._master_port = masterInfo.port
logger.info(
'Scale scheduler registered as framework %s with Mesos master at %s:%i',
self._framework_id, self._master_hostname, self._master_port)
initialize_system()
Scheduler.objects.update_master(self._master_hostname,
self._master_port)
# Initial database sync
self._job_type_manager.sync_with_database()
self._scheduler_manager.sync_with_database()
self._workspace_manager.sync_with_database()
# Start up background threads
self._db_sync_thread = DatabaseSyncThread(self._driver,
self._job_exe_manager,
self._job_type_manager,
self._node_manager,
self._scheduler_manager,
self._workspace_manager)
db_sync_thread = threading.Thread(target=self._db_sync_thread.run)
db_sync_thread.daemon = True
db_sync_thread.start()
self._recon_thread = ReconciliationThread(self._driver)
recon_thread = threading.Thread(target=self._recon_thread.run)
recon_thread.daemon = True
recon_thread.start()
self._scheduling_thread = SchedulingThread(
self._driver, self._job_exe_manager, self._job_type_manager,
self._node_manager, self._offer_manager, self._scheduler_manager,
self._workspace_manager)
scheduling_thread = threading.Thread(
target=self._scheduling_thread.run)
scheduling_thread.daemon = True
scheduling_thread.start()
self._reconcile_running_jobs()
def setUp(self):
django.setup()
Scheduler.objects.initialize_scheduler()
self._driver = MagicMock()
scheduler_mgr.sync_with_database()
offer_mgr.clear()
self.node_agent_1 = 'agent_1'
self.node_agent_2 = 'agent_2'
self.slave_infos = [
SlaveInfo('host_1', slave_id=self.node_agent_1),
SlaveInfo('host_2', slave_id=self.node_agent_2)
]
node_mgr.clear()
node_mgr.register_agent_ids([self.node_agent_1, self.node_agent_2])
with patch('scheduler.node.manager.api.get_slaves') as mock_get_slaves:
mock_get_slaves.return_value = self.slave_infos
node_mgr.sync_with_database('master_host', 5050)
# Ignore initial cleanup tasks and health check tasks
for node in node_mgr.get_nodes():
node._last_heath_task = now()
node._initial_cleanup_completed()
node._update_state()
self.queue_1 = queue_test_utils.create_queue(cpus_required=4.0,
mem_required=1024.0,
disk_in_required=100.0,
disk_out_required=200.0,
disk_total_required=300.0)
self.queue_2 = queue_test_utils.create_queue(cpus_required=8.0,
mem_required=512.0,
disk_in_required=400.0,
disk_out_required=45.0,
disk_total_required=445.0)
job_type_mgr.sync_with_database()
self._scheduling_thread = SchedulingThread(self._driver, '123')
def initialize(self):
"""Initializes the scheduler and gets it ready to connect to Mesos. This method should only ever be called once.
"""
initialize_system()
# Initial database sync
logger.info('Performing initial sync with Scale database')
logger.info('Retrieving errors...')
reset_error_cache()
logger.info('Retrieving job execution metrics...')
job_exe_mgr.init_with_database()
logger.info('Retrieving job types...')
job_type_mgr.sync_with_database()
logger.info('Retrieving workspaces...')
workspace_mgr.sync_with_database()
logger.info('Retrieving scheduler settings...')
scheduler_mgr.sync_with_database()
# Start up background threads
self._threads = []
logger.info('Starting up background threads')
self._messaging_thread = MessagingThread()
restart_msg = RestartScheduler()
restart_msg.when = now()
self._messaging_thread.add_initial_messages([restart_msg])
messaging_thread = threading.Thread(target=self._messaging_thread.run)
messaging_thread.daemon = True
messaging_thread.start()
self._threads.append(messaging_thread)
self._recon_thread = ReconciliationThread()
recon_thread = threading.Thread(target=self._recon_thread.run)
recon_thread.daemon = True
recon_thread.start()
self._threads.append(recon_thread)
self._scheduler_status_thread = SchedulerStatusThread()
scheduler_status_thread = threading.Thread(target=self._scheduler_status_thread.run)
scheduler_status_thread.daemon = True
scheduler_status_thread.start()
self._threads.append(scheduler_status_thread)
self._scheduling_thread = SchedulingThread(self._client)
scheduling_thread = threading.Thread(target=self._scheduling_thread.run)
scheduling_thread.daemon = True
scheduling_thread.start()
self._threads.append(scheduling_thread)
self._sync_thread = SyncThread(self._driver)
sync_thread = threading.Thread(target=self._sync_thread.run)
sync_thread.daemon = True
sync_thread.start()
self._threads.append(sync_thread)
self._task_handling_thread = TaskHandlingThread(self._driver)
task_handling_thread = threading.Thread(target=self._task_handling_thread.run)
task_handling_thread.daemon = True
task_handling_thread.start()
self._threads.append(task_handling_thread)
self._task_update_thread = TaskUpdateThread()
task_update_thread = threading.Thread(target=self._task_update_thread.run)
task_update_thread.daemon = True
task_update_thread.start()
self._threads.append(task_update_thread)
def registered(self, driver, frameworkId, masterInfo):
"""
Invoked when the scheduler successfully registers with a Mesos master.
It is called with the frameworkId, a unique ID generated by the
master, and the masterInfo which is information about the master
itself.
See documentation for :meth:`mesos_api.mesos.Scheduler.registered`.
"""
self._driver = driver
self._framework_id = frameworkId.value
self._master_hostname = masterInfo.hostname
self._master_port = masterInfo.port
logger.info(
'Scale scheduler registered as framework %s with Mesos master at %s:%i',
self._framework_id, self._master_hostname, self._master_port)
initialize_system()
Scheduler.objects.update_master(self._master_hostname,
self._master_port)
scheduler_mgr.update_from_mesos(
self._framework_id,
HostAddress(self._master_hostname, self._master_port))
recon_mgr.driver = self._driver
# Initial database sync
logger.info('Performing initial sync with Scale database')
reset_error_cache()
job_exe_mgr.init_with_database()
job_type_mgr.sync_with_database()
scheduler_mgr.sync_with_database()
workspace_mgr.sync_with_database()
# Start up background threads
self._messaging_thread = MessagingThread()
restart_msg = RestartScheduler()
restart_msg.when = now()
self._messaging_thread.add_initial_messages([restart_msg])
messaging_thread = threading.Thread(target=self._messaging_thread.run)
messaging_thread.daemon = True
messaging_thread.start()
self._recon_thread = ReconciliationThread()
recon_thread = threading.Thread(target=self._recon_thread.run)
recon_thread.daemon = True
recon_thread.start()
self._scheduler_status_thread = SchedulerStatusThread()
scheduler_status_thread = threading.Thread(
target=self._scheduler_status_thread.run)
scheduler_status_thread.daemon = True
scheduler_status_thread.start()
self._scheduling_thread = SchedulingThread(self._driver)
scheduling_thread = threading.Thread(
target=self._scheduling_thread.run)
scheduling_thread.daemon = True
scheduling_thread.start()
self._sync_thread = SyncThread(self._driver)
sync_thread = threading.Thread(target=self._sync_thread.run)
sync_thread.daemon = True
sync_thread.start()
self._task_handling_thread = TaskHandlingThread(self._driver)
task_handling_thread = threading.Thread(
target=self._task_handling_thread.run)
task_handling_thread.daemon = True
task_handling_thread.start()
self._task_update_thread = TaskUpdateThread()
task_update_thread = threading.Thread(
target=self._task_update_thread.run)
task_update_thread.daemon = True
task_update_thread.start()
self._reconcile_running_jobs()