def _perform_scheduling(self):
"""Performs task reconciliation with the Mesos master
:returns: The number of Mesos tasks that were scheduled
:rtype: int
"""
when = now()
# Get updated node and job type models from managers
nodes = node_mgr.get_nodes()
cleanup_mgr.update_nodes(nodes)
offer_mgr.update_nodes(nodes)
offer_mgr.ready_new_offers()
self._job_types = job_type_mgr.get_job_types()
# Look at job type limits and determine number available to be scheduled
self._job_type_limit_available = {}
for job_type in self._job_types.values():
if job_type.max_scheduled:
self._job_type_limit_available[job_type.id] = job_type.max_scheduled
for running_job_exe in job_exe_mgr.get_running_job_exes():
if running_job_exe.job_type_id in self._job_type_limit_available:
self._job_type_limit_available[running_job_exe.job_type_id] -= 1
self._consider_node_tasks(when)
self._consider_running_job_exes()
self._consider_new_job_exes()
return self._schedule_accepted_tasks()
def _perform_scheduling(self):
"""Performs task reconciliation with the Mesos master
:returns: The number of Mesos tasks that were scheduled
:rtype: int
"""
# Get updated node and job type models from managers
nodes = node_mgr.get_nodes()
cleanup_mgr.update_nodes(nodes)
offer_mgr.update_nodes(nodes)
offer_mgr.ready_new_offers()
self._job_types = job_type_mgr.get_job_types()
# Look at job type limits and determine number available to be scheduled
self._job_type_limit_available = {}
for job_type in self._job_types.values():
if job_type.max_scheduled:
self._job_type_limit_available[job_type.id] = job_type.max_scheduled
for running_job_exe in running_job_mgr.get_all_job_exes():
if running_job_exe.job_type_id in self._job_type_limit_available:
self._job_type_limit_available[running_job_exe.job_type_id] -= 1
self._send_tasks_for_reconciliation()
self._consider_cleanup_tasks()
self._consider_running_job_exes()
self._consider_new_job_exes()
return self._schedule_accepted_tasks()
def setUp(self):
django.setup()
reset_error_cache()
self.framework_id = '1234'
Scheduler.objects.initialize_scheduler()
Scheduler.objects.update(
num_message_handlers=0
) # Prevent message handler tasks from scheduling
self._client = MagicMock()
scheduler_mgr.sync_with_database()
scheduler_mgr.update_from_mesos(framework_id=self.framework_id)
resource_mgr.clear()
job_exe_mgr.clear()
self.agent_1 = Agent('agent_1', 'host_1')
self.agent_2 = Agent('agent_2', 'host_2')
self.agent_3 = Agent('agent_3', 'host_2')
node_mgr.clear()
node_mgr.register_agents([self.agent_1, self.agent_2])
node_mgr.sync_with_database(scheduler_mgr.config)
# Ignore initial cleanup, health check, and image pull tasks
for node in node_mgr.get_nodes():
node._last_health_task = now()
node._initial_cleanup_completed()
node._is_image_pulled = True
node._update_state()
if node.agent_id == 'agent_1':
self.node_1_id = node.id
cleanup_mgr.update_nodes(node_mgr.get_nodes())
self.node_1 = Node.objects.get(id=self.node_1_id)
# Ignore system tasks
system_task_mgr._is_db_update_completed = True
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.queue_large = queue_test_utils.create_queue(
resources=NodeResources([Cpus(
125.0), Mem(12048.0), Disk(12048.0)]))
job_type_mgr.sync_with_database()
def _execute(self):
"""See :meth:`scheduler.threads.base_thread.BaseSchedulerThread._execute`
"""
scheduler_mgr.sync_with_database()
job_type_mgr.sync_with_database()
job_exe_mgr.sync_with_database()
workspace_mgr.sync_with_database()
node_mgr.sync_with_database(scheduler_mgr.config)
cleanup_mgr.update_nodes(node_mgr.get_nodes())
mesos_master = scheduler_mgr.mesos_address
resource_mgr.sync_with_mesos(mesos_master.hostname, mesos_master.port)
# Handle canceled job executions
for finished_job_exe in job_exe_mgr.sync_with_database():
cleanup_mgr.add_job_execution(finished_job_exe)
if settings.SECRETS_URL:
secrets_mgr.sync_with_backend()
def _execute(self):
"""See :meth:`scheduler.threads.base_thread.BaseSchedulerThread._execute`
"""
scheduler_mgr.sync_with_database()
job_type_mgr.sync_with_database()
workspace_mgr.sync_with_database()
node_mgr.sync_with_database(scheduler_mgr.config)
cleanup_mgr.update_nodes(node_mgr.get_nodes())
mesos_master = scheduler_mgr.mesos_address
resource_mgr.sync_with_mesos(mesos_master.hostname, mesos_master.port)
# Kill running tasks for canceled job executions
for task_to_kill in job_exe_mgr.sync_with_database():
pb_task_to_kill = mesos_pb2.TaskID()
pb_task_to_kill.value = task_to_kill.id
logger.info('Killing task %s', task_to_kill.id)
self._driver.killTask(pb_task_to_kill)
if settings.SECRETS_URL:
secrets_mgr.sync_with_backend()
