def _perform_sync(self):
"""Performs the sync with the database
"""
scheduler_mgr.sync_with_database()
job_type_mgr.sync_with_database()
workspace_mgr.sync_with_database()
mesos_master = scheduler_mgr.mesos_address
node_mgr.sync_with_database(mesos_master.hostname, mesos_master.port)
self._sync_running_job_executions()
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
job_type_mgr.sync_with_database()
scheduler_mgr.sync_with_database()
workspace_mgr.sync_with_database()
# Start up background threads
self._db_sync_thread = DatabaseSyncThread(self._driver)
db_sync_thread = threading.Thread(target=self._db_sync_thread.run)
db_sync_thread.daemon = True
db_sync_thread.start()
self._recon_thread = ReconciliationThread()
recon_thread = threading.Thread(target=self._recon_thread.run)
recon_thread.daemon = True
recon_thread.start()
self._scheduling_thread = SchedulingThread(self._driver, self._framework_id)
scheduling_thread = threading.Thread(target=self._scheduling_thread.run)
scheduling_thread.daemon = True
scheduling_thread.start()
self._task_thread = TaskUpdateThread()
task_thread = threading.Thread(target=self._task_thread.run)
task_thread.daemon = True
task_thread.start()
self._reconcile_running_jobs()
def _perform_sync(self):
"""Performs the sync with the database
"""
scheduler_mgr.sync_with_database()
job_type_mgr.sync_with_database()
workspace_mgr.sync_with_database()
mesos_master = scheduler_mgr.mesos_address
node_mgr.sync_with_database(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()
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()
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()
