def handle_worker_offline(event):
"""
Celery event handler for 'worker-offline' events.
The 'worker-offline' event is emitted when a worker gracefully shuts down. It is not
emitted when a worker is killed instantly.
The event is first parsed and logged. If this event is from the resource manager, there is
no further processing to be done. Otherwise, a worker is shutting down, and a
_delete_worker() task is dispatched so that the resource manager will remove the record,
and handle any work cleanup associated with a worker going offline. Logging at the info
and debug level is also done.
:param event: A celery event to handle.
:type event: dict
"""
event_info = _parse_and_log_event(event)
# if this is the resource_manager do nothing
if _is_resource_manager(event):
return
msg = _("Worker '%(worker_name)s' shutdown") % event_info
_logger.info(msg)
_delete_worker.apply_async(args=(event_info['worker_name'], ),
kwargs={'normal_shutdown': True},
queue=RESOURCE_MANAGER_QUEUE)
def check_workers(self):
"""
Look for missing workers, and dispatch a cleanup task if one goes missing.
To find a missing worker, filter the Workers model for entries older than
utcnow() - WORKER_TIMEOUT_SECONDS. The heartbeat times are stored in native UTC, so this is
a comparable datetime.
For each missing worker found, dispatch a _delete_worker task requesting that the resource
manager delete the Worker and cleanup any associated work.
This method logs and the debug and error levels.
"""
msg = _('Looking for workers missing for more than %s seconds'
) % self.WORKER_TIMEOUT_SECONDS
_logger.debug(msg)
oldest_heartbeat_time = datetime.utcnow() - timedelta(
seconds=self.WORKER_TIMEOUT_SECONDS)
worker_criteria = Criteria(
filters={'last_heartbeat': {
'$lt': oldest_heartbeat_time
}},
fields=('_id', 'last_heartbeat', 'num_reservations'))
worker_list = list(resources.filter_workers(worker_criteria))
for worker in worker_list:
msg = _(
"Workers '%s' has gone missing, removing from list of workers"
) % worker.name
_logger.error(msg)
_delete_worker.apply_async(args=(worker.name, ),
queue=RESOURCE_MANAGER_QUEUE)
def handle_worker_offline(event):
"""
Celery event handler for 'worker-offline' events.
The 'worker-offline' event is emitted when a worker gracefully shuts down. It is not
emitted when a worker is killed instantly.
The event is first parsed and logged. If this event is from the resource manager, there is
no further processing to be done. Otherwise, a worker is shutting down, and a
_delete_worker() task is dispatched so that the resource manager will remove the record,
and handle any work cleanup associated with a worker going offline. Logging at the info
and debug level is also done.
:param event: A celery event to handle.
:type event: dict
"""
event_info = _parse_and_log_event(event)
# if this is the resource_manager do nothing
if _is_resource_manager(event):
return
msg = _("Worker '%(worker_name)s' shutdown") % event_info
_logger.info(msg)
_delete_worker.apply_async(args=(event_info['worker_name'],),
kwargs={'normal_shutdown': True},
queue=RESOURCE_MANAGER_QUEUE)
def check_workers(self):
"""
Look for missing workers, and dispatch a cleanup task if one goes missing.
To find a missing worker, filter the Workers model for entries older than
utcnow() - WORKER_TIMEOUT_SECONDS. The heartbeat times are stored in native UTC, so this is
a comparable datetime.
For each missing worker found, dispatch a _delete_worker task requesting that the resource
manager delete the Worker and cleanup any associated work.
This method logs and the debug and error levels.
"""
msg = _('Looking for workers missing for more than %s seconds') % self.WORKER_TIMEOUT_SECONDS
_logger.debug(msg)
oldest_heartbeat_time = datetime.utcnow() - timedelta(seconds=self.WORKER_TIMEOUT_SECONDS)
worker_criteria = Criteria(filters={'last_heartbeat': {'$lt': oldest_heartbeat_time}},
fields=('_id', 'last_heartbeat', 'num_reservations'))
worker_list = list(resources.filter_workers(worker_criteria))
for worker in worker_list:
msg = _("Workers '%s' has gone missing, removing from list of workers") % worker.name
_logger.error(msg)
_delete_worker.apply_async(args=(worker.name,), queue=RESOURCE_MANAGER_QUEUE)
