def process_task(self, task: Task):
auxiliary = self.auxiliary['process_tasks'][-1]
auxiliary['computers'] = []
config = yaml_load(task.dag_rel.config)
executor = config['executors'][task.executor]
computers = self._process_task_get_computers(executor, task, auxiliary)
if len(computers) == 0:
return
to_send = self._process_task_to_send(executor, task, computers)
auxiliary['to_send'] = to_send[:5]
additional_info = yaml_load(task.additional_info)
rank = 0
master_port = None
if len(to_send) > 0:
master_port = self.find_port(
to_send[0][0], to_send[0][1].split('_')[1]
)
computer_names = {c['name'] for c, _, __ in to_send}
if len(computer_names) == 1:
task.computer_assigned = list(computer_names)[0]
for computer, queue, gpu_assigned in to_send:
main_cmp = to_send[0][0]
# noinspection PyTypeChecker
ip = 'localhost' if computer['name'] == main_cmp['name'] \
else main_cmp['ip']
distr_info = {
'master_addr': ip,
'rank': rank,
'local_rank': gpu_assigned,
'master_port': master_port,
'world_size': len(to_send),
'master_computer': main_cmp['name']
}
service_task = self.create_service_task(
task,
distr_info=distr_info,
gpu_assigned=gpu_assigned,
resume=additional_info.get('resume')
)
self.process_to_celery(service_task, queue, computer)
rank += 1
main_cmp['ports'].add(master_port)
if len(to_send) > 0:
task.status = TaskStatus.Queued.value
self.sent_tasks += len(to_send)
def process_to_celery(self, task: Task, queue: str, computer: dict):
r = execute.apply_async((task.id, ), queue=queue)
task.status = TaskStatus.Queued.value
task.computer_assigned = computer['name']
task.celery_id = r.id
if task.gpu_assigned is not None:
for g in map(int, task.gpu_assigned.split(',')):
computer['gpu'][g] = task.id
computer['cpu'] -= task.cpu
computer['memory'] -= task.memory * 1024
self.provider.update()
def process_to_celery(self, task: Task, queue: str, computer: dict):
r = execute.apply_async((task.id, ), queue=queue, retry=False)
task.status = TaskStatus.Queued.value
task.computer_assigned = computer['name']
task.celery_id = r.id
if task.gpu_assigned is not None:
for g in map(int, task.gpu_assigned.split(',')):
computer['gpu'][g] = task.id
computer['cpu'] -= task.cpu
computer['memory'] -= task.memory * 1024
self.logger.info(
f'Sent task={task.id} to celery. Queue = {queue} '
f'Task status = {task.status} Celery_id = {r.id}',
ComponentType.Supervisor)
self.provider.update()