class ExecuteBuilder:
def __init__(self, id: int, repeat_count: int = 1, exit=True):
self.session = Session.create_session(key='ExecuteBuilder')
self.id = id
self.repeat_count = repeat_count
self.logger = create_logger(self.session, 'ExecuteBuilder')
self.logger_db = create_logger(self.session,
'ExecuteBuilder.db',
console=False)
self.exit = exit
self.provider = None
self.library_provider = None
self.storage = None
self.task = None
self.dag = None
self.executor = None
self.hostname = None
self.docker_img = None
self.worker_index = None
self.queue_personal = None
self.config = None
self.executor_type = None
def info(self, msg: str, step=None):
self.logger.info(msg, ComponentType.Worker, self.hostname, self.id,
step)
def error(self, msg: str, step=None):
self.logger.error(msg, ComponentType.Worker, self.hostname, self.id,
step)
def warning(self, msg: str, step=None):
self.logger.warning(msg, ComponentType.Worker, self.hostname, self.id,
step)
def debug(self, msg: str, step=None):
self.logger.debug(msg, ComponentType.Worker, self.hostname, self.id,
step)
def create_base(self):
self.info('create_base')
if app.current_task:
app.current_task.update_state(state=states.SUCCESS)
app.control.revoke(app.current_task.request.id, terminate=True)
self.provider = TaskProvider(self.session)
self.library_provider = DagLibraryProvider(self.session)
self.storage = Storage(self.session)
self.task = self.provider.by_id(
self.id, joinedload(Task.dag_rel, innerjoin=True))
if not self.task:
raise Exception(f'task with id = {self.id} is not found')
self.dag = self.task.dag_rel
self.executor = None
self.hostname = socket.gethostname()
self.docker_img = DOCKER_IMG
self.worker_index = os.getenv('WORKER_INDEX', -1)
self.queue_personal = f'{self.hostname}_{self.docker_img}_' \
f'{self.worker_index}'
self.config = Config.from_yaml(self.dag.config)
set_global_seed(self.config['info'].get('seed', 0))
self.executor_type = self.config['executors'][
self.task.executor]['type']
executor = self.config['executors'][self.task.executor]
cuda_visible_devices = os.getenv('CUDA_VISIBLE_DEVICES', '')
self.info(f'Env.before execution '
f'CUDA_VISIBLE_DEVICES={cuda_visible_devices}')
if cuda_visible_devices.strip() != '':
gpu_assigned = self.task.gpu_assigned or ''
cuda_visible_devices = cuda_visible_devices.split(',')
cuda_visible_devices = ','.join([
cuda_visible_devices[int(g)] for g in gpu_assigned.split(',')
if g.strip() != ''
])
else:
cuda_visible_devices = self.task.gpu_assigned
cuda_visible_devices = cuda_visible_devices or ''
env = {
'MKL_NUM_THREADS': 1,
'OMP_NUM_THREADS': 1,
'CUDA_VISIBLE_DEVICES': cuda_visible_devices
}
env.update(executor.get('env', {}))
for k, v in env.items():
os.environ[k] = str(v)
self.info(f'Set env. {k} = {v}')
def check_status(self):
self.info('check_status')
assert self.dag is not None, 'You must fetch task with dag_rel'
if self.task.status >= TaskStatus.InProgress.value:
msg = f'Task = {self.task.id}. Status = {self.task.status}, ' \
f'before the execute_by_id invocation.'
if app.current_task:
msg += f' Request Id = {app.current_task.request.id}'
self.error(msg)
return True
def change_status(self):
self.info('change_status')
self.task.computer_assigned = self.hostname
self.task.pid = os.getpid()
self.task.worker_index = self.worker_index
self.task.docker_assigned = self.docker_img
self.provider.change_status(self.task, TaskStatus.InProgress)
def download(self):
self.info('download')
if not self.task.debug:
folder = self.storage.download(task=self.id)
else:
folder = os.getcwd()
os.chdir(folder)
libraries = self.library_provider.dag(self.task.dag)
executor_type = self.executor_type
self.info('download. folder changed')
mlcomp_executors_folder = join(dirname(abspath(__file__)), 'executors')
mlcomp_base_folder = os.path.abspath(
join(mlcomp_executors_folder, '../../../'))
imported, was_installation = self.storage.import_executor(
mlcomp_executors_folder, mlcomp_base_folder, executor_type)
if not imported:
imported, was_installation = self.storage.import_executor(
folder, folder, executor_type, libraries)
if not imported:
raise Exception(f'Executor = {executor_type} not found')
self.info('download. executor imported')
if was_installation and not self.task.debug:
if self.repeat_count > 0:
self.info('was installation. '
'set task status to Queued. '
'And resending the task to a queue')
self.task.status = TaskStatus.Queued.value
self.provider.commit()
try:
execute.apply_async((self.id, self.repeat_count - 1),
queue=self.queue_personal,
retry=False)
except Exception:
pass
finally:
sys.exit()
assert Executor.is_registered(executor_type), \
f'Executor {executor_type} was not found'
def create_executor(self):
self.info('create_executor')
additional_info = yaml_load(self.task.additional_info) \
if self.task.additional_info else dict()
self.executor = Executor.from_config(executor=self.task.executor,
config=self.config,
additional_info=additional_info,
session=self.session,
logger=self.logger,
logger_db=self.logger_db)
def execute(self):
self.info('execute start')
res = self.executor(task=self.task,
task_provider=self.provider,
dag=self.dag)
self.info('execute executor finished')
res = res or {}
self.task.result = yaml_dump(res)
self.provider.commit()
if 'stage' in res and 'stages' in res:
index = res['stages'].index(res['stage'])
if index < len(res['stages']) - 1:
self.executor.info(f'stage = {res["stage"]} done. '
f'Go to the stage = '
f'{res["stages"][index + 1]}')
time.sleep(3)
self.executor.info(f'sending {(self.id, self.repeat_count)} '
f'to {self.queue_personal}')
self.task.status = TaskStatus.Queued.value
self.provider.commit()
execute.apply_async((self.id, self.repeat_count),
queue=self.queue_personal,
retry=False)
return
self.executor.step.finish()
self.provider.change_status(self.task, TaskStatus.Success)
self.info('execute end')
def build(self):
try:
self.create_base()
bad_status = self.check_status()
if bad_status:
return
self.change_status()
self.download()
self.create_executor()
self.execute()
except Exception as e:
step = self.executor.step.id if \
(self.executor and self.executor.step) else None
if Session.sqlalchemy_error(e):
Session.cleanup(key='ExecuteBuilder')
self.session = Session.create_session(key='ExecuteBuilder')
self.logger.session = create_logger(self.session,
'ExecuteBuilder')
self.error(traceback.format_exc(), step)
if self.task.status <= TaskStatus.InProgress.value:
self.provider.change_status(self.task, TaskStatus.Failed)
raise e
finally:
if app.current_task:
app.close()
if self.exit:
# noinspection PyProtectedMember
os._exit(0)
class ExecuteBuilder:
def __init__(self, id: int, repeat_count: int = 1, exit=True):
self.session = Session.create_session(key='ExecuteBuilder')
self.id = id
self.repeat_count = repeat_count
self.logger = create_logger(self.session, 'ExecuteBuilder')
self.exit = exit
self.provider = None
self.library_provider = None
self.storage = None
self.task = None
self.dag = None
self.executor = None
self.hostname = None
self.docker_img = None
self.worker_index = None
self.queue_personal = None
self.config = None
self.executor_type = None
def info(self, msg: str, step=None):
self.logger.info(msg, ComponentType.Worker, self.hostname, self.id,
step)
def error(self, msg: str, step=None):
self.logger.error(msg, ComponentType.Worker, self.hostname, self.id,
step)
def warning(self, msg: str, step=None):
self.logger.warning(msg, ComponentType.Worker, self.hostname, self.id,
step)
def debug(self, msg: str, step=None):
self.logger.debug(msg, ComponentType.Worker, self.hostname, self.id,
step)
def create_base(self):
self.info('create_base')
self.provider = TaskProvider(self.session)
self.library_provider = DagLibraryProvider(self.session)
self.storage = Storage(self.session)
self.task = self.provider.by_id(
self.id, joinedload(Task.dag_rel, innerjoin=True))
if not self.task:
raise Exception(f'task with id = {self.id} is not found')
self.dag = self.task.dag_rel
self.executor = None
self.hostname = socket.gethostname()
self.docker_img = DOCKER_IMG
self.worker_index = os.getenv('WORKER_INDEX', -1)
self.queue_personal = f'{self.hostname}_{self.docker_img}_' \
f'{self.worker_index}'
self.config = Config.from_yaml(self.dag.config)
self.executor_type = self.config['executors'][
self.task.executor]['type']
def check_status(self):
self.info('check_status')
assert self.dag is not None, 'You must fetch task with dag_rel'
if self.task.status > TaskStatus.InProgress.value:
msg = f'Task = {self.task.id}. Status = {self.task.status}, ' \
f'before the execute_by_id invocation'
self.error(msg)
raise Exception(msg)
def change_status(self):
self.info('change_status')
self.task.computer_assigned = self.hostname
self.task.pid = os.getpid()
self.task.worker_index = self.worker_index
self.task.docker_assigned = self.docker_img
self.provider.change_status(self.task, TaskStatus.InProgress)
def download(self):
self.info('download')
if not self.task.debug:
folder = self.storage.download(task=self.id)
else:
folder = os.getcwd()
os.chdir(folder)
libraries = self.library_provider.dag(self.task.dag)
executor_type = self.executor_type
mlcomp_executors_folder = join(dirname(abspath(__file__)), 'executors')
mlcomp_base_folder = os.path.abspath(
join(mlcomp_executors_folder, '../../../'))
imported, was_installation = self.storage.import_executor(
mlcomp_executors_folder, mlcomp_base_folder, executor_type)
if not imported:
imported, was_installation = self.storage.import_executor(
folder, folder, executor_type, libraries)
if not imported:
raise Exception(f'Executor = {executor_type} not found')
if was_installation and not self.task.debug:
if self.repeat_count > 0:
try:
self.warning(traceback.format_exc())
execute.apply_async((self.id, self.repeat_count - 1),
queue=self.queue_personal)
except Exception:
pass
finally:
sys.exit()
assert Executor.is_registered(executor_type), \
f'Executor {executor_type} was not found'
def create_executor(self):
self.info('create_executor')
additional_info = yaml_load(self.task.additional_info) \
if self.task.additional_info else dict()
self.executor = Executor.from_config(executor=self.task.executor,
config=self.config,
additional_info=additional_info,
session=self.session,
logger=self.logger)
def execute(self):
self.info('execute start')
res = self.executor(task=self.task,
task_provider=self.provider,
dag=self.dag)
self.info('execute executor finished')
res = res or {}
self.task.result = yaml_dump(res)
self.provider.commit()
if 'stage' in res and 'stages' in res:
index = res['stages'].index(res['stage'])
if index < len(res['stages']) - 1:
self.executor.info(f'stage = {res["stage"]} done. '
f'Go to the stage = '
f'{res["stages"][index + 1]}')
time.sleep(3)
self.executor.info(f'sending {(self.id, self.repeat_count)} '
f'to {self.queue_personal}')
execute.apply_async((self.id, self.repeat_count),
queue=self.queue_personal)
return
self.executor.step.finish()
self.provider.change_status(self.task, TaskStatus.Success)
self.info('execute end')
def build(self):
try:
self.create_base()
self.check_status()
self.change_status()
self.download()
self.create_executor()
self.execute()
except Exception as e:
if Session.sqlalchemy_error(e):
Session.cleanup(key='ExecuteBuilder')
self.session = Session.create_session(key='ExecuteBuilder')
self.logger.session = create_logger(self.session,
'ExecuteBuilder')
step = self.executor.step.id if \
(self.executor and self.executor.step) else None
self.error(traceback.format_exc(), step)
self.provider.change_status(self.task, TaskStatus.Failed)
raise e
finally:
if app.current_task:
app.current_task.update_state(state=states.SUCCESS)
app.close()
if self.exit:
# noinspection PyProtectedMember
os._exit(0)