def stop_container(self, container_info: ContainerInfo) -> None:
for container in self.client.containers.list():
if container.name == get_name_from_alias(container_info.name):
container.kill()
delete_alias(container_info.name)
return
raise ContainerNotFound(additional_message="Job Not Killed ",
container_name=container_info.name)
def tensorboard_refresh(self, container_info: ContainerInfo) -> None:
if get_name_from_alias(container_info.name) in map(
lambda container: container.name,
self.client.containers.list()):
self.training_api_tensorboard_service.refresh_training_api_tensorboard(
container_info)
else:
self.tensorboard_service.tensorboard_stop(container_info)
self.tensorboard_service.tensorboard_start(container_info)
def get_container_logs(self, container_info: ContainerInfo) -> List[str]:
logs: str = ""
container_alias: str = get_name_from_alias(alias=container_info.name)
try:
for container in self.client.containers.list():
if container.name == container_alias:
logs = container.logs()
log_list: List[str] = logs.decode('utf-8').splitlines()
return log_list
except Exception:
raise ContainerNotFound(container_name=container_info.name)
def get_tensorboard_port(self,
container_info: ContainerInfo) -> TensorboardPort:
tensorboard_port: int = None
api_port: int = None
try:
for container in self.client.containers.list():
if (container.name == get_name_from_alias(
container_info.name)):
tensorboard_port = int(
container.ports['6006/tcp'][0]['HostPort'])
api_port = int(container.ports['5252/tcp'][0]['HostPort'])
return TensorboardPort(tensorboard_port=tensorboard_port,
api_port=api_port)
except Exception:
raise ContainerNotFound(container_name=container_info.name)
def refresh_training_api_tensorboard(
self, container_info: ContainerInfo) -> None:
api_port: int = None
try:
for container in self.client.containers.list():
if container.name == get_name_from_alias(container_info.name):
api_port = int(container.ports['5252/tcp'][0]['HostPort'])
refresh_url = self.url.base_url + str(
api_port) + self.url.training_ip_tensorboard_refresh
response = requests.get(refresh_url)
except requests.exceptions.RequestException as e:
print(e)
raise ApiRequestException(additional_message=e.__str__(),
container_name=container_info.name)
except Exception as e:
raise ContainerNotFound(additional_message=e.__str__(),
container_name=container_info.name)