diff --git a/src/bentoml/serve.py b/src/bentoml/serve.py index 7f03952074a..75b94d5bd3d 100644 --- a/src/bentoml/serve.py +++ b/src/bentoml/serve.py @@ -667,6 +667,7 @@ def serve_grpc_production( reflection: bool = Provide[BentoMLContainer.grpc.reflection.enabled], ) -> None: from circus.sockets import CircusSocket + from bentoml import load from bentoml.exceptions import UnprocessableEntity diff --git a/src/bentoml/start.py b/src/bentoml/start.py index 2b4c555f809..ee00785dd0e 100644 --- a/src/bentoml/start.py +++ b/src/bentoml/start.py @@ -35,10 +35,11 @@ def start_runner_server( """ Experimental API for serving a BentoML runner. """ - from bentoml import load from circus.sockets import CircusSocket from circus.watcher import Watcher + from bentoml import load + from .serve import ensure_prometheus_dir from ._internal.utils import reserve_free_port from ._internal.utils.circus import create_standalone_arbiter