fix(runner): Eliminate the exception during shutdown

Signed-off-by: Frost Ming <[email protected]>

src/bentoml/_internal/marshal/dispatcher.py

@@ -1,25 +1,26 @@
 from __future__ import annotations
 
-import time
-import typing as t
 import asyncio
-import logging
+import collections
 import functools
+import logging
+import time
 import traceback
-import collections
+import typing as t
 
 import attr
 import numpy as np
 
 from ..utils import cached_property
 from ..utils.alg import TokenBucket
 
+
 logger = logging.getLogger(__name__)
 
 
 if t.TYPE_CHECKING:
-    from ..runner.utils import Params
     from ..runner.container import Payload
+    from ..runner.utils import Params
 
 
 class NonBlockSema:
@@ -235,7 +236,7 @@ async def train_optimizer(
                     info.enqueue_time = now
                 self._loop.create_task(self.outbound_call(inputs_info))
         except asyncio.CancelledError:
-            return
+            raise
         except Exception as e:  # pylint: disable=broad-except
             logger.error(traceback.format_exc(), exc_info=e)
 

tests/e2e/bento_server_http/tests/test_serve.py

@@ -20,7 +20,7 @@ def test_http_server(bentoml_home: str):
 
     server.stop()
 
-    timeout = 60
+    timeout = 10
     start_time = time.time()
     while time.time() - start_time < timeout:
         retcode = server.process.poll()
@@ -48,7 +48,7 @@ def test_http_server_ctx(bentoml_home: str):
 
         assert res == {"more_test": "and more json"}
 
-    timeout = 60
+    timeout = 10
     start_time = time.time()
     while time.time() - start_time < timeout:
         retcode = server.process.poll()