small cleanups

Signed-off-by: Edward Oakes <[email protected]>

python/ray/serve/_private/constants.py

@@ -327,8 +327,8 @@
     os.environ.get("RAY_SERVE_GRPC_MAX_MESSAGE_SIZE", (2 * 1024 * 1024 * 1024) - 1)
 )
 
-# Serve's gRPC server options.
-SERVE_GRPC_OPTIONS = [
+# Default options passed when constructing gRPC servers.
+DEFAULT_GRPC_SERVER_OPTIONS = [
     ("grpc.max_send_message_length", RAY_SERVE_GRPC_MAX_MESSAGE_SIZE),
     ("grpc.max_receive_message_length", RAY_SERVE_GRPC_MAX_MESSAGE_SIZE),
 ]

python/ray/serve/_private/grpc_util.py

@@ -1,9 +1,9 @@
-from typing import Sequence
+from typing import Callable, List, Optional, Sequence, Tuple
 
 import grpc
 from grpc.aio._server import Server
 
-from ray.serve._private.constants import SERVE_GRPC_OPTIONS
+from ray.serve._private.constants import DEFAULT_GRPC_SERVER_OPTIONS
 
 
 class gRPCServer(Server):
@@ -51,7 +51,11 @@ def add_generic_rpc_handlers(
         super().add_generic_rpc_handlers(generic_rpc_handlers)
 
 
-def create_serve_grpc_server(service_handler_factory):
+def create_serve_grpc_server(
+    service_handler_factory: Callable,
+    *,
+    extra_options: Optional[List[Tuple[str, str]]] = None
+):
     """Custom function to create Serve's gRPC server.
 
     This function works similar to `grpc.server()`, but it creates a Serve defined
@@ -63,7 +67,7 @@ def create_serve_grpc_server(service_handler_factory):
         thread_pool=None,
         generic_handlers=(),
         interceptors=(),
-        options=SERVE_GRPC_OPTIONS,
+        options=DEFAULT_GRPC_SERVER_OPTIONS + (extra_options or []),
         maximum_concurrent_rpcs=None,
         compression=None,
         service_handler_factory=service_handler_factory,