[Frontend] change default log request behavior

docs/models/supported_models.md

@@ -253,7 +253,7 @@ export https_proxy=http://your.proxy.server:port
 https_proxy=http://your.proxy.server:port huggingface-cli download <model_name>
 
 # or use vllm cmd directly
-https_proxy=http://your.proxy.server:port  vllm serve <model_name> --disable-log-requests
+https_proxy=http://your.proxy.server:port  vllm serve <model_name>
 ```
 
 - Set the proxy in Python interpreter:

tests/engine/test_arg_utils.py

@@ -290,6 +290,22 @@ def test_prefix_cache_default():
     assert not engine_args.enable_prefix_caching
 
 
+def test_log_request_default_and_legacy_flag(caplog_vllm):
+    parser = AsyncEngineArgs.add_cli_args(FlexibleArgumentParser())
+
+    args = parser.parse_args([])
+    caplog_vllm.clear()
+    engine_args = AsyncEngineArgs.from_cli_args(args=args)
+    assert engine_args.disable_log_requests
+    assert "--disable-log-requests" in caplog_vllm.text
+
+    caplog_vllm.clear()
+    args = parser.parse_args(["--enable-legacy-log-requests"])
+    engine_args = AsyncEngineArgs.from_cli_args(args=args)
+    assert not engine_args.disable_log_requests
+    assert caplog_vllm.text == ""
+
+
 # yapf: disable
 @pytest.mark.parametrize(("arg", "expected", "option"), [
     (None, None, "mm-processor-kwargs"),

vllm/engine/arg_utils.py

@@ -881,6 +881,17 @@ class EngineArgs:
         attrs = [attr.name for attr in dataclasses.fields(cls)]
         # Set the attributes from the parsed arguments.
         engine_args = cls(**{attr: getattr(args, attr) for attr in attrs})
+        if isinstance(engine_args, AsyncEngineArgs):
+            if getattr(args, 'disable_log_requests', False):
+                engine_args.disable_log_requests = True
+            elif getattr(args, 'enable_legacy_log_requests', False):
+                engine_args.disable_log_requests = False
+            else:
+                engine_args.disable_log_requests = True
+                logger.warning(
+                    "'--disable-log-requests' is now enabled by default. "
+                    "Use --enable-legacy-log-requests to restore the previous "
+                    "request logging behavior.")
         return engine_args
 
     def create_model_config(self) -> ModelConfig:
@@ -1701,7 +1712,8 @@ class EngineArgs:
 @dataclass
 class AsyncEngineArgs(EngineArgs):
     """Arguments for asynchronous vLLM engine."""
-    disable_log_requests: bool = False
+    disable_log_requests: bool = True
+    enable_legacy_log_requests: bool = False
 
     @staticmethod
     def add_cli_args(parser: FlexibleArgumentParser,
@@ -1715,6 +1727,9 @@ class AsyncEngineArgs(EngineArgs):
         parser.add_argument('--disable-log-requests',
                             action='store_true',
                             help='Disable logging requests.')
+        parser.add_argument('--enable-legacy-log-requests',
+                            action='store_true',
+                            help='Enable legacy request logging behavior.')
         current_platform.pre_register_and_update(parser)
         return parser
 

vllm/entrypoints/openai/api_server.py

@@ -1448,13 +1448,14 @@ async def init_app_state(
     vllm_config: VllmConfig,
     state: State,
     args: Namespace,
+    engine_args: AsyncEngineArgs,
 ) -> None:
     if args.served_model_name is not None:
         served_model_names = args.served_model_name
     else:
         served_model_names = [args.model]
 
-    if args.disable_log_requests:
+    if engine_args.disable_log_requests:
         request_logger = None
     else:
         request_logger = RequestLogger(max_log_len=args.max_log_len)
@@ -1704,12 +1705,16 @@ async def run_server_worker(listen_address,
     if log_config is not None:
         uvicorn_kwargs['log_config'] = log_config
 
-    async with build_async_engine_client(args, client_config) as engine_client:
+    engine_args = AsyncEngineArgs.from_cli_args(args)
+    async with build_async_engine_client_from_engine_args(
+            engine_args, args.disable_frontend_multiprocessing,
+            client_config) as engine_client:
         maybe_register_tokenizer_info_endpoint(args)
         app = build_app(args)
 
         vllm_config = await engine_client.get_vllm_config()
-        await init_app_state(engine_client, vllm_config, app.state, args)
+        await init_app_state(engine_client, vllm_config, app.state, args,
+                             engine_args)
 
         logger.info("Starting vLLM API server %d on %s", server_index,
                     listen_address)

vllm/entrypoints/openai/run_batch.py

@@ -326,7 +326,7 @@ async def main(args):
         for name in served_model_names
     ]
 
-    if args.disable_log_requests:
+    if engine_args.disable_log_requests:
         request_logger = None
     else:
         request_logger = RequestLogger(max_log_len=args.max_log_len)