[Bugfix] Bind api server port before starting engine (vllm-project#8491)

dtrifiro · Sep 16, 2024 · 47f5e03 · 47f5e03
1 parent 2759a43
commit 47f5e03
Showing 1 changed file with 6 additions and 0 deletions.
diff --git a/vllm/entrypoints/openai/api_server.py b/vllm/entrypoints/openai/api_server.py
@@ -5,6 +5,7 @@
 import os
 import re
 import signal
+import socket
 import tempfile
 from argparse import Namespace
 from contextlib import asynccontextmanager
@@ -525,6 +526,9 @@ async def run_server(args, **uvicorn_kwargs) -> None:
     logger.info("vLLM API server version %s", VLLM_VERSION)
     logger.info("args: %s", args)
 
+    temp_socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+    temp_socket.bind(("", args.port))
+
     def signal_handler(*_) -> None:
         # Interrupt server on sigterm while initializing
         raise KeyboardInterrupt("terminated")
@@ -541,6 +545,8 @@ def signal_handler(*_) -> None:
         model_config = await async_engine_client.get_model_config()
         init_app_state(async_engine_client, model_config, app.state, args)
 
+        temp_socket.close()
+
         shutdown_task = await serve_http(
             app,
             limit_concurrency=async_engine_client.limit_concurrency,