Fix server launch for jupyter notebook (#186)

python/sglang/srt/server.py

@@ -464,7 +464,6 @@ def launch_server(server_args, pipe_finish_writer):
     assert proc_router.is_alive() and proc_detoken.is_alive()
 
     def _launch_server():
-        # Launch api server
         uvicorn.run(
             app,
             host=server_args.host,
@@ -474,9 +473,7 @@ def launch_server(server_args, pipe_finish_writer):
             loop="uvloop",
         )
 
-    t = threading.Thread(target=_launch_server)
-    t.start()
-
+    def _wait_and_warmup():
         url = server_args.url()
         for _ in range(60):
             time.sleep(1)
@@ -518,6 +515,13 @@ def launch_server(server_args, pipe_finish_writer):
         if pipe_finish_writer is not None:
             pipe_finish_writer.send("init ok")
 
+    t = threading.Thread(target=_wait_and_warmup)
+    t.start()
+    try:
+        _launch_server()
+    finally:
+        t.join()
+
 
 class Runtime:
     def __init__(