Different versions of Uvicorn can affect whether request.is_disconnected() in FastAPI executes correctly

[QZJ-2003]

My program is as follows:

@app.post("/v1/chat/completions", response_class=StreamingResponse)
async def chat(request: ChatRequest, req: Request) -> StreamingResponse:
    """
    Streaming chat interface.
    :param session_id: Session ID
    :param request: Request body, containing messages, temperature, top_p
    :return: SSE-formatted streaming response
    """
    session_id = request.session_id
    session = session_db.get_session(session_id)
    if not session:
        session = session_db.create_session(session_id)

    chat_history: List[Dict] = session["history"]
    chat_history.append({"role": "user", "content": request.messages[-1]['content']})
    generator = llm_model.generate_stream(
        chat_history, 
        temperature=request.temperature, 
        top_p=request.top_p, 
        max_tokens=8192, 
        stop= None, 
    )

    async def event_generator():
        full_response = ""
        interrupt_flag = False
        async for chunk in generator:
            if await req.is_disconnected():  # must use uvicorn==0.29.0 instead of 0.34.0(latest)
                print("break")
                interrupt_flag = True
                break
            if chunk is None or not chunk.outputs:
                continue
            out = chunk.outputs[-1]
            formatted_chunk = {
                "id": chunk.request_id,
                "object": "chat.completion.chunk",
                "created": get_cst_time(),
                "model": "QwQ-32B",
                "choices": [
                    {
                        "index": 0,
                        "delta": {
                            # any stop caused by stop sequences is not a real stop
                            "stop": chunk.finished and out.finish_reason == "stop" and out.stop_reason is None,
                            "role": "assistant",
                            "content": out.text[len(full_response):],
                        },
                        "finish_reason": out.finish_reason,
                        "stop_reason": out.stop_reason,
                    }
                ],
                "finished": chunk.finished,
            }
            full_response = out.text
            yield f"data: {json.dumps(formatted_chunk)}\n\n"

        # update session history
        if not interrupt_flag:
            chat_history.append({"role": "assistant", "content": full_response})
            session_db.update_session(session_id, chat_history)
            yield "data: [DONE]\n\n"

    return StreamingResponse(
        event_generator(), 
        media_type="text/event-stream",
        headers={
            "Cache-Control": "no-cache",
            "Connection": "keep-alive"
        }
    )

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=8000)

When I use Uvicorn version 0.34.0, the program does not enter the req.is_disconnected() check and never reaches the break statement. However, when I downgrade Uvicorn to version 0.29.0, this issue no longer occurs.