Function stream_request

src/api/vertex.py:595–787

流式请求入口。 body 格式: {"model": str, "request": {...gemini payload...}} Yields: Response 对象（错误时）或 str/bytes（SSE 格式数据）

(
    body: Dict[str, Any],
    native: bool = False,
)

# ==================== 核心请求函数 ====================

async def stream_request(
    body: Dict[str, Any],
    native: bool = False,
):
    """
    流式请求入口。

    body 格式: {"model": str, "request": {...gemini payload...}}

    Yields:
        Response 对象（错误时）或 str/bytes（SSE 格式数据）
    """
    if not WREQ_AVAILABLE:
        yield Response(
            content='{"error":{"code":503,"message":"wreq not installed, vertex channel unavailable","status":"UNAVAILABLE"}}',
            status_code=503,
            media_type="application/json",
        )
        return

    model = body.get("model", "")
    gemini_payload = body.get("request", {})

    max_retries = 3
    recaptcha_token: Optional[str] = None
    is_first_auth = True

    for attempt in range(max_retries + 1):
        log.debug(f"[VERTEX STREAM] attempt {attempt + 1}/{max_retries + 1}, model={model}")

        if not recaptcha_token:
            recaptcha_token = await fetch_recaptcha_token()
            is_first_auth = True

        if not recaptcha_token:
            if attempt >= max_retries:
                yield Response(
                    content=json.dumps({"error": {"code": 401, "message": "Could not fetch reCAPTCHA token", "status": "UNAUTHENTICATED"}}),
                    status_code=401,
                    media_type="application/json",
                )
                return
            await asyncio.sleep(1)
            continue

        payload = _build_request_payload(model, gemini_payload, recaptcha_token)
        headers = _batch_graphql_headers()

        content_yielded = False
        need_retry = False
        auth_retry = False
        quota_retry = False

        try:
            resp = await wreq.post(
                BATCH_GRAPHQL_URL,
                json=payload,
                headers=headers,