chore(translate): 降频 2秒/次 + 改 spark 为 wss WebSocket 鉴权(智谱/zhipu=第一)

backend/app/config.py

@@ -70,19 +70,28 @@ class Settings(BaseSettings):
     tencent_tmt_quota_buffer: float = 0.05
     tencent_tmt_max_chars_per_req: int = 4500
 
-    # ===== 星火 Spark(优先翻译;Lite/免费)=====
+    # ===== 智谱 GLM(OpenAI 兼容,翻译主通道)=====
-    # 留空 = 不启用星火(直接走腾讯 TMT)
+    # 用法:智谱开放平台 GLM-4 系列,通过 OpenAI 协议调用
-    spark_api_password: str = ""
+    # 留空 api_key = 不启用该 provider
-    spark_base_url: str = "https://spark-api-open.xf-yun.com/v1"
-    spark_model: str = "lite"
-    spark_interval_sec: float = 1.0
-
-    # ===== 智谱 GLM(第二序位翻译;GLM-4-Flash 免费)=====
-    # 留空 = 不启用智谱(spark 不可用时直接降级到 tencent)
     zhipu_api_key: str = ""
     zhipu_base_url: str = "https://open.bigmodel.cn/api/paas/v4"
-    zhipu_model: str = "glm-4-flash"
+    zhipu_chat_model: str = "glm-4-flash"
-    zhipu_interval_sec: float = 1.0
+    zhipu_model: str = "glm-4-flash"  # 兼容旧字段名
+    # 2 秒/次(用户要求 6/11,降低频率避免触发限流)
+    zhipu_interval_sec: float = 2.0
+
+    # ===== 讯飞星火(WebSocket,翻译二级通道)=====
+    # 用法:讯飞星火 v1.1 Spark Lite,WebSocket 鉴权需要 APPID + APIKey + APISecret
+    # 留空任意一个 = 不启用该 provider
+    # (历史实现用的是 OpenAI 兼容 + APIPassword,已切换为 WebSocket 鉴权,字段名换)
+    spark_appid: str = ""
+    spark_api_key: str = ""  # WebSocket 鉴权用的 APIKey
+    spark_api_secret: str = ""  # WebSocket 鉴权用的 APISecret
+    spark_domain: str = "lite"  # v1.1 Spark Lite
+    # 兼容旧字段名(留空,只在没填 WebSocket 字段时起提示作用)
+    spark_api_password: str = ""
+    # 2 秒/次(用户要求 6/11,降低频率避免触发限流)
+    spark_interval_sec: float = 2.0
 
     @field_validator("tencent_tmt_quota_buffer")
     @classmethod
@@ -102,8 +111,8 @@ class Settings(BaseSettings):
     tencent_maas_api_key: str = ""
     tencent_maas_base_url: str = "https://maas-api.hivoice.cn/v1"
     tencent_maas_model: str = "u2"
-    # 每篇调用间隔(秒),与 LLM 客户端解耦
+    # 2 秒/次(与智谱/星火统一节流)
-    tencent_maas_interval_sec: float = 1.0
+    tencent_maas_interval_sec: float = 2.0
 
     # ===== 抓取 =====
     fetch_global_qps: int = 4

backend/app/services/translation/spark.py

@@ -1,19 +1,27 @@
-"""星火 Spark 翻译(OpenAI 兼容协议)。
+"""讯飞星火 Spark Lite 翻译后端(WebSocket 协议,用户 6/11 要求改回 wss)。
 
-- 端点:https://spark-api-open.xf-yun.com/v1/chat/completions
+- 端点:wss://spark-api.xf-yun.com/v1.1/chat
-- 模型:lite(也支持 generalv3.5 等,但该项目用 Lite,免费额度)
+- 模型:Spark Lite(v1.1 通用轻量版,限时免费)
-- 鉴权:Bearer token(APIPassword 直接当 Bearer 用)
+- 鉴权:URL QueryString 带 authorization(HMAC-SHA256 签名)
-- 请求:POST /chat/completions,system prompt 告诉模型做翻译
+  - 需要 APPID + APIKey + APISecret
+  - 签名算法见 _build_auth_url()
 
-设计上独立于 LlmClient,专门走 spark_* 配置,避免和 LLM 智能增强共用 client 的节流。
+设计上独立于 LlmClient(不在通用 OpenAI 协议内),
+鉴权 URL 每次调用前重算(因为 date 是当前时间)。
 """
 from __future__ import annotations
 
 import asyncio
+import base64
+import hashlib
+import hmac
+import json
 import logging
 import random
+from datetime import datetime
+from urllib.parse import urlencode
 
-import httpx
+import websockets
 
 from app.config import settings
 from app.services.translation.base import BaseTranslator, TranslationResult
@@ -21,84 +29,108 @@ from app.services.translation.base import BaseTranslator, TranslationResult
 logger = logging.getLogger("news.translate.spark")
 
 
-# 经过实测,这套 prompt 在 Spark Lite 上输出稳定(不夹带 reasoning / 注释)。
+# 讯飞 v1.1 域
-# Lite 不支持 system 角色(早期 Lite 模型),但 v1 OpenAI 兼容的 Lite 接受 system;
+_SPARK_HOST = "spark-api.xf-yun.com"
-# 保留 system prompt 走 4.0Ultra/Max/Pro 时也通用。
+_SPARK_PATH = "/v1.1/chat"
-_SYSTEM_PROMPT = """你是一个即时翻译助手。对于用户输入的英文或日文文章,请直接输出对应的简体中文译文。严格遵守:
 
-不要进行任何分步思考、不要输出分析、不要添加注释或说明。
 
-只输出中文译文本身,不包含任何额外文字(如"翻译结果:"、"以下是中文:"等)。
+# 讯飞做翻译的 system / user prompt 包装
+_SYSTEM_PROMPT = (
+    "你是一个翻译助手。请将用户输入的英文或日文文本翻译成简体中文。"
+    "严格遵守:不要输出分析、不要输出注释、不要添加任何包裹文字,只输出译文本身。"
+)
 
-如果输入内容既非英文也非日文,仅回复:"仅支持英文或日文翻译为中文。" """
+
+def _build_auth_url(api_key: str, api_secret: str) -> str:
+    """构造带鉴权 query 的 WebSocket URL。
+
+    算法来自讯飞开放平台官方文档:
+    1) date = 当前 GMT 时间 (RFC 1123 格式)
+    2) signature_origin = "host: {host}\\ndate: {date}\\n GET {path} HTTP/1.1"
+    3) signature_sha = HMAC-SHA256(api_secret, signature_origin)
+    4) signature = base64(signature_sha)
+    5) authorization_origin = "api_key=\\"{api_key}\\", algorithm=\\"hmac-sha256\\", "
+       "headers=\\"host date request-line\\", signature=\\"{signature}\\""
+    6) authorization = base64(authorization_origin)
+    7) 最终 URL: wss://{host}{path}?host={host}&date={urlencoded_date}&authorization={urlencoded_authorization}
+    """
+    now = datetime.utcnow()
+    date = now.strftime("%a, %d %b %Y %H:%M:%S GMT")
+    signature_origin = f"host: {_SPARK_HOST}\ndate: {date}\n GET {_SPARK_PATH} HTTP/1.1"
+    signature_sha = hmac.new(
+        api_secret.encode("utf-8"),
+        signature_origin.encode("utf-8"),
+        digestmod=hashlib.sha256,
+    ).digest()
+    signature = base64.b64encode(signature_sha).decode("utf-8")
+    authorization_origin = (
+        f'api_key="{api_key}", algorithm="hmac-sha256", '
+        f'headers="host date request-line", signature="{signature}"'
+    )
+    authorization = base64.b64encode(authorization_origin.encode("utf-8")).decode("utf-8")
+    # 注意 date 里带冒号 / 逗号,必须 url-encode
+    return f"wss://{_SPARK_HOST}{_SPARK_PATH}?{urlencode({'host': _SPARK_HOST, 'date': date, 'authorization': authorization})}"
 
 
 class SparkTranslator(BaseTranslator):
-    """星火 Spark 翻译(OpenAI 兼容协议,模型 lite / generalv3.5 等)。"""
+    """讯飞星火 Spark Lite 翻译后端(WebSocket)。"""
 
     name = "spark"
 
     def __init__(self):
-        if not settings.spark_api_password:
+        if not settings.spark_appid or not settings.spark_api_key or not settings.spark_api_secret:
-            raise RuntimeError("Spark APIPassword missing")
+            raise RuntimeError("讯飞星火 APPID / APIKey / APISecret 未配置(需要 WSS 鉴权三件套)")
-        self.api_password = settings.spark_api_password
+        self.appid = settings.spark_appid
-        self.base_url = settings.spark_base_url.rstrip("/")
+        self.api_key = settings.spark_api_key
-        self.model = settings.spark_model
+        self.api_secret = settings.spark_api_secret
+        self.domain = settings.spark_domain
         self.interval_sec = settings.spark_interval_sec
 
     def is_configured(self) -> bool:
-        return bool(self.api_password)
+        return bool(self.appid and self.api_key and self.api_secret)
 
     async def translate(
         self, text: str, source: str = "auto", target: str = "zh"
     ) -> TranslationResult:
-        """翻译接口。source/target 当前固定 EN/JA → ZH(由 prompt 控制)。"""
         if not text.strip():
             return TranslationResult(text=text, engine=self.name, chars=0)
 
         if not self.is_configured():
-            raise RuntimeError("Spark APIPassword missing")
+            raise RuntimeError("讯飞星火未配置(APPID/APIKey/APISecret)")
 
-        system = _SYSTEM_PROMPT
+        # 长度截断:Spark Lite 单轮 8K context,留余量
-        user = text
+        max_input = 4000
+        truncated = text[:max_input] if len(text) > max_input else text
 
-        url = f"{self.base_url}/chat/completions"
+        # 构造请求体
         payload = {
-            "model": self.model,
+            "header": {"app_id": self.appid, "uid": "translator"},
-            "messages": [
+            "parameter": {
-                {"role": "system", "content": system},
+                "chat": {
-                {"role": "user", "content": user},
+                    "domain": self.domain,
-            ],
+                    "temperature": 0.1,  # 翻译低温度
-            "temperature": 0.0,
+                    "max_tokens": 4096,
-            "max_tokens": max(256, len(text) * 3),
+                }
-        }
+            },
-        headers = {
+            "payload": {
-            "Authorization": f"Bearer {self.api_password}",
+                "message": {
-            "Content-Type": "application/json",
+                    "text": [
+                        {"role": "system", "content": _SYSTEM_PROMPT},
+                        {"role": "user", "content": truncated},
+                    ]
+                }
+            },
         }
 
-        # 简单串行 + 重试 1 次
+        # 鉴权 URL(每次重算)
+        auth_url = _build_auth_url(self.api_key, self.api_secret)
+
+        # 简单重试 1 次
         last_exc: Exception | None = None
         for attempt in range(2):
             try:
-                async with httpx.AsyncClient(timeout=60.0) as client:
+                content = await self._send_once(auth_url, payload)
-                    r = await client.post(url, json=payload, headers=headers)
+                # 节流(避免被限流 — 2 秒/次)
-                if r.status_code >= 500:
-                    raise RuntimeError(f"Spark 5xx: {r.status_code} {r.text[:200]}")
-                if r.status_code != 200:
-                    raise RuntimeError(f"Spark {r.status_code}: {r.text[:300]}")
-                data = r.json()
-                # 错误响应:{"error": {...}}
-                if "error" in data:
-                    raise RuntimeError(f"Spark error: {data['error']}")
-                content = (
-                    data.get("choices", [{}])[0]
-                    .get("message", {})
-                    .get("content", "")
-                    .strip()
-                )
-                if not content:
-                    raise RuntimeError(f"Spark empty content: {r.text[:300]}")
                 await asyncio.sleep(self.interval_sec)
                 return TranslationResult(
                     text=content, engine=self.name, chars=len(text), cached=False
@@ -112,3 +144,39 @@ class SparkTranslator(BaseTranslator):
                     raise
         assert last_exc is not None
         raise last_exc
+
+    async def _send_once(self, auth_url: str, payload: dict) -> str:
+        """单次 WebSocket 调用,聚合流式响应,返回完整文本。"""
+        # websockets 13+ 异步上下文(关闭旧版 serve / connect 双 API 模糊)
+        async with websockets.connect(auth_url, ping_interval=None) as ws:
+            await ws.send(json.dumps(payload, ensure_ascii=False))
+
+            collected: list[str] = []
+            while True:
+                raw = await ws.recv()
+                data = json.loads(raw)
+                # 错误响应(header.code != 0)
+                header = data.get("header", {})
+                code = header.get("code", 0)
+                if code != 0:
+                    msg = header.get("message", "")
+                    raise RuntimeError(f"Spark 错误 {code}: {msg}")
+
+                # 流式增量
+                choices = data.get("payload", {}).get("choices", {})
+                status = choices.get("status", 0)
+                text_parts = choices.get("text", [])
+                if text_parts:
+                    for t in text_parts:
+                        content = t.get("content", "")
+                        if content:
+                            collected.append(content)
+
+                # status=2 表示结束
+                if status == 2:
+                    break
+
+            full = "".join(collected).strip()
+            if not full:
+                raise RuntimeError("Spark 返回空 content")
+            return full

backend/pyproject.toml

@@ -30,6 +30,7 @@ dependencies = [
     "python-dateutil>=2.9.0",
     # translation
     "tencentcloud-sdk-python>=3.0.1200",
+    "websockets>=13.0",  # 讯飞星火 WebSocket 鉴权 / 调用
     # scheduling
     "apscheduler>=3.10.4",
     # observability