feat: LLM/Embedding 重试次数纳入 system.yaml 统一配置

- system.yaml: llm.max_retries(死配置) 改为 llm.max_attempts；embedding 段新增 max_attempts
- llm/client.py: LLMConfig 新增 max_attempts（默认 3 兜底），load_llm_config 读取配置
- llm/extractor.py: translate_and_extract(_async) max_attempts=None 时取 config.max_attempts
- embedding/client.py: load_embedding_config 读取 embedding.max_attempts
- scheduler/reporter.py: _call_llm_simple max_retries=None 时取 max_attempts-1
- 新增 2 个配置读取测试；已同步 pi5 验证

configs/system.yaml

@@ -43,10 +43,10 @@ dedup:
 # ── LLM 翻译+事件抽取 ────────────────────────────────
 llm:
   provider: "deepseek"
-  deepseek_model: "deepseek-chat"
+  deepseek_model: "deepseek-v4-flash"
   qwen_model: "qwen-plus"
   timeout_sec: 60
-  max_retries: 3
+  max_attempts: 3          # 单篇总尝试次数（含首次），失败后指数退避重试；日报 AI 摘要同用此值
   max_tokens: 8192
   temperature: 0.1
   concurrency: 3
@@ -57,6 +57,7 @@ embedding:
   dashscope_model: "text-embedding-v3"
   dimension: 1024
   batch_size: 10
+  max_attempts: 3          # 单批总尝试次数（含首次），失败后指数退避重试
   timeout_sec: 30
 
 # ── Qdrant ──────────────────────────────────────────

embedding/client.py

@@ -92,6 +92,7 @@ def load_embedding_config(
     dimension = int(sys_cfg.get("dimension", DASHSCOPE_DEFAULT_DIM))
     batch_size = min(int(sys_cfg.get("batch_size", DASHSCOPE_BATCH_LIMIT)), DASHSCOPE_BATCH_LIMIT)
     timeout = float(sys_cfg.get("timeout_sec", 30.0))
+    max_attempts = int(sys_cfg.get("max_attempts", DEFAULT_MAX_ATTEMPTS))
 
     if not api_key:
         raise EmbeddingError("DASHSCOPE_API_KEY 未配置，请检查 .env")
@@ -104,6 +105,7 @@ def load_embedding_config(
         dimension=dimension,
         batch_size=batch_size,
         timeout_sec=timeout,
+        max_attempts=max_attempts,
     )
 
 

llm/client.py

@@ -25,6 +25,9 @@ _QWEN_DEFAULT_BASE = "https://dashscope.aliyuncs.com/compatible-mode/v1"
 _DEEPSEEK_DEFAULT_MODEL = "deepseek-chat"
 _QWEN_DEFAULT_MODEL = "qwen-plus"
 
+# 默认单次调用总尝试次数（含首次；system.yaml llm.max_attempts 未配置时兜底）
+_DEFAULT_MAX_ATTEMPTS = 3
+
 
 def _load_system_config() -> dict:
     """加载 configs/system.yaml 中 llm 段配置。"""
@@ -50,6 +53,7 @@ class LLMConfig:
     timeout_sec: float = 60.0
     temperature: float = 0.1
     max_tokens: int = 8192
+    max_attempts: int = _DEFAULT_MAX_ATTEMPTS  # 单次调用总尝试次数（含首次）
 
     def __post_init__(self) -> None:
         if not self.api_key:
@@ -93,6 +97,7 @@ def load_llm_config(
     timeout = float(config.get("timeout_sec", 60.0))
     temperature = float(config.get("temperature", 0.1))
     max_tokens = int(config.get("max_tokens", 8192))
+    max_attempts = int(config.get("max_attempts", _DEFAULT_MAX_ATTEMPTS))
 
     return LLMConfig(
         provider=p,
@@ -102,6 +107,7 @@ def load_llm_config(
         timeout_sec=timeout,
         temperature=temperature,
         max_tokens=max_tokens,
+        max_attempts=max_attempts,
     )
 
 

llm/extractor.py

@@ -226,7 +226,7 @@ def translate_and_extract(
     article: ProcessedArticle,
     *,
     template: PromptTemplate | None = None,
-    max_attempts: int = DEFAULT_MAX_ATTEMPTS,
+    max_attempts: int | None = None,
 ) -> EnTranslatedArticle:
     """同步翻译 + 事件抽取（单篇文章，带重试）。
 
@@ -235,7 +235,8 @@ def translate_and_extract(
         config: LLM 配置
         article: 待处理的英文新闻
         template: Prompt 模板，默认加载 prompts/translation_and_extraction.md
-        max_attempts: 最大重试次数
+        max_attempts: 最大尝试次数（含首次）；None 时取 config.max_attempts
+            （来自 system.yaml llm.max_attempts）
 
     Returns:
         EnTranslatedArticle 含双语内容 + 事件
@@ -243,6 +244,8 @@ def translate_and_extract(
     Raises:
         LLMCallError: 所有重试均失败
     """
+    if max_attempts is None:
+        max_attempts = config.max_attempts
     tpl = template or PromptTemplate()
     system_prompt, user_prompt = tpl.render(article)
 
@@ -305,10 +308,12 @@ async def translate_and_extract_async(
     article: ProcessedArticle,
     *,
     template: PromptTemplate | None = None,
-    max_attempts: int = DEFAULT_MAX_ATTEMPTS,
+    max_attempts: int | None = None,
     semaphore: asyncio.Semaphore | None = None,
 ) -> EnTranslatedArticle:
     """异步翻译 + 事件抽取（批处理用），与同步版逻辑等价。"""
+    if max_attempts is None:
+        max_attempts = config.max_attempts
     tpl = template or PromptTemplate()
     system_prompt, user_prompt = tpl.render(article)
 

scheduler/reporter.py

@@ -283,7 +283,7 @@ def _call_llm_simple(
     system_prompt: str,
     user_prompt: str,
     max_tokens: int = 600,
-    max_retries: int = 2,
+    max_retries: int | None = None,
 ) -> str:
     """封装 LLM 调用，带重试和客户端复用。
 
@@ -291,7 +291,8 @@ def _call_llm_simple(
         system_prompt: system role 内容
         user_prompt: user role 内容
         max_tokens: 最大输出 token
-        max_retries: 最大重试次数（不含首次调用）
+        max_retries: 重试次数（不含首次）；None 时取
+            config.max_attempts - 1（system.yaml llm.max_attempts）
 
     Returns:
         LLM 输出文本；所有重试均失败返回空字符串
@@ -308,6 +309,9 @@ def _call_llm_simple(
     config = _llm_client_cache["config"]
     client = _llm_client_cache[cache_key]
 
+    if max_retries is None:
+        max_retries = max(config.max_attempts - 1, 0)
+
     last_err: str = ""
     for attempt in range(1, max_retries + 2):  # 首次 + max_retries 次重试
         try:

tests/test_embedding.py

@@ -87,6 +87,12 @@ class TestLoadEmbeddingConfig:
         assert cfg.provider == "dashscope"
         assert cfg.api_key == "sk-dashscope-test"
 
+    def test_max_attempts_from_system_yaml(self, monkeypatch):
+        """重试次数取自 system.yaml embedding.max_attempts（当前 3）。"""
+        monkeypatch.setenv("DASHSCOPE_API_KEY", "sk-dashscope-test")
+        cfg = load_embedding_config()
+        assert cfg.max_attempts == 3
+
     def test_fallback_to_qwen_key(self, monkeypatch):
         monkeypatch.delenv("DASHSCOPE_API_KEY", raising=False)
         monkeypatch.setenv("QWEN_API_KEY", "sk-qwen-key")

tests/test_llm.py

@@ -457,6 +457,12 @@ class TestLoadLLMConfig:
         with pytest.raises(ValueError, match="API key 未配置"):
             load_llm_config(provider="deepseek")
 
+    def test_max_attempts_from_system_yaml(self, monkeypatch):
+        """重试次数取自 system.yaml llm.max_attempts（当前 3）。"""
+        monkeypatch.setenv("DEEPSEEK_API_KEY", "sk-deepseek-test-key")
+        config = load_llm_config(provider="deepseek")
+        assert config.max_attempts == 3
+
 
 # --------------------------------------------------------------------------- #
 # translate_and_extract（mock LLM）