feat：优化执行步骤

app/clients/llm.py

@@ -23,6 +23,7 @@ def _get_base_llm() -> ChatOpenAI:
         base_url=settings.llm_base_url,
         timeout=settings.llm_timeout,
         max_retries=settings.llm_max_retries,
+        reasoning_effort=settings.llm_reasoning_effort,
     )
 
 

app/clients/user_center.py

@@ -1,5 +1,6 @@
 """用户中心客户端：用请求携带的 token 换取当前用户，结果短时缓存"""
 
+import asyncio
 import hashlib
 import time
 from functools import cache
@@ -14,6 +15,9 @@ from app.schemas.user import CurrentUser
 # 用户中心 401 时提示调用方用 Bearer token 重新认证
 BEARER_CHALLENGE = {"WWW-Authenticate": "Bearer"}
 
+# 瞬时故障的重试间隔（秒），长度即重试次数
+RETRY_DELAYS = (0.2, 0.5)
+
 
 class UserCenterClient:
     def __init__(
@@ -55,23 +59,41 @@ class UserCenterClient:
             del self._cache[stale]
         self._cache[key] = (now + ttl, user)
 
+    async def _request_with_retry(self, token: str) -> httpx.Response:
+        """校验是幂等的 GET：连接失败、超时和 5xx 属于瞬时故障，短间隔重试"""
+        attempts = 1 + len(RETRY_DELAYS)
+        for attempt in range(attempts):
+            try:
+                response = await self._get_client().get(
+                    "/api/users/me", headers={"Authorization": f"Bearer {token}"}
+                )
+            except httpx.HTTPError as error:
+                logger.opt(exception=error).warning(
+                    "用户中心请求失败 attempt={}/{}", attempt + 1, attempts
+                )
+                if attempt == attempts - 1:
+                    raise BizException(503, "用户中心不可用") from error
+            else:
+                if response.status_code < 500:
+                    return response
+                logger.warning(
+                    "用户中心返回 HTTP {} attempt={}/{}",
+                    response.status_code,
+                    attempt + 1,
+                    attempts,
+                )
+                if attempt == attempts - 1:
+                    raise BizException(503, "用户中心不可用")
+            await asyncio.sleep(RETRY_DELAYS[attempt])
+        raise AssertionError("unreachable")
+
     async def fetch_current_user(self, token: str) -> CurrentUser:
         key = hashlib.sha256(token.encode()).hexdigest()
         user = self._cached(key)
         if user is not None:
             return user
 
-        try:
-            response = await self._get_client().get(
-                "/api/users/me", headers={"Authorization": f"Bearer {token}"}
-            )
-        except httpx.HTTPError as error:
-            logger.opt(exception=error).warning("用户中心请求失败")
-            raise BizException(503, "用户中心不可用") from error
-
-        if response.status_code >= 500:
-            logger.warning("用户中心返回 HTTP {}", response.status_code)
-            raise BizException(503, "用户中心不可用")
+        response = await self._request_with_retry(token)
         try:
             body = response.json()
         except ValueError: