From d11b059943776a2ed1656d49abfe8b6fc0bbe949 Mon Sep 17 00:00:00 2001
From: piexian <64474352+piexian@users.noreply.github.com>
Date: Tue, 10 Feb 2026 18:36:19 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BF=AE=E5=A4=8D=20Token=20=E8=87=AA?=
 =?UTF-8?q?=E5=8A=A8=E5=88=B7=E6=96=B0=E5=A4=B1=E8=B4=A5=E5=AF=BC=E8=87=B4?=
 =?UTF-8?q?=E6=8C=81=E7=BB=AD=20504/429=20=E9=94=99=E8=AF=AF?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 修复 Scheduler 刷新使用错误的默认模型名 (grok-4-1-thinking-1129 → grok-3)，
  避免 basic token 查询配额时返回 403
- 调整 Scheduler 启动后立即执行首轮刷新，而非等待整个 interval (2-8h)
- 移除刷新失败分支中的 mark_synced() 调用，使失败 token 在下一轮
  循环中被重新尝试而非被跳过
- Token 池耗尽时自动触发 refresh_cooling_tokens()，尝试恢复
  冷却中的 token，而非直接返回 429
---
 app/services/grok/services/chat.py | 21 +++++++++++++++------
 app/services/token/manager.py      |  6 +-----
 app/services/token/scheduler.py    |  5 ++++-
 3 files changed, 20 insertions(+), 12 deletions(-)

diff --git a/app/services/grok/services/chat.py b/app/services/grok/services/chat.py
index dcd36cc..ea88209 100644
--- a/app/services/grok/services/chat.py
+++ b/app/services/grok/services/chat.py
@@ -394,12 +394,21 @@ class ChatService:
                 break
 
         if not token:
-            raise AppException(
-                message="No available tokens. Please try again later.",
-                error_type=ErrorType.RATE_LIMIT.value,
-                code="rate_limit_exceeded",
-                status_code=429,
-            )
+            logger.info("No available tokens, attempting to refresh cooling tokens...")
+            result = await token_mgr.refresh_cooling_tokens()
+            if result.get("recovered", 0) > 0:
+                for pool_name in ModelService.pool_candidates_for_model(model):
+                    token = token_mgr.get_token(pool_name)
+                    if token:
+                        break
+
+            if not token:
+                raise AppException(
+                    message="No available tokens. Please try again later.",
+                    error_type=ErrorType.RATE_LIMIT.value,
+                    code="rate_limit_exceeded",
+                    status_code=429,
+                )
 
         # 解析参数
         think = {"enabled": True, "disabled": False}.get(thinking)
diff --git a/app/services/token/manager.py b/app/services/token/manager.py
index b5e28f1..d72cbd4 100644
--- a/app/services/token/manager.py
+++ b/app/services/token/manager.py
@@ -646,7 +646,7 @@ class TokenManager:
                 # 重试逻辑：最多 2 次重试
                 for retry in range(3):  # 0, 1, 2
                     try:
-                        result = await usage_service.get(token_str)
+                        result = await usage_service.get(token_str, model_name="grok-3")
 
                         if result and "remainingTokens" in result:
                             new_quota = result["remainingTokens"]
@@ -666,7 +666,6 @@ class TokenManager:
                                 "expired": False,
                             }
 
-                        token_info.mark_synced()
                         return {"recovered": False, "expired": False}
 
                     except Exception as e:
@@ -688,16 +687,13 @@ class TokenManager:
                                     f"marking as expired"
                                 )
                                 token_info.status = TokenStatus.EXPIRED
-                                token_info.mark_synced()
                                 return {"recovered": False, "expired": True}
                         else:
                             logger.warning(
                                 f"Token {token_info.token[:10]}...: refresh failed ({e})"
                             )
-                            token_info.mark_synced()
                             return {"recovered": False, "expired": False}
 
-                token_info.mark_synced()
                 return {"recovered": False, "expired": False}
 
         # 批量处理
diff --git a/app/services/token/scheduler.py b/app/services/token/scheduler.py
index ad4da7a..1427713 100644
--- a/app/services/token/scheduler.py
+++ b/app/services/token/scheduler.py
@@ -23,7 +23,6 @@ class TokenRefreshScheduler:
 
         while self._running:
             try:
-                await asyncio.sleep(self.interval_seconds)
                 storage = get_storage()
                 lock_acquired = False
                 lock = None
@@ -44,6 +43,7 @@ class TokenRefreshScheduler:
 
                 if not lock_acquired:
                     logger.info("Scheduler: skipped (lock not acquired)")
+                    await asyncio.sleep(self.interval_seconds)
                     continue
 
                 try:
@@ -65,10 +65,13 @@ class TokenRefreshScheduler:
                         except Exception:
                             pass
 
+                await asyncio.sleep(self.interval_seconds)
+
             except asyncio.CancelledError:
                 break
             except Exception as e:
                 logger.error(f"Scheduler: refresh error - {e}")
+                await asyncio.sleep(self.interval_seconds)
 
     def start(self):
         """启动调度器"""