fix(rate_limit): differentiate between tenant plan and API Key QPS limit errors

- Add logic to detect tenant plan QPS limits and return a specific error message when triggered.
- Simplify boolean check in model activation quota validation.

api/app/core/api_key_auth.py

@@ -108,6 +108,8 @@ def require_api_key(
                 # 根据错误消息判断限流类型
                 if "Daily" in error_msg:
                     code = BizCode.API_KEY_DAILY_LIMIT_EXCEEDED
+                elif "Tenant" in error_msg:
+                    code = BizCode.API_KEY_QPS_LIMIT_EXCEEDED  # 租户套餐速率超限，同属 QPS 类
                 elif "QPS" in error_msg:
                     code = BizCode.API_KEY_QPS_LIMIT_EXCEEDED
                 else:

api/app/core/quota_manager.py

@@ -488,7 +488,7 @@ def check_model_activation_quota(func: Callable) -> Callable:
             logger.warning("模型激活配额检查失败：缺少 model_id 或 model_data 参数")
             return await func(*args, **kwargs)
 
-        if model_data.is_active is True:
+        if model_data.is_active:
             try:
                 from app.services.model_service import ModelConfigService
 
@@ -522,7 +522,7 @@ def check_model_activation_quota(func: Callable) -> Callable:
             logger.warning("模型激活配额检查失败：缺少 model_id 或 model_data 参数")
             return func(*args, **kwargs)
 
-        if model_data.is_active is True:
+        if model_data.is_active:
             try:
                 from app.services.model_service import ModelConfigService
 

api/app/services/api_key_service.py

@@ -386,7 +386,12 @@ class RateLimiterService:
         # 用最终有效限额做 QPS 检查
         qps_ok, qps_info = await self.check_qps(api_key.id, effective_limit)
         if not qps_ok:
-            return False, "QPS limit exceeded", {
+            # 判断是套餐限额触发还是 api_key 自身限额触发
+            if tenant_limit and effective_limit == tenant_limit and api_key.rate_limit > tenant_limit:
+                error_msg = "Tenant QPS limit exceeded"
+            else:
+                error_msg = "QPS limit exceeded"
+            return False, error_msg, {
                 "X-RateLimit-Limit-QPS": str(qps_info["limit"]),
                 "X-RateLimit-Remaining-QPS": str(qps_info["remaining"]),
                 "X-RateLimit-Reset": str(qps_info["reset"])