refactor: Improve model request rate limit middleware execution

middleware/model-rate-limit.go

@@ -153,21 +153,23 @@ func memoryRateLimitHandler(duration int64, totalMaxCount, successMaxCount int)
 
 
 // ModelRequestRateLimit 模型请求限流中间件
 // ModelRequestRateLimit 模型请求限流中间件
 func ModelRequestRateLimit() func(c *gin.Context) {
 func ModelRequestRateLimit() func(c *gin.Context) {
-	// 如果未启用限流，直接放行
+	return func(c *gin.Context) {
+		// 在每个请求时检查是否启用限流
 		if !setting.ModelRequestRateLimitEnabled {
 		if !setting.ModelRequestRateLimitEnabled {
-		return defNext
+			c.Next()
+			return
 		}
 		}
 
 
 		// 计算限流参数
 		// 计算限流参数
 		duration := int64(setting.ModelRequestRateLimitDurationMinutes * 60)
 		duration := int64(setting.ModelRequestRateLimitDurationMinutes * 60)
 		totalMaxCount := setting.ModelRequestRateLimitCount
 		totalMaxCount := setting.ModelRequestRateLimitCount
-
 		successMaxCount := setting.ModelRequestRateLimitSuccessCount
 		successMaxCount := setting.ModelRequestRateLimitSuccessCount
 
 
-	// 根据存储类型选择限流处理器
+		// 根据存储类型选择并执行限流处理器
 		if common.RedisEnabled {
 		if common.RedisEnabled {
-		return redisRateLimitHandler(duration, totalMaxCount, successMaxCount)
+			redisRateLimitHandler(duration, totalMaxCount, successMaxCount)(c)
 		} else {
 		} else {
-		return memoryRateLimitHandler(duration, totalMaxCount, successMaxCount)
+			memoryRateLimitHandler(duration, totalMaxCount, successMaxCount)(c)
+		}
 	}
 	}
 }
 }