在 .NET 中限制 HTTP 处理程序的速率

项目
03/22/2023

本文介绍如何创建一个客户端 HTTP 处理程序，该处理程序对发送的请求数进行速率限制。你将看到访问 "www.example.com" 资源的 HttpClient。资源由依赖它们的应用使用，当应用对单个资源发出过多请求时，它可能会导致资源争用。资源争用在过多应用使用资源时发生，并且该资源无法为请求它的所有应用提供服务。这可能导致用户体验不佳，在某些情况下，甚至可能导致拒绝服务 (DoS) 攻击。有关 DoS 的详细信息，请参阅 OWASP：拒绝服务。

什么是速率限制？

速率限制是限制可以访问的资源量的概念。例如，你可能知道应用访问的数据库每分钟可以安全地处理 1,000 个请求，但它可能处理不了更多。可以在应用中放置一个速率限制器，每分钟只允许 1,000 个请求，并在它们可以访问数据库之前拒绝更多请求。因此，限制数据库速率并允许应用处理安全数量的请求。这是分布式系统中的一种常见模式，其中你可能有多个应用实例正在运行，并且你希望确保它们不会同时尝试访问数据库。有多个不同的速率限制算法来控制请求流。

若要在 .NET 中使用速率限制，请引用 System.Threading.RateLimiting NuGet 包。

实现 `DelegatingHandler` 子类

若要控制请求流，请实现自定义 DelegatingHandler 子类。这是一种 HttpMessageHandler，允许在请求发送到服务器之前截获和处理请求。还可以在响应返回到调用方之前截获和处理响应。在此示例中，你将实现一个自定义 DelegatingHandler 子类，该子类限制可以发送到单个资源的请求数。请考虑以下自定义 ClientSideRateLimitedHandler 类：

internal sealed class ClientSideRateLimitedHandler(
    RateLimiter limiter)
    : DelegatingHandler(new HttpClientHandler()), IAsyncDisposable
{
    protected override async Task<HttpResponseMessage> SendAsync(
        HttpRequestMessage request, CancellationToken cancellationToken)
    {
        using RateLimitLease lease = await limiter.AcquireAsync(
            permitCount: 1, cancellationToken);

        if (lease.IsAcquired)
        {
            return await base.SendAsync(request, cancellationToken);
        }

        var response = new HttpResponseMessage(HttpStatusCode.TooManyRequests);
        if (lease.TryGetMetadata(
                MetadataName.RetryAfter, out TimeSpan retryAfter))
        {
            response.Headers.Add(
                "Retry-After",
                ((int)retryAfter.TotalSeconds).ToString(
                    NumberFormatInfo.InvariantInfo));
        }

        return response;
    }

    async ValueTask IAsyncDisposable.DisposeAsync()
    { 
        await limiter.DisposeAsync().ConfigureAwait(false);

        Dispose(disposing: false);
        GC.SuppressFinalize(this);
    }

    protected override void Dispose(bool disposing)
    {
        base.Dispose(disposing);

        if (disposing)
        {
            limiter.Dispose();
        }
    }
}

上述 C# 代码：

继承 DelegatingHandler 类型。
实现 IAsyncDisposable 接口。
定义从构造函数分配的 RateLimiter 字段。
替代 SendAsync 方法，以便在请求发送到服务器之前截获和处理请求。
重写 DisposeAsync() 方法以释放 RateLimiter 实例。

仔细观察 SendAsync 方法，你会发现它：

依靠 RateLimiter 实例从 AcquireAsync 获取 RateLimitLease。
当 lease.IsAcquired 属性为 true 时，请求会发送到服务器。
否则，将返回带有 429 状态代码的 HttpResponseMessage，如果 lease 包含 RetryAfter 值，则 Retry-After 标头将设置为该值。

模拟多个并发请求

要将此自定义 DelegatingHandler 子类投入测试，你将创建一个模拟许多并发请求的控制台应用。此 Program 类使用自定义 ClientSideRateLimitedHandler 创建一个 HttpClient：

var options = new TokenBucketRateLimiterOptions
{ 
    TokenLimit = 8, 
    QueueProcessingOrder = QueueProcessingOrder.OldestFirst,
    QueueLimit = 3, 
    ReplenishmentPeriod = TimeSpan.FromMilliseconds(1), 
    TokensPerPeriod = 2, 
    AutoReplenishment = true
};

// Create an HTTP client with the client-side rate limited handler.
using HttpClient client = new(
    handler: new ClientSideRateLimitedHandler(
        limiter: new TokenBucketRateLimiter(options)));

// Create 100 urls with a unique query string.
var oneHundredUrls = Enumerable.Range(0, 100).Select(
    i => $"https://example.com?iteration={i:0#}");

// Flood the HTTP client with requests.
var floodOneThroughFortyNineTask = Parallel.ForEachAsync(
    source: oneHundredUrls.Take(0..49), 
    body: (url, cancellationToken) => GetAsync(client, url, cancellationToken));

var floodFiftyThroughOneHundredTask = Parallel.ForEachAsync(
    source: oneHundredUrls.Take(^50..),
    body: (url, cancellationToken) => GetAsync(client, url, cancellationToken));

await Task.WhenAll(
    floodOneThroughFortyNineTask,
    floodFiftyThroughOneHundredTask);

static async ValueTask GetAsync(
    HttpClient client, string url, CancellationToken cancellationToken)
{
    using var response =
        await client.GetAsync(url, cancellationToken);

    Console.WriteLine(
        $"URL: {url}, HTTP status code: {response.StatusCode} ({(int)response.StatusCode})");
}

在前面的控制台应用中：

TokenBucketRateLimiterOptions 配置令牌限制为 8，队列处理顺序为 OldestFirst，队列限制为 3，补货周期为 1 毫秒，每个周期的令牌值为 2，自动补充值为 true。
使用 ClientSideRateLimitedHandler（配置有 TokenBucketRateLimiter）创建了一个 HttpClient。
为了模拟 100 个请求，Enumerable.Range 创建了 100 个 URL，每个 URL 都有一个唯一的查询字符串参数。
从 Parallel.ForEachAsync 方法分配了两个 Task 对象，将 URL 分成两组。
HttpClient 用于向每个 URL 发送 GET 请求，并将响应写入控制台。
Task.WhenAll 等待两个任务完成。

由于 HttpClient 配置了 ClientSideRateLimitedHandler，因此并非所有请求都会到达服务器资源。可以通过运行控制台应用来测试此断言。你将看到只有一小部分请求发送到服务器，其余请求被拒绝并返回 HTTP 状态代码 429。尝试更改用于创建 TokenBucketRateLimiter 的 options 对象，以查看发送到服务器的请求数如何变化。

请参考以下示例输出：

URL: https://example.com?iteration=06, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=60, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=55, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=59, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=57, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=11, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=63, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=13, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=62, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=65, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=64, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=67, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=14, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=68, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=16, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=69, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=70, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=71, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=17, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=18, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=72, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=73, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=74, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=19, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=75, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=76, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=79, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=77, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=21, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=78, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=81, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=22, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=80, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=20, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=82, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=83, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=23, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=84, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=24, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=85, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=86, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=25, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=87, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=26, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=88, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=89, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=27, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=90, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=28, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=91, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=94, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=29, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=93, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=96, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=92, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=95, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=31, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=30, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=97, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=98, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=99, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=32, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=33, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=34, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=35, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=36, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=37, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=38, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=39, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=40, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=41, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=42, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=43, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=44, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=45, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=46, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=47, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=48, HTTP status code: TooManyRequests (429)
URL: https://example.com?iteration=15, HTTP status code: OK (200)
URL: https://example.com?iteration=04, HTTP status code: OK (200)
URL: https://example.com?iteration=54, HTTP status code: OK (200)
URL: https://example.com?iteration=08, HTTP status code: OK (200)
URL: https://example.com?iteration=00, HTTP status code: OK (200)
URL: https://example.com?iteration=51, HTTP status code: OK (200)
URL: https://example.com?iteration=10, HTTP status code: OK (200)
URL: https://example.com?iteration=66, HTTP status code: OK (200)
URL: https://example.com?iteration=56, HTTP status code: OK (200)
URL: https://example.com?iteration=52, HTTP status code: OK (200)
URL: https://example.com?iteration=12, HTTP status code: OK (200)
URL: https://example.com?iteration=53, HTTP status code: OK (200)
URL: https://example.com?iteration=07, HTTP status code: OK (200)
URL: https://example.com?iteration=02, HTTP status code: OK (200)
URL: https://example.com?iteration=01, HTTP status code: OK (200)
URL: https://example.com?iteration=61, HTTP status code: OK (200)
URL: https://example.com?iteration=05, HTTP status code: OK (200)
URL: https://example.com?iteration=09, HTTP status code: OK (200)
URL: https://example.com?iteration=03, HTTP status code: OK (200)
URL: https://example.com?iteration=58, HTTP status code: OK (200)
URL: https://example.com?iteration=50, HTTP status code: OK (200)

你会注意到第一个记录的条目始终是立即返回 429 响应，最后一个条目始终是 200 响应。这是因为速率限制是在客户端遇到的，并且避免了对服务器的 HTTP 调用。这是一件好事，因为这意味着服务器不会请求泛滥。这还意味着所有客户端上都一致地执行速率限制。

另请注意，每个 URL 的查询字符串都是唯一的：请检查 iteration 参数以查看它是否针对每个请求增加 1。此参数帮助说明 429 响应不是来自第一个请求，而是来自达到速率限制后发出的请求。 200 响应的完成时间较晚，但这些请求在达到限制之前完成了。

若要更好地了解各种速率限制算法，请尝试重写此代码以接受不同的 RateLimiter 实现。除了 TokenBucketRateLimiter，还可以尝试：

ConcurrencyLimiter
FixedWindowRateLimiter
PartitionedRateLimiter
SlidingWindowRateLimiter

总结

本文介绍了如何实现自定义 ClientSideRateLimitedHandler。此模式可用于为已知具有 API 限制的资源实现速率限制的 HTTP 客户端。通过这种方式，可以防止客户端应用向服务器发出不必要的请求，并且还可以防止应用被服务器阻止。此外，通过使用元数据来存储重试计时值，还可以实现自动重试逻辑。

通过

在 .NET 中限制 HTTP 处理程序的速率

什么是速率限制？

实现 `DelegatingHandler` 子类

模拟多个并发请求

总结

另请参阅

反馈

其他资源

通过

在 .NET 中限制 HTTP 处理程序的速率

什么是速率限制？

实现 DelegatingHandler 子类

模拟多个并发请求

总结

另请参阅

反馈

其他资源

实现 `DelegatingHandler` 子类