AspNet Core 3 HttpClientFactory：TaskCanceledException

Question

我在 Azure Kubernetes 服务中使用了一个 aspnet core 3 反向代理，大致基于 ProxyKit，它在以前的集群中工作得很好。在新集群上（我能看到的唯一区别是旧集群使用 kubenet，新集群使用 azure 虚拟网络）当上游请求花费超过几秒钟。

这是抛出异常的相关方法：

return await _httpClient.SendAsync(
                    UpstreamRequest,
                    HttpCompletionOption.ResponseContentRead,
                    HttpContext.RequestAborted)
                    .ConfigureAwait(false);

HttpClient 由 HttpClientFactory 使用类型化的 AddHttpClient 中间件提供。

到目前为止我尝试过的事情：

• 为 HttpClient
明确设置 30 秒超时
• 没有将 CancelationToken 传递给 SendAsync 方法
• 按照 this article
中的建议实施自定义超时处理

HttpClientFactory 之前是这样配置的：

var httpClientBuilder = services
                .AddHttpClient<ProxyKitClient>()
                .ConfigurePrimaryHttpMessageHandler(sp => new HttpClientHandler
                    {
                        AllowAutoRedirect = false,
                        UseCookies = false
                    });

这是现在的配置：

var httpClientBuilder = services
                .AddHttpClient<ProxyKitClient>(o => o.Timeout = Timeout.InfiniteTimeSpan)
                .ConfigurePrimaryHttpMessageHandler(sp => new TimeoutHandler
                {
                    InnerHandler = new HttpClientHandler
                    {
                        AllowAutoRedirect = false,
                        UseCookies = false
                    }
                });

行为没有任何改变。

如何确保 HttpClient 等待上游请求完成？ Kestrel 和 HttpClient 默认超时比现在中止的请求高得多。

附带说明一下，当我恢复到 aspnet core 2.2 时，行为完全相同。

Answer 1

我评论了：

Task cancellation for something like SendAsync is going to occur when the client closes the connection. As such, I think you're looking the wrong place. You need to figure out why the clients are closing the connection prematurely.

因此，OP 能够确定此问题与 Azure 应用程序网关有关：

Makes sense, and you are totally right. Was pulling my remaining hair out the whole day and the culprit is Azure Application Gateway with a request timeout of 1 second.