简体繁体 English

PoolingHttpClientConnectionManager和保持活动设置

[英]PoolingHttpClientConnectionManager and keep-alive setting

原文 2017-11-29 23:10:43 3 1 java/ multithreading/ performance/ apache-httpclient-4.x

I have performance critical application with http requests pool. 我有带有HTTP请求池的性能至关重要的应用程序。 My application service is bursty processing batch files. 我的应用程序服务是突发处理批处理文件。 Does PoolingHttpClientConnectionManager re-use connection for multiple requests? PoolingHttpClientConnectionManager是否为多个请求重用连接？ If yes, how do i calculate optimal keep-alive time? 如果是，我该如何计算最佳保持时间？ My http request response time is avg 25 ms. 我的http请求响应时间平均为25毫秒。

1 个解决方案

Yes, it will reuse connections for multiple requests. 是的，它将对多个请求重用连接。 Note that it pools connections with a maximum number per route and the default is 2, so you will almost certainly want to increase that or it will become a bottleneck! 请注意，它汇集了每个路由的最大连接数，默认值为2，因此您几乎可以肯定希望增加连接数，否则它将成为瓶颈！

Calculating the best keep-alive time is non-trivial as it depends on many things. 计算最佳的保持活动时间并非易事，因为它取决于很多事情。 How often are the bursts, what keep-alive time does the web front or application server support and so on. 突发频率多久一次，Web前端或应用程序服务器支持什么保持活动时间等等。 Generally I would start with something like 5-10 seconds and measure. 通常，我会从5-10秒开始测量。 If you know that the bursts are 8 seconds apart, aim for something slightly larger. 如果您知道爆发间隔为8秒，则将目标对准更大一点的目标。 If the bursts are more than a few minutes apart, keeping the connections alive for that long will be problematic as they tend to time out (firewalls, server-side close, ...). 如果突发间隔不超过几分钟，那么保持连接的活动时间会很长，因为它们往往会超时（防火墙，服务器端关闭等）。