问题
My understanding, all implementations of ClientConnectionManager persist connections base on route. This results in basically no persistent connections if a proxy is involved. For example, the HttpClient needs to visit 1000 different domains via a HTTP proxy with an fix IP, it has to establish at least 1000 connection to the proxy instead of creating 1 persistent connection to the proxy and reuse that for the 1000 requests.
I'm simulating multiple users visiting thousands of domains (fake domains, all dns resolved to a couple of IPs, the resolving happen after the proxy, so nothing to do with HttpClient). The above behavior quickly use up all available ports in the localhost as I increase the # of users and domains, the Address Bind errors occur as result.
Is there a way to make the HttpClient to persist connection on proxy basis? ie. A HttpClient only maintain specified number of connections to a given proxy.
回答1:
After intensive research, it seems that Apache HttpClient doesn't support this behavior out-of-box. I have to modify the HttpClient/HttpCore source in order to have this feature, ie. maintain persistent connections based only on localAddress and First Proxy address.
The classes I modified are:
org.apache.http.conn.routing.HttpRounte.java and org.apache.http.conn.routing.BasicRouteDirector.java.
Basically I changed the hashCode and equal method in HttpRoute (which is used as a key to hashtable for persistent conn lookup), so the lookup doesn't consider target address if a proxy is involved.
Initial test results of above modification shows about 100 times improvement in terms of request throughput in my scenario. So far it works fine for me.
Kevin
来源：https://stackoverflow.com/questions/13425327/apache-httpclient-4-persistent-connection-per-proxy-instead-of-per-route