On 06/12/2007, sebb <[EMAIL PROTECTED]> wrote: > On 06/12/2007, sebb <[EMAIL PROTECTED]> wrote: > > On 06/12/2007, Roland Weber <[EMAIL PROTECTED]> wrote: > > > Hi Sebastian, > > > > > > > If HttpClient (or my app) is failing to close sockets at the end, > > > > perhaps it needs to be enhanced - otherwise it will be causing the > > > > server to go into TIME_WAIT instead? > > > > > > I'm not sure whether the server needs to go into TIME_WAIT > > > if it gets a reset packet. But it can't hurt to call > > > SimpleHttpConnectionManager.shutdown() when exiting > > > the application. > > > > Oops I forgot: > > JMeter 2.3.1 calls closeIdleConnections(-1000) in the threadFinished() code. > > > > The latest (unreleased) code calls shutdown() instead. > > > > I'll add that to the test application and see if it helps. > > That seems to have solved it: > > With the added shutdown() call in the test app, HttpClient now behaves > the same as Java, i.e. the connections to google.com and .co.uk are > both left in TIME_WAIT state if keep-alive is selected, but not if > keep-alive is not. > > So your analysis was spot on. > > Thanks for all the replies.
Just for completeness, I've also tried running the test application without the Java disconnect(). It behaves the same as HttpClient, i.e. the google.com connection goes to TIME_WAIT and the final .co.uk connection disappears. With disconnect(), Java leaves both connections in TIME_WAIT state. So in order to ensure an orderly closure, "Java" users should call disconnect(), and HttpClient users should call shutdown() before exitting. When using keep-alive, this will result in TIME_WAITs on the client; if not using keep-alive the client does not see TIME_WAIT (because the server will have disconnected first). I hope that's now the full story ;-) > > > Looks like we forgot to add that method > > > to the interface, but it's too late for 3.1 now. > > > > I noticed that. Perhaps worth a Jira in case a bugfix release is needed? > > > > > I wouldn't want to mess with shutdown hooks [1] by default, > > > but I will remember the problem when we start working on > > > an HttpClient 4.0 Tutorial. > > > > Maybe add a shutdown method to the HttpClient API? > > [Perhaps that exists anyway] > > > > In 3.1, one has to get the conn manager and check if it is one of the > > mgrs that supports shutdown. > > > > > cheers, > > > Roland > > > > > > [1] > > > http://java.sun.com/j2se/1.5.0/docs/api/java/lang/Runtime.html#addShutdownHook(java.lang.Thread) > > > > > > > > > > > > --------------------------------------------------------------------- > > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
