[jboss-jira] [JBoss JIRA] Commented: (JBREM-1010) Add feature to declaratively turn on leasing for 2.2.2

Add feature to declaratively turn on leasing for 2.2.2 ------------------------------------------------------ Key: JBREM-1010 URL: https://jira.jboss.org/jira/browse/JBREM-1010 Project: JBoss Remoting Issue Type: Feature Request Security Level: Public(Everyone can see) Components: general Affects Versions: 2.2.2.SP7 Reporter: Jay Howell Assignee: Ron Sigal Fix For: 2.2.2.SP9 We have a request to declaratively turn on leasing. Currently you can only do it via debugging. Here's the conversation from Ron so far. It says more than I could. -------------------------------------------------------------- The problem lies in org.jboss.aspects.remoting.InvokeRemoteInterceptor, which (1) opens an org.jboss.remoting.Client, (2) makes the invocation, and (3) closes the Remoting client. I guess that's because there's no notion of a "session" or a "lifetime" with EJBs, so there's no other natural place to close the Client. The other part of the story is the relationship between Client and the actual invoker, org.jboss.remoting.transport.socket.MicroSocketClientInvoker in this case. Multiple Clients connected to the same server will share an invoker, increasing its reference count. So if multiple threads were calling on the same EJB at the same time, closing the connection after an invocation wouldn't close the MicroSocketClientInvoker (and its cache of connections) if another thread was in the middle of another invocation. The gap in my explanation is the lack of evidence of multithreading in the example. I don't know what method1() is doing. But if there were some simultaneous processing going on, then enabling leasing would add an extra invocation to the process of connecting and disconnecting, thereby changing the timing of things, which might then lead to keeping a MicroSocketClientInvoker alive long enough to be reused. Unfortunately, there's no declarative way to turn on leasing. Just an oversight, I guess, and it's never come up so I never thought about fixing it. I could, though. However, there's a more direct approach. I put a feature in Remoting 2.4 which makes it possible to configure Client.disconnect() to delay the destruction of the client invoker by putting it in a TimerTask: JBREM-877 "New Socket Connection is being Created for Every Client Request to the Server". On the EJB3 forum thread http://www.jboss.com/index.html?module=bb&op=viewtopic&t=130264, I asked the EJB3 guys to create a JBPAPP issue if they wanted the same feature in Remoting 2.2, but it never happened so I never did it. Since it's certainly a Remoting feature, and not really an EJB3 bug, I guess it's a judgment call whether to go ahead and make the change in Remoting 2.2. But Jay, if you create the JIRA issue(s), I'll be happy to do it and attach a preview jboss-remoting.jar. --------------------------------------------------------------