[jboss-jira] [JBoss JIRA] Created: (JBAS-4505) Inter-jar dependencies between clustered EJBs leads to bean not accepting requests

[Brian Stansberry (JIRA)]

Inter-jar dependencies between clustered EJBs leads to bean not accepting requests
----------------------------------------------------------------------------------

                 Key: JBAS-4505
                 URL: http://jira.jboss.com/jira/browse/JBAS-4505
             Project: JBoss Application Server
          Issue Type: Bug
      Security Level: Public (Everyone can see)
          Components: Clustering, EJB2
    Affects Versions: JBossAS-4.2.0.GA
            Reporter: Brian Stansberry
         Assigned To: Brian Stansberry


Scenario:  2 clustered EJB2 beans, A and B, deployed in A.jar and B.jar.  A declares a dependency on B via a <depends> in jboss.xml.  A.jar gets deployed before B.jar.

Problem:  The latch in bean A's HATarget that allows invocations to reach the bean never gets released.  The bean is available in JNDI, but any thread that attempts to invoke on it will block (forever) on the latch.  The bean becomes a black hole for threads.

Cause: ProxyFactoryHA registers a JMX NotificationListener to listen for a service started event.  Receipt of that event triggers release of the latch.  Problem is it listeners for the start of the EJB module (A.jar) rather than the target bean.  Reason for this is described in the ProxyFactoryHA.create() method:

// ************************************************************************ 
// NOTE: We could also subscribe for the container service events instead of the
// ejbModule service events. This problem comes from beans using other beans
// in the same ejbModule: we may receive an IllegalStateException thrown
// by the Container implementation. Using ejbModule events solve this
// problem. 
// ************************************************************************
this.container.getServer ().
         addNotificationListener (this.container.getEjbModule ().getServiceName (), 
                                  new ProxyFactoryHA.StateChangeListener (), 
                                  filter, 
                                  null);

My interpretation of the comment is that if you allow access to the bean before the entire module is started, and the bean calls into a 2nd bean in the same module, then a request could come it via the 1st bean that tries to prematurely access the 2nd bean. Waiting for the entire module to start is intended to prevent this.  The assumption here is that all beans within the module will complete the start phase before the module itself.

Problem is the depends element on a bean breaks this assumption.  What happens is you get the following start order:

A.jar
bean B
bean A
B.jar

Effect is the notification for A.jar start happens before the HATarget for bean A even exists. The latch that A's HATarget later creates will thus never get released.

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira