[JBoss-dev] jsr-77 and undeploying ejbs

[Rod B. <rod...@we...>]

There's a bug in here somewhere, but I'm not sure exactly where because I'm
not sure exactly about the intent.  Here's the process, as I read it:

 - when an ejb is deployed, the LocalJBossServerDomain is notified and a
statistics mbean is registered for the ejb
   - this is handled by EJBModuleFactory.createEJB() and EJB.create()
 - if the ejb is undeployed, any notification is ignored by
LocalJBossServerDomain, and the stats mbean remains registered
 - if the ejb is redeployed, EJBModuleFactory.createEJB() and EJB.create()
again create and attempt to register a stats mbean
 - registration fails because the object is still in use by the original
stats mbean
 - EJB.create() returns null
 - EJBModuleFactory.createEJB() puts this null value into it's
containerToModuleNameMap
   - the factory here loses it's reference to the original stats mbean
 - stats mbean unregistration is attempted upon notification of ejb service
destroy
   - but it doesn't happen because of the null value previously put into
containerToModuleNameMap

Any stats mbean method invocation that goes through the mbean server will
operate on the original stats mbean, accumulating stats across deployments.
This stats mbean won't be unregistered when the actual ejb is destroyed.

If ejb stats are supposed to accumulate over multiple deployments (as they
now are), then handling of the 'mbean already registered' situation should
be changed.  Maybe EJB.create() should return the existing object name, or
maybe EJBModuleFactory.createEJB() should check for create() returning a
null reference, or maybe there's another alternative.

If ejb stats are NOT supposed to accumulate, but to cover only a single
deployment, then the server domain should pay attention to the notification.

It looks to me like there are three options:
  a) there's a bug associated with creation of the stats ejb mbean
  b) there's a bug associated with destruction of the stats ejb mbean
  c) I'm misreading something

Which is it? a?, b?, c?, a and c?, b and c?


Rod Burgett 
Senior Software Engineer 

webMethods, Inc. 
3930 Pender Drive 
Fairfax, VA  22030  USA 
Ph: 703.460.5819  (tty only) 

"It's all just 0s & 1s - 
 the trick is getting them lined up in the proper order"