[Design of POJO Server] - Re: Handling component references
by wolfc
Andrew Lee Rubinger (new EJB 3 dev) wants to build a service locator. This depends on an interface registry. So he's going to create one in projects/ejb3 ejb3-registrar. I think the deployers also need such a registry to resolve interface dependencies. Do you concur with that dependency?
Should we put it in projects/ejb3? Or in integration?
I want to limit the functionality of the component to (scoped) interface resolution only. Not ejb-links. So you'll get something like:
interface Registry {
| JavaEEComponent getComponentByBusinessInterface(Class<?> businessInterface);
| }
Currently for ejb-links I've taken an approach where a JavaEEComponent asks it's JavaEEModule to resolve an ejb-link (in a weird way, via createObjectName). This should become: getComponent.
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4084339#4084339
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4084339
17 years, 1 month
[Design of Clustering on JBoss (Clusters/JBoss)] - Re: JBAS-4717 - Problem with @JMX for clustering
by bstansberry@jboss.com
I think the solution may be simpler. Seems like there are two problems:
1) ServiceMBeanSupport.jbossInternalCreate() doesn't have the kind of state transition validity checking logic that jbossInternalStart()/Stop()/Destroy() do. So, two calls to create() result in two calls to createService(). This won't happen with the other service lifecycle methods. Do you know any reason why ServiceMBeanSupport.jbossInternalCreate() doesn't start with something like
| if (state == CREATED || state == STARTING || state == STARTED
| || state == STOPPING || state == STOPPED)
| {
| return;
| }
|
2) ClusterPartition is doing naughty stuff by making invocations on another service in createService(). That's going to be moved to startService().
I've been testing a fix based on #1 and I believe I avoided the problem you posted. But it looks like something's changed in the deployment ordering that's exposed a bug in how DRM coordinates with ClusterPartition. Effect is the server start hangs. So, I've got to sort that before I can confirm that fixing the above 2 issues gets rid of the problem.
View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4084291#4084291
Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4084291
17 years, 1 month