See Thread at: http://www.techienuggets.com/Detail?tx=52347 Posted on behalf of
a User
I think you'll want to modify the EJB stack defined in
server/all/deploy/ejb3-interceptors-aop.xml.
You can add your custom interceptor or simply remove/comment-out the "IsLocal"
interceptor.
For example, if you want to make the change for clustered stateless session
beans the applicable section could look like this:
<stack name="ClusteredStatelessSessionClientInterceptors">
<!-- interceptor-ref
name="org.jboss.ejb3.remoting.ClusteredIsLocalInterceptor"/ -->
<interceptor-ref
name="org.jboss.aspects.security.SecurityClientInterceptor"/>
<interceptor-ref
name="org.jboss.aspects.tx.ClientTxPropagationInterceptor"/>
<interceptor-ref
name="org.jboss.aspects.remoting.ClusterChooserInterceptor"/>
<interceptor-ref
name="org.jboss.aspects.remoting.InvokeRemoteInterceptor"/>
</stack>
Notice the commented out ClusteredIsLocalInterceptor.
I've been trying to do the exact same thing as you (i.e. cluster ejbs called
from within the same JBoss 4.2.3 server). And I just figured this out today.
I think it does what I want, but haven't fully "exercised" it in any way. Let
me know if this helps.
-Randy
In Response To:
Hi All,
After much reading up, it appears that I need to write a custom
InvokerInterceptor to achieve clustering of Session beans in JBoss (I am
running on 4.2.3) and I have a couple of questions before I venture down this
road (although it is not a lot of code when looking at it)
I have an application that runs entirely in the container, I no external
client, the system is implemented in Sessions beans, Entity Beans, JMX MBeans
and MDB Beans (it also has a minimal web client built on tapestry 5 for mainly
management and configuration tasks), all in EJB3, and is a fairly highly
stressed system which has become cpu bound, hence we are looking at gaining a
performance increase by clustering.
As the application is fully contained within the container, we have the
scenario where clients and session beans run in the same JVM (MDB's and JMX
beans act as clients as well as session beans dependent on other session beans)
so I predict a large performance increase by being able to load balance our
session bean calls between multiple servers, however the limitation of the
"optimization" where by the InvokerInterceptor routes calls to local instances
only is a big limiting factor and does not allow us to distribute the load,
hence looking at implementing a custom InvokerInterceptor.
1. When overriding hasLocalTarget and simply returning false, will the round
robin process still send some requests to the "local" session bean instance as
well as the "remote" instances in the cluster, or will it send ALL requests to
the remote instances / nodes for processing and ignore the local instance all
together?
2. If the answer to 1 is "Yes, it will send all requests to remote nodes" under
the scenario of simply returning false, is the InvokerInterceptor "statefull".
What i mean by this is would something like the following work:
| public class CustomInvokerInterceptor extends InvokerInterceptor {
|
| private boolean _local = false;
|
| public boolean hasLocalTarget() {
| if (_local) {
| _local = false;
| return true;
| } else {
| _local = true;
| return false;
| }
| }
| }
|
such that between calls to hasLocalTarget, the state of _local will be
maintained, or will the be a fresh instantiation between invocations and thus
always return false? If this is not the case and all calls end up going remote,
we would essentially have an under utilised node on the cluster that simply
acts as a client to the other nodes (I hate wasted resources :) )
3. Based on the fact that there are a number of threads regarding this issue
over some time now, is it still an issue? IE, in 4.2.3 or even 5, do we still
need to implement a custom InvokerInterceptor to "de-optimize" JBoss so we can
optimize out application, or has it now been made configurable such that we can
change a config setting to tell the container that we want a certain behavior
or some other way to achieve RoundRobin load balancing? I understand the fact
that it is more resource intensive to route a call to a remote node when it
could be serviced locally, but that rule doesn't cater very well for the
scenario where the local node is 100% CPU stressed and an application has
become CPU bound.
Cheers,
Ben
_______________________________________________
jboss-user mailing list
jboss-user@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/jboss-user
