Problem with Connection Load Balancing

[vpothnis]

Hello,

I have a cluster of 3 HornetQ servers with server discovery. I have 'cluster connection' set up and the following is the configuration in hornetq-jms.xml.

 <connection-factory name="ConnectionFactory">
 <discovery-group-ref discovery-group-name="dg-group1"/>
 <entries>
 <entry name="ConnectionFactory"/>
 </entries>
 <connection-load-balancing-policy-class-name>
 org.hornetq.core.client.impl.RoundRobinConnectionLoadBalancingPolicy
 </connection-load-balancing-policy-class-name>
 <discovery-initial-wait-timeout>5000</discovery-initial-wait-timeout>
 <failover-on-server-shutdown>true</failover-on-server-shutdown>
 </connection-factory>

On the client side, I create 3 connections and from these I create 12 consumers/message listeners.

But, the connections and listeners do not get load balanced always. Some times they are load balanced but some times they are not.

What could be the cause for this?

Thanks
Vinay

[vpothnis]

I am using JConsole to check/verify that the connections have been established in a load balanced way to all the 3 HornetQ servers.

[vpothnis]

Update:
I noticed that when i am debugging through the loop that creates multiple connections, the connections were load balanced.

So, when I introduced a Thread.sleep(1000) in the loop where I was creating the connections, the connections were load balanced across the hornetQ servers in the cluster.

Has anybody noticed this? Any inputs on this?

[timfox]

Probably you haven't waited long enough for discovery information to reach the client.

Each server broadcasts its whereabouts every x milliseconds, when the client gets a broadcast from a server it will know about that server.

Until it gets a broadcast from a server it won't know it exists.

The broadcast period is configurable in the broad cast group config (broadcast-period).

Also there is a discovery-initial-wait setting on the session factory.

See user manual for more info.

[vpothnis]

Oh! That might explain this. Thanks Tim.
I will look into the "discovery-initial-wait" and the broadcast group configuration.

Thanks
Vinay

[vpothnis]

I think this is related to the defect in RoundRobin implementation.
http://www.jboss.org/index.html?module=bb&op=viewtopic&t=164343

I tested with a local fix on RoundRobinConnectionLoadBalancingPolicy.java and the load balancing worked correctly.