Some additional info
Today I again spend some time for testing the loadbalancing of the HQ-2.2.5 cluster. For it I downloaded a fresh jboss-as-6.1.final distribution and created two all-configurations (s1, s2). All what I edited in the hornetq-configuration.xml of both servers is setting them to use the paging and redistribution-delay=0. To run both jboss on the same machine I used the port-offset of the binding tool. The servers started cleanly…
The first strange cluster behavior is that the loadbalancing appears to lose messages. I connected a standalone jms client to the server that I started first and sent 1000 short textmessages. Like expected the 500 messages arrive the s1, BUT on s2 are only ~100 (as you can see in the jboss-admin-console). If I start the s2-server first, then the s1 lose messages (it seems to depend on the order the nodes started).
The redistribution has also big problems to work properly: 200 Messages are sent to s1. On both nodes are 100 messages to see. Then I connect my client to s2 and like expected all messages from s1 are redistributed to s2. Then I cleanly close the client connection and establish it on s1. The redistribution starts (s2->s1), BUT only 100 messages travel to s1 from s2. The other 100 messages rest on s2. Moreover it is not possible to remove them using the jboss-admin-console.
In both cases there were no warning or error output.
All it seems for me like a critical bugs.
Dear hq-developers, can you confirm it? Can it be that my configuration or even my environment is responsible for this behavior? Or do I maybe misunderstand something?
On my HQ 2.2.14 within jboss as 6.1 final the redistribution doesn’t work.
Strange, but now it works, with the same configuration. Anyway, the disconnecting causes the attached warnings.