Address-full-policy in cluster nodes set to BLOCK but not blocking.

[mb69ou812]

Using HornetQ 2.4.5.FINAL within Wildfly 8.2.0 Final, I have a three-node symmetric cluster.  There is a producer sending messages to node 1 but node 1 has no queues.  For load balancing, nodes 2 and 3 have the address, queues, and consumers.  (I shut down the consumers to test address full behavior.)

When the address-full-policy is set to PAGE for the address in question, paging works as expected in nodes 2 and 3.  But when it is set to BLOCK, nodes 2 and 3 do not block nor do they page.  They just keep accepting messages!

This is what I am really after.  Can the producer attached to node 1 be blocked or notified when the address memory is full in either of nodes 2 or 3?  Ideally without paging or messages being dropped.

Thank you.

[jbertram]

Are the messages actually being distributed to nodes 2 and 3 when they should be blocked or are they simply being stored in the store-and-forward queue on node 1?  In other words, are you seeing the message counts on node 2 and 3 go up and up without ceasing?

[mb69ou812]

Message counts go up on 2 and 3 without ceasing.

[jbertram]

Could you work up a simple example so that I could reproduce what you're seeing, perhaps based on the "symmetric-cluster" example shipped with HornetQ?

[mb69ou812]

I have not been able to reproduce it using the symmetric-cluster example, but its configuration is quite different from the environment where I have the issue.  As I have time I will continue to try to create a test case that reproduces it.