large messages in core bridge stay in 'delivering' although delivered

[carstenlenz]

Hi,

we have a simple setup with two hornetq servers in separate JBoss AS where messages from server A are bridged to a queue in server B via a core bridge.

When small messages are sent everything is ok, but large messages seem to 'hang' in delivering (the DeliveringCount JMX property is incremented and never goes down) although they are delivered correctly and consumed on server B. When restarting server A all messages which stayed in 'delivering' are resend and get thrown away on B because of duplicate message detection. As said, these message are large (in our production case from 5-30MB) and cause a lot of (unnecessary) traffic (connection between servers is WAN).

I tried the core bridge example and simply changed the first message to be sent into a TextMessage and set a large String as text, around 9 MB in my case. This shows exactly the same behaviour! (No configs changed)

Is there anything we can do about this? I consider replacing the core bridge with a JMS bridge and accepting the poorer performance but I would rather stay with the core bridge.

We use HornetQ 2.1.2 (tried also with 2.1.1).

thank you for your excellent product !

cheers

Carsten

[ataylor]

can you try with Trunk, this sounds very familiar and may be fixed

[carstenlenz]

Hi,

thanks for the fast response.

I will try with trunk today.

If it sounds familiar, is there by any chance a workaround without resorting to a JMS-Bridge?

[ataylor]

try setting the confirmation-window-size on the bridge to >0

[carstenlenz]

I tried on trunk with the same result. Sending a large message (9MB) results in DeliveringCount=1 continuously on the source queue. Setting confirmation-window-size > 0 (I tried 102400) doesn't help. Stopping and starting the bridge via JMX-Console gets the successfully consumed message being redelivered.

For our production problem I now configured a JMS-Bridge which doesn't show the same behaviour (after setting cache-large-message-client=true for the connection factory - without there is a exception already documented somewhere here in the forum).

[ataylor]

ok, i see, its specific to large messages, could you raise a jira please with instructions on how to recreate

[carstenlenz]

I created JIRA HORNETQ-638 and attached a patch for the core-bridge-example to reproduce.

Thanks for your help!

Carsten

[clebert.suconic]

Can you try with this branch please?

https://svn.jboss.org/repos/hornetq/branches/Branch_2_2_EAP/

Build it and test it.. we have made several changes on it. It would help us if you could verify it for us.

[carstenlenz]

I tried with Branch_2_2_EAP revision 10185 but I got the same result

You can apply the patch in the JIRA.