Hi,

Suppose we have store-and-forward network of two brokers (A and B) that
share a topic (T1).  Each broker has a single consumer for T1 (ConsumerA and
ConsumerB).  A and B communicate by producing and consuming messages from
T1.  Since ConsumerA does not want to consume the messages that it produces,
and vice-versa with ConsumerB, each consumer is created with noLocal=true.

This works great, except that each broker can only have one such consumer
(i.e., there is no load balancing) as described here:

http://activemq.apache.org/virtual-destinations.html

To allow multiple consumers, we "virtualized" T1 so that each broker has its
own private queue analog (Consumers.A.T1 and Consumers.B.T1).  In order to
load balance the traffic arriving through the topic, A has multiple
consumers for Consumers.A.T1 and B hsa multiple consumers for
Consumers.B.T1.

This works great if the consumers have noLocal=false, but if the consumers
have noLocal=true (as per our original desire to avoid processing reflected
messages), we run into the following problem:

A producer on A sends a message M to T1.  AMQ replicates M so that it
appears in broker A's Consumers.A.T1 queue and in broker B's Consumers.B.T1
queue.  Since the consumers for Consumers.A.T1 have noLocal=true, M is never
dispatched to a consumer on A.  This happens indefinitely until
Consumers.A.T1 is filled up (i.e., through jconsole we see the memory usage
on the queue as 100%).

Questions
=======
1) Is this behaviour intended?  I'm guessing that it is intended more as a
consequence of the semantics of virtual topics.
2) Is there any way to configure AMQ so that locally-produced messages to a
virtual topic are not copied to the local consumer queues? I'm guessing the
answer is no.
3) Are there any other workarounds besides using noLocal=false or setting an
expiry on the published messages?

Thanks,
Stirling
-- 
View this message in context: 
http://old.nabble.com/Problem-with-Virtual-Topics-and-noLocal%3Dtrue-consumers-tp26416480p26416480.html
Sent from the ActiveMQ - User mailing list archive at Nabble.com.

Reply via email to