Hi Peter, how does your failover URI looks like? Client (producer/consumer) should only try to reconnect to the urls contained in the failover uri.
Cheers -- Dejan Bosanac - http://twitter.com/dejanb Open Source Integration - http://fusesource.com/ ActiveMQ in Action - http://www.manning.com/snyder/ Blog - http://www.nighttale.net On Wed, Oct 20, 2010 at 9:46 AM, Pieter C <[email protected]> wrote: > > Hi ActiveMQ, > > I'm trying to make a setup with four brokers, two on side A and two on side > B for high availability. So at each side one broker is active, the other one > is waiting for the kahadb-lock of a shared disk. In between side A and B > the brokers connect with failover through SSL. The producers and consumers > in side A and B connect via failover through openwire. Producers, consumers > and brokers use ActiveMQ 5.4.1 > > The happy-path works fine, but because of the network of brokers the > producer at side A gets informed about the openwire connection at side B (I > debugged the source and it happens via a BrokerInfo object that is > transferred to the producer during the startup of the connection). During a > failover the producer at side A, who is producing to a failover string, > suddenly tries (and succeeds) to connect to the openwire port of a broker at > side B. This is not the behavior I want, the consumer should only know > about the connections at side A, not about the SSL connections between A and > B and the openwire connection at side B. Is there a way to let the network > not propagate the topology of the network? For the borkers and for the > clients? This problem could be solved via firewall settings but there must > be a cleaner way? > > Some smaller questions: > - What does the parameter trackMessages of the failover uri do exactly? I > use this because I thought it was logic that it should be true. In a > transactional connection I sometimes lose a complete transaction during > failover, this happens less with the trackMessages on true. > - Am I correct in saying that when you use a transactional connection, the > producer (eg java program) should handle the TransactionRolledBackException > you get during a failover itself? > - I think the bug AMQ-2803 still exists in 5.4.1, the zombie messages, > because after a failover (while sending eg 1000 messages to a queue) the > pending message size stays one or two (connection is not transactional) > - Should the producer and consumer use a failover uri? Or can it just > contain the two openwire connections? > > I could be wrong, but it’s difficult to get a good grip on the stability of > this complete failover setup, and that is just what I need, stability! > > Thanks in advance > > Pieter > -- > View this message in context: > http://activemq.2283324.n4.nabble.com/Questions-about-network-of-brokers-with-failover-tp3003405p3003405.html > Sent from the ActiveMQ - User mailing list archive at Nabble.com. >
