masterslave: is an alias for failover: with maxReconnects=0. (There might be another URI option included in the alias, I don't remember; I think the details are in the JIRA where Gary added the failover transport, if you're curious.) So there's no need to try using failover explicitly, since the configuration you're using already used the URI option I was concerned about.
That's strange that you're seeing different throughput (based on different numbers of pending messages) based on which direction messages flow between the clusters. It might be due to the network issues you referenced; if not, then hopefully you can figure out which link is the slow one by finding the last point where messages are piling up. Is there a non-trivial amount of latency (more that 15ms, let's say) on any of the links between brokers or the links between clients and brokers? I've had to do quite a bit of tuning to get ActiveMQ to run efficiently across a high-latency WAN, so if you have a bad network link in your setup you may need to make some adjustments to improve throughput. Also, just to confirm: were you comparing pending queue sizes based on which role (producer-side or consumer-side) the cluster was being used for your test? (So comparing amq1-3 in your first setup with amq4-6 in your second setup and vice versa.) Make sure your comparisons were apples to apples between the tests, otherwise the conclusion of lower throughput might not be valid. On Fri, Oct 17, 2014 at 4:20 AM, uromahn <ulr...@ulrichromahn.net> wrote: > Quick update... > > I re-ran my tests as suggested. > > First my producer connected to amq4 and the consumer to amq1. That setup > worked quite well without any error or timeout. > Then I re-configured it again with messages being sent to amq1 and consumed > from amq4. To my surprise it worked this time (I re-ran the test three > times > yesterday and all three failed the same way!). > However, I noticed that when transmitting messages from amq1 to amq4, it > appears that the bridge is slower than the other way around since I saw on > average 10k pending messages in all queues - the other way around, there > were on average less than 1,500 msg pending. > > On another note: I mentioned that I setup this environment in our private > cloud infrastructure. This morning I saw a note from our infrastructure > guys > that we are having some network issues in that data center hosting this > environment. It is certainly possible that my issues yesterday may have > been > a side-effect of the network problems. > > So, I will keep testing but for now, my suspicion is that it is likely to > be > a network problem and not an issue within ActiveMQ. > > I will follow-up in case the issue shows up again. > > Until then, sorry for potentially raising a false alarm. > > > > -- > View this message in context: > http://activemq.2283324.n4.nabble.com/Potential-Bug-in-Master-Slave-with-Replicated-LevelDB-Store-tp4686450p4686492.html > Sent from the ActiveMQ - User mailing list archive at Nabble.com. >
