Your directory still was able to confirm that data is indeed being written to the KahaDB store, so thank you for that. One thing I notice is that you've got a large number of sequential files that haven't been deleted, which implies that there are messages in those files that aren't being consumed. Are you certain that during your performance test, all messages published were being consumed? I'd recommend you try to confirm what you believe to be true by viewing the JMX MBeans that represent your topics, to make sure that the count of dequeued messages really is increasing at the expected rate.
It would be useful to know whether the growth in memory happens gradually over the full length of your performance run or happens in one or more "spikes" of additional memory usage. I've found I get the most useful memory/GC information out of JConsole; you can attach it to the broker when you start the performance test, and then just leave it running and watch what the graph on the memory tab does over time. Also, did the "Failed to load next journal location" warnings disappear with 5.15.0? https://issues.apache.org/jira/browse/AMQ-6432 appears to be what you were hitting, and it should be fixed under 5.15.0. If it's gone in 5.15.0 but the OutOfMemoryError is still happening, then that implies that the two are not related. Tim On Tue, Jul 11, 2017 at 1:27 AM, Hidekazu <hi_de...@yahoo.co.jp> wrote: > Hi > > I faced the same problem with ActiveMQ 5.15.0. > Configurations and environments are almost the same with 5.14.3. > > Hidekazu > > > > -- > View this message in context: http://activemq.2283324.n4. > nabble.com/AcativeMQ-becomes-slow-suddenly-in-60-minutes- > after-starting-a-performance-test-tp4728103p4728434.html > Sent from the ActiveMQ - User mailing list archive at Nabble.com. >
