Quick update: I have enabled G1GC for the JVM running the broker and since then had no problem again. The master broker stays master even under very heavy load.
So, my suggestion and recommendation when using replicated LevelDB would be to use the G1 garbage collector significantly reducing "stop-the-world" GCs causing a time-out on the connection to Zookeeper which will ultimately "demote" a broker to Slave without real reason. However, while running my tests, I noticed the following "Warnings" in the log files of both slaves: 2014-10-21 09:44:37,694 | WARN | Invalid log position: 1569963680 | org.apache.activemq.leveldb.LevelDBClient | Thread-2 There are probably hundreds of those (didn't actually count them) with obviously changing log positions. I will retry my tests by switching to the Java implementation of LevelDB. -- View this message in context: http://activemq.2283324.n4.nabble.com/Potential-Bug-in-Master-Slave-with-Replicated-LevelDB-Store-tp4686450p4686580.html Sent from the ActiveMQ - User mailing list archive at Nabble.com.
