I have a two node symmetric cluster (without backup nodes) that both use ramdisk to write the journals. The ramdisk is set to 1 GB on both machines. Under high loads when both machines are taking traffic (both machines have producers and consumers), I see new journals created very frequently and then later cleaned up. I've had to adjust the journal-compact-percentage to 75 just so the journals are compacted well before I run out of space on the ramdisk. For example, over a couple of days I see the journal id (the id in the file name) get into the 200k to 300k range.
Oddly enough, when I force all consumers and producers to be on just one of the two nodes, I see very little new journal creation and cleanup.
I have 22 queues, which about 6 have about 2000 messages go in and out of them every second (each machine has about 60 Mbps of traffic in and out split evenly). Each queue has max-size-bytes set to 10485760, so it shouldn't be that journals should fill up any more than 220 MiB, right? My journal-min-files is set to 10 and journal size is 10485760, so I start out with 100 MiB in journal space.
I also see this message in the logs, but not sure if it's related -
[2012-04-28 00:16:57,325] 183289211 WARN [Thread-0 (HornetQ-server-HornetQServerImpl::serverUUID=b67431cf-8f1c-11e1-b7e8-000c29848f2f-31583366)] org.hornetq.core.server.impl.QueueImpl - Couldn't finish waiting executors. Try increasing the thread pool size
Attaching my hornetq config files, as well as the start script I use to bring it up (where I read the local.properties file) and the hornetq log of when the ramdisk did fill up.