Hi, Several weeks ago I discovered that some odd bug was preventing the JournalPersistenceAdapter from being notified when the UsageManager registered a memory increase. The outcome of this behavior is that when memory gets filled with persistent messages, producers block until the periodic (5 min. I believe) checkpoint task stores all messages and evicts the in-memory copies.
I was unable to discern exactly what was happening, but it appeared that the CopyOnWriteArrayList that the UsageManager uses for managing its listeners was getting trashed. Since ActiveMQ uses a back-ported implementation of this class, I suspected some subtle misbehavior between JVM and API. When I checked out the source from trunk several weeks ago, I found that this issue had disappeared. Now, playing with RC3, I note that it is back. Does anyone know anything about this? Chris
