Daniel Patterson wrote:
I've turned off persistance "<broker persistance="false">" and that
makes this problem go away, but
the broker then uses large amounts of memory, seemingly uncapped.
I've moved this process to
a directory on local disk and re-enabled persistence, and the problem
returns.
I've tried using the 4.1.0-RC2 release, the same symptoms occur. I've
also kept a JMX client attached and
watched the heap size over several hours. The order of events goes
something like this:
1. ActiveMQ startup, everything working fine. I'm using mostly
topics and I have 32 durable subscribers
(all active for the duration of the test).
2. After a couple of hours of operation, and 20k messages, an
"IOException: Not started" error is thrown
by the journal code.
3. After this, memory usage slowly increases, in proportion to the
number of messages recieved. While we're
below the maximum heap size, messages continue to flow.
4. We hit the maximum heap size. There are no errors logged
anywhere, but ActiveMQ stops responding
and message flow halts. Clients don't recieve any errors but
simply block. CPU usage is low, so
it doesn't look like GC thrashing.
I don't have a solid test-case to reproduce this, but should I go
ahead and file a bug?
daniel
