Aidan Skinner wrote:
Hi,
I ran across a situation where the broker behaves rather poorly. If a large
number of unprocessed Jobs and Events build up (such as a fast producer and
a slow message store) they can cause high memory usage, which causes the GC
to kick in a lot and the whole thing gets a bit messy.
I've written up a design for temporarily suspending the production of those
when memory is low, comments welcome here:
http://cwiki.apache.org/confluence/display/qpid/Broker+job+queue+limits
I'd prefer to keep the discussion on the wiki so it's all in one place.
Thanks,
- Aidan
The C++ broker has threshold for this case -- yes it is the BIG hammer.
would be interested to
compare strategies.
Carl.
