Alan, thanks for the reply. I've done more digging into the issue we are having and it became apparent that its only happening when we have qpidd operating in a clustered mode. If I run qpidd with no clustering enabled its working perfectly, the memory use never exceeds relative queue depth for the most part. Once I enabled clustering to one additional node the memory use of qpidd will start to grow out of control. One thing I notice when running qpid too is the metrics of this object
org.apache.qpid.broker:queue 3 0 When in cluster mode there is a very excessive amount of Deleted queue objects that constantly increases. This seems like its due to failover.* queues. This is really unfortunate as the main reason the decision was made to go with qpid was because of the clustering support, and again, when no clustering is enabled memory use is fine. Any suggestions or thoughts? Version wise, we are using the 0.5 release of qpid, with openais 0.80.3. Our clients are running against the C++ API. Etrik --- On Thu, 6/11/09, Alan Conway <[email protected]> wrote: > From: Alan Conway <[email protected]> > Subject: Re: Qpid C++ Broker memory use > To: [email protected] > Date: Thursday, June 11, 2009, 12:03 AM > > qpidd is currently using the standard C++ new/delete to > manage memory, so memory freed by qpidd is not > necessarily returned to the OS immediately, particularly if > the heap is fragmented. I'm not surprised at your results, > but I haven't done much heap profiling to date so I can't > say much about when you would start to see memory being > released back to the OS. > --------------------------------------------------------------------- Apache Qpid - AMQP Messaging Implementation Project: http://qpid.apache.org Use/Interact: mailto:[email protected]
