[rabbitmq-discuss] problem with new persiter with 1000 topics/queues

[alex chen]

Matthias,


> Delivered but unacknowledged messages are held in RAM. This is something
we are going to change in a future release. Meanwhile, are you setting a prefetch count? That will bound the number of unacked messages and thus the memory taken up by them.

I am setting prefetch count to 50.  each message is 20 KB.  so total usage should be
1000 queue * 20 KB * 50 which is about 1GB.   After all the messages were consumed, the broker's memory still remained at 4GB.  Is this expected or a problem with memory leak?   

Another observation with 2.1.1 is that the most rdq files in msg_store_persistent were not deleted until all the messages were consumed.  because the broker was so busy "rabbitmqctl list_queues" could not return the queue length, so i am not sure if the delay in deleting rdq files was caused by some consumers were lagging behind.  When i used release 2.0.0, the rdq files were deleted while the consumers were running.  Not sure if this contributed to the memory growth or not.

thank for looking into these problems!  it is greatly appreciated.

-alex


      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20101028/2664a490/attachment.htm>