[openstack-dev] RabbitMQ Scaling

Ray Pekowski pekowski at gmail.com
Tue Nov 27 17:48:26 UTC 2012


Andrea,

On Fri, Nov 16, 2012 at 3:41 AM, Rosa, Andrea (HP Cloud Services) <
andrea.rosa at hp.com> wrote:

>
> I am really surprised that you have the same results. The configuration
> with RAM nodes should have an impact on the performance.
>


My guess as to why performance did not improve with RAM nodes in my testing
is that the dynamically created queues and exchanges are defined as auto
delete.  They go away when the RabbitMQ server/node is stopped.  Guessing
that auto-delete things don’t need their configuration information written
to disk is why I don’t see disk I/O during their dynamic creation/deletion
and thus no improvement when going to RAM nodes.



Here are the options used for the dynamically created call response
queues/exchanges (from impl_kombu.py):



        options = {'durable': False,

                   'auto_delete': True,

                   'exclusive': True}



Here are the options used for the static call request queues/exchanges
(also from impl_kombu.py):



        options = {'durable': conf.rabbit_durable_queues,

                   'auto_delete': False,

                   'exclusive': False}



That durable conf option is defined as false:



    cfg.BoolOpt('rabbit_durable_queues',

                default=False,

                help='use durable queues in RabbitMQ')



So, all queues are defined as non-durable, but the auto-delete-ness of them
varies by what they are used for.

Ray
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20121127/6d0e08a2/attachment.html>


More information about the OpenStack-dev mailing list