[openstack-dev] RabbitMQ Scaling

Russell Bryant rbryant at redhat.com
Tue Nov 13 19:33:51 UTC 2012


On 11/13/2012 02:24 PM, Vishvananda Ishaya wrote:
> I guess I'm not seeing what is scary about this. It seems like we can just
> create a single response queue, and keep a dictionary of calls by message id.
> When we get a response we wake up the greenthread that was waiting on the
> response. If there isn't one, we log a warning and move on. This seems
> preferrable to just having the message sit in the queue forever (which can
> happen currently).

The reply queues should get destroyed once a caller gives up waiting for
a response, so messages shouldn't be able to just sit in a queue forever.

I do see your point in just tagging responses with an ID for the call
they are associated with.  I guess I was only thinking about the way
responses are formatted right now, which is just a raw return value and
the queue it shows up in was the request ID.  Seems like it should work ...

-- 
Russell Bryant



More information about the OpenStack-dev mailing list