[Openstack] Messaging reliability/durability expectations

Gordon Sim gsim at redhat.com
Thu Oct 16 11:23:02 UTC 2014


On 10/14/2014 10:48 PM, Aaron Knister wrote:
> The fixes to all 3 of these issues seem to be patches to the rabbit
> driver for oslo. Are the other drivers (e.g. qpid) any more robust or
> are they just not heavily used so more bugs may be lurking there?

As mentioned, the qpid driver does not use acknowledgements for messages 
received from the broker at all, which means messages can be lost in 
transit (sent, but never received). It does share some code/concepts 
with the rabbit driver, including the same concept of configurable 
durability and autodelete which applies only in certain contexts (and 
not for example in reply queues, again allowing message loss).

So certainly using the qpid driver will not decrease the chance of a 
failure of some type resulting in an RPC failure.

My view is that a clearer statement of intent is required around the 
design for reliability, against which that design and indeed the 
implementation(s) can be evaluated.







More information about the Openstack mailing list