Open Stack

On 06/18/2013 09:07 AM, Sebastien Han wrote:
> Hi Jay,
>
> Could you please elaborate on this one:
>
>> * do NOT use the SQL driver for Keystone catalog. Use the templated
>> catalog driver.

If you have multiple regions and availability zones returning the same 
service types  -- e.g. compute -- and you point your 
nova|quantum|cinder|glanceclient to a Keystone endpoint in one of those 
zones that is sharing the identity database with other zones, then the 
Keystone endpoint will return the service catalog for all of the regions 
and zones. At this point, the X-client will happily go to the first 
matching service type endpoint for its service type, irrespective of 
which region or availability zone you actually originally hit.

Using the templated catalog driver means you can set up your Keystone 
endpoint to only return the region and availability zone that that 
particular endpoint is in, and not everything in the services and 
endpoints tables.

It's the reason that I brought up at the design summit that the service 
catalog is pretty much a broken model, and proposed what I thought was a 
relatively minor fix/change to the API:

https://review.openstack.org/#/c/27563/

That was subsequently poo-poo'd.

Best,
-jay