[Openstack] Fwd: [openstack] [keystone] [horizon] regions setup

Jay Pipes jaypipes at gmail.com
Thu Jan 2 18:14:12 UTC 2014


On 01/02/2014 12:32 PM, Xu (Simon) Chen wrote:
> A few questions..
>
> First, I am a little confused by this post:
> http://docs.openstack.org/trunk/openstack-ops/content/segregate_cloud.html
>
> On the one hand, it says different regions should have no interactions
> among them. On the other hand, it says keystone should be shared across
> regions. I can see that sharing credentials is useful, but replicating
> things like tokens across region seems to be a hassle to deal with - I
> don't want to replicate the tokens that are specific to regions via WAN..
>
> Second, I am confused about Horizon's multi-region support. There are
> two ways of informing a horizon instance about multiple regions. One way
> is to configure the AVAILABLE_REGIONS variable in local_settings.py,
> where I can put keystone endpoints associated to different regions. Then
> something would show up in the top right corner of horizon, that I can
> switch to a different region, log in, and it works. The second way is to
> configure the endpoints of another region in the keystone instance local
> to horizon. Then, a drop down list would show up on the left side of the
> page, right beneath the list of projects. This however doesn't work,
> since the openstack_auth package seems to be performing a simple
> redirect assuming the same token would work across regions (my two
> regions have completely separate keystone deployments.)
>
> Any ideas on the best practice here?

Hello there, Simon! :) Happy New Year!

My best advice to you would be to share identity/role/group information 
across regions (just so your users don't have to deal with separate 
creds in each region), but use the memcached token backend in each 
region's Keystone service. That way, you get the advantage of shared 
credentials but get decent token performance. As you point out, 
replicating tokens across the WAN is deadly for performance, as just a 
small number of users can quickly swamp the replicated database traffic 
from millions of tokens created and replicated.

I have no played with the AVAILABLE_REGIONS thing in Horizon yet, as I 
was under the impression that it relied on shared-region tokens 
(otherwise, users would have to grab a different token when doing things 
in different regions..)

Our users so far have not complained about simply going to the Horizon 
dashboard of the particular region they are working with, but I 
understand from Ryan Lane and others that that isn't a great user story!

All the best,
-jay




More information about the Openstack mailing list