← Back to team overview

openstack team mailing list archive

Re: [keystone] Why are we returing such a big payload in validate token?

 

On 01/31/2013 10:57 PM, Vishvananda Ishaya wrote:
On Jan 31, 2013, at 6:37 PM, "Ali, Haneef" <haneef.ali@xxxxxx <mailto:haneef.ali@xxxxxx>> wrote:

Isn’t signed token an optional feature? If so validateToken is going to be a high frequency call. Also “Service Catalog” is a constant, the services can cache it. It doesn’t need to be part of validateToken.

Service catalog is not a constant. That said the only time it is used is when a service needs to proxy a call to another service using the same token. If we had a reasonable way to make requests on behalf of other users we don't really need it as the service could just keep its own catalog and make requests on behalf of the requesting user.

I'm working on it.  It is called "trusts" and there is a WIP posted here:

https://review.openstack.org/#/c/20289/

Blueprint is here:

https://blueprints.launchpad.net/keystone/+spec/trusts


Vish

Thanks
Haneef
*From:*openstack-bounces+haneef.ali=hp.com@xxxxxxxxxxxxxxxxxxx <mailto:openstack-bounces+haneef.ali=hp.com@xxxxxxxxxxxxxxxxxxx> [mailto:openstack-bounces+haneef.ali=hp.com@xxxxxxxxxxxxxxxxxxx <mailto:bounces+haneef.ali=hp.com@xxxxxxxxxxxxxxxxxxx>]*On Behalf Of*Adam Young
*Sent:*Thursday, January 31, 2013 6:25 PM
*To:*openstack@xxxxxxxxxxxxxxxxxxx <mailto:openstack@xxxxxxxxxxxxxxxxxxx>
*Subject:*Re: [Openstack] [keystone] Why are we returing such a big payload in validate token?
On 01/31/2013 07:44 PM, Ali, Haneef wrote:

    Hi,
    As of now  v3  validateToken response has “tokens, service
    catalog, users, project , roles and  domains.  (i.e)  Except for
    groups we are returning everything.  We also discussed about the
    possibility of 100s of endpoints.  ValidateToken is supposed to
    be a high frequency call .    This is


Validate token should not going be a high frequency call. The information is encapsulated inside the signed token for just that reason.

I would agree with the sentiment, however, that we are cramming a lot of info into the token. TOkens should be scoped much, much more finely: by default one service or endpoint, and one tenant.

The only thing that should require the full service catalog is the initial request of an unsigned token, and that should merely go back to the client.


going to be a huge performance impact . What is the use case for such a big payload when compared with v2? If a service needs catalog , then the service can always ask for the catalog.
Thanks
Haneef



_______________________________________________
Mailing list:https://launchpad.net/~openstack  <https://launchpad.net/%7Eopenstack>
Post to     :openstack@xxxxxxxxxxxxxxxxxxx  <mailto:openstack@xxxxxxxxxxxxxxxxxxx>
Unsubscribe :https://launchpad.net/~openstack  <https://launchpad.net/%7Eopenstack>
More help   :https://help.launchpad.net/ListHelp
_______________________________________________
Mailing list: https://launchpad.net/~openstack <https://launchpad.net/%7Eopenstack> Post to : openstack@xxxxxxxxxxxxxxxxxxx <mailto:openstack@xxxxxxxxxxxxxxxxxxx> Unsubscribe : https://launchpad.net/~openstack <https://launchpad.net/%7Eopenstack>
More help   : https://help.launchpad.net/ListHelp



References