openstack team mailing list archive

[Maru Newby <mnewby@xxxxxxxxxxxx>]

On 2012-03-09, at 7:52 AM, Jay Pipes wrote:

>> The data is stored in a python dictionary, inside of the metadata table.  You will not be able to use SQL without an unwieldy wildcard search.  IMO this seems overly complicated for a core function of the tool, and possibly the reason why listing user/tenant roles hasn't been implemented.
> 
> ++
> 
> I suspect the existing SQL schema has more to do with the default of using a key-value store until recently.
> 
> I think that storing in the roles relationships in the "extra" column is a bit of premature optimization that is a little ill-conceived at this point -- it sacrifices functionality for a perceived performance improvement. I don't believe there's any evidence that the join to a roles table (or two joins for a mapping many-to-many relationship table) had an adverse impact on performance in the legacy Keystone.
> 
> -jay


++

Storing data in a non-relational way when it is useful to access it in a relational way is doubleplusungood.  If a cache in the "extra" column is needed for performance it can always be maintained in addition to a table with transactional updates.

Thanks,


Maru