Open Stack

On 21 November 2014 08:50, Michael Krotscheck <krotscheck at gmail.com> wrote:
> For API clients that require access to large parts of the data set, I've
> actually proposed a solution that doesn't involve the REST API at all. It's
> likely that this particular spec will also be the backbone of storyboard's
> federation (assuming we don't argue it into oblivion).
>
> https://review.openstack.org/#/c/105252/

That will certainly mitigate the frequency at which API clients scan,
but initial sync, and every time the socket connection is lost, will
trigger a full scan of the interest area of the client - as the spec
is written today anyway.

An architecture more like pubsubhubbub would mitigate that on
disconnects, if that is relevant. (Basically have any subscribable
thing have a last-changed field that can be read, and then api clients
can sync state by reading back until the last mod time is low than
their last fully-synced datestamp). Initial sync will still happen,
and that may be a concern.

Also, IME the clients doing full syncs as described are often naive
one-shot data analysis tools. They'd likely be better served by server
side map-reduce (e.g. js functions running in node).

-Rob

-- 
Robert Collins <rbtcollins at hp.com>
Distinguished Technologist
HP Converged Cloud