On 07/25/2013 08:30 PM, Joshua Harlow wrote:
When you have so much state to maintain then aren't the APIs incorrect??
Yes, the EC2 APIs are incorrect in being silly and using ints for ids
for so many things, also for supporting people to make GET requests with
16k get strings. But there isn't much we can do about that. :)
Or can there be new API's that expose this translation, something
seems/feels wrong if there is so much state to maintain u can't do a
translation layer.
Most of this is about id allocation and translation. OpenStack uses
UUIDs, AWS uses ints. UUIDs is a better design point, and means you
don't need to have a global auto allocator which you can guaruntee,
which is good.
Also there are EC2 design points that have request lengths greater than
what Apache (or any other web front end) is compiled to support, as they
have the possibility of enourmous GET strings (16K at least). Again,
instead of sensibly requiring to move to POST in those cases. I know we
had to land a change for CERN to allow bigger requests on EC2 calls for
just this reason (we did keep the get length apache sized on OSAPI, so
we didn't break people's attempts to get this behind a real web server).
Translation is never exact, go talk to the WINE folks about that one.
I'm personally fine either way, proxy or embedded in openstack. Which
approach isn't really the issue. It's that no one is doing the work.
Actions speak much louder than words (well... except in pundit echo
chambers), so I'd much rather have people with strong opinions on this
express how strongly those are by having a big patch queue for me to review.
-Sean
My 2 cents.
On 7/25/13 4:36 PM, "Michael Still" <mi...@stillhq.com> wrote:
On Fri, Jul 26, 2013 at 8:30 AM, Russell Bryant <rbry...@redhat.com>
wrote:
If an external proxy (like AWSOME) is what you want, one of those
already exists (at least for the EC2 API).
http://deltacloud.apache.org/
It supports EC2 on the frontend and the OpenStack compute API on the
backend. I'm not sure how using this compares to the EC2 implementation
in nova, though.
I am sceptical of the external proxy approach as there is a lot of
state to maintain (uuid to id mappings for example) which is hard to
do right in a proxy. I like the idea of the AWS APIs being secondary
APIs within nova. However, its fair to say that there hasn't been much
work done on them recently.
Michael
--
Rackspace Australia
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
--
Sean Dague
http://dague.net
_______________________________________________
OpenStack-dev mailing list
OpenStack-dev@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
