Open Stack

Thanks Matt for all the help so far and also for taking the time to write up this summary! I would like to join in saying thanks to the handful of people listed below who formed an amazing cross-project team during the past two years.

I strongly believe that by hunting for multi-attach support we stepped on the right track in terms of the interactions between the API’s of Cinder and Nova and better readability and maintainability of the code base in both projects on the long run with the changes we’ve made so far.

I hope we can grow our team onwards as testing will become even more crucial as we move forward with the remaining implementation work to reach the minimum functionality for multi-attach.

Thank you All once more for the hard work! And see you on the weekly meeting next Thursday. :)

Thanks,
Ildikó
(IRC: ildikov)


> On 2017. Dec 9., at 13:15, Matt Riedemann <mriedemos at gmail.com> wrote:
> 
> I just wanted to take a minute to recognize that the patch [1] to make Nova use the new-style Cinder volume attach flow has merged.
> 
> As one can tell from the patch set count alone, this has been in the works for a long time now.
> 
> We were talking about volume multi-attach support back in Mitaka at the midcycle in Bristol. The early talks / patches would have piled more technical debt onto Nova and further baked in Nova's need to manage volume state, which is something we wanted to avoid, so at the Newton summit in Austin we changed course and decided to prioritize working on a new data model and API in Cinder, which eventually came out as the 3.27 volume attachments API in Ocata.
> 
> There was then serious work on both sides in Pike to start using the new 3.27 API and we found out we needed some more changes to Cinder, which became the 3.44 Cinder API in Queens. Now Nova has merged the change at the end of a very long series of changes to enable the new flow once 3.44 is available to Nova and computes are all upgraded to understand the new flow. One can appreciate the complexity here if you read through the Nova spec for the new attach flow [2].
> 
> I was really happy to see [1] merge this week because I knew we needed to get that in by the queens-2 milestone if we were going to have a shot at (1) flushing out stability issues before Queens RC1 and (2) a good chance to get multi-attach support into Nova in Queens.
> 
> We have a plan for multi-attach support [3] which I think is doable before feature freeze. The Cinder 3.48 API is now available too which Nova needs to correctly detach a multi-attach volume.
> 
> I want to thank everyone that's helped push this along for their dedication and patience, especially John Griffith, Ildiko Vancsa, Steve Noyes and John Garbutt. And thanks to Sean McGinnis, Jay Bryant, Walter Boring and Balazs Gibizer for review support.
> 
> We've had weekly meetings between the Nova and Cinder teams for at least two years now and I can finally see the end so let's keep up the momentum.
> 
> [1] https://review.openstack.org/#/c/330285/
> [2] https://specs.openstack.org/openstack/nova-specs/specs/queens/approved/cinder-new-attach-apis.html
> [3] https://specs.openstack.org/openstack/nova-specs/specs/queens/approved/cinder-volume-multi-attach.html
> 
> -- 
> 
> Thanks,
> 
> Matt
> 
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev