[openstack-dev] [nova] Reflections on the pike-1 milestone

Matt Riedemann mriedemos at gmail.com
Wed Apr 19 21:42:14 UTC 2017


Hey everyone,

Now that the pike-1 milestone is behind us I wanted to have a recap of 
the milestone to compare what progress we made against goals we set at 
the PTG, and to look forward to the pike-2 milestone.

First some highlights of things accomplished in the pike-1 milestone in 
no particular order:

- Jay Pipes got the Ironic virt driver reporting custom resource classes 
into the Placement service for compute node inventory.
- There is good progress on the os-traits library and Alex Xu got the 
/traits API merged into the placement endpoint.
- Sean Dague got high-level agreement on unifying limits in Keystone 
which is a foundation for supporting hierarchical quotas.
- We merged the spec and plan for integrating Searchlight into nova-api. 
At this point that's all just spec, but it was a pretty complicated spec 
to work through and we have a plan going into pike-2.
- Sean Dague got uwsgi working in devstack now and Chris Dent is working 
on making nova-api run under uwsgi per the Pike community goal.
- Dan Smith has made good progress on enabling multi-cell support in the 
REST API and getting devstack to run and pass tests with a fleet of 
conductors. We'll be discussing this at the Forum [1].
- We merged Ildiko Vancsa's patch to remove the check_attach code from 
Nova, and we merged John Garbutt's spec for integrating the new Cinder 
attachment APIs into Nova. Progress has been made on the code for using 
the new APIs too.
- Chris Dent has been sending weekly emails giving updates on the work 
going on with placement, and Balazs Gibizer has been doing similar for 
the versioned notifications work. This has been helpful for keeping 
focus, recording decisions, and giving those outside the day-to-day 
involvement an idea of the progress made and where they can help.
- Good progress from the OSIC team on documenting the various policy 
rules [2].
- We have 62 blueprints/specs approved, 3 completed, and several with 
code up for review.

Some targets we missed in pike-1:

- We aren't as far along as we'd like to be with the counting quotas 
work, but to be fair, some of that was redone after initial review to 
make it easier to integrate. And we did approve the spec for putting a 
/usages API into placement which the quotas work will leverage.
- We don't have the additional-notification-fields-for-searchlight 
blueprint done yet. We hit some snags during review but those have been 
ironed out now, so we should be able to finish this early in pike-2.
- We never had a spec for using Cinder as an ephemeral backend. However, 
we will be discussing this at the Forum [3] so hopefully we'll have a 
plan going into Queens.
- The versioned notifications transformation has been slowing down, 
probably due to a lack of reviews.
- I never delivered a spec for deprecating personality files from the 
compute REST API (but I'm deprecating some other things from the API, so 
that counts, right?).
- We didn't merge a spec to support the concept of service-locked 
instances. There is a draft work in progress spec though to pick up in 
Queens [4].
- Little to no progress on merging the network-aware scheduling series 
which has been carried over since Newton. This is needed to support 
Neutron routed networks.
- The PowerVM driver series has not landed a single change yet due to 
lack of reviews.

Looking to pike-2 we have a few priority things to get done:

- Get a dsvm job running with nova + searchlight and start writing the 
proof of concept for searchlight integration with nova-api. The goal 
here is going to be finding out what issues we didn't anticipate in the 
spec, even though there were plenty of issues already identified in the 
spec. We will also be discussing this at the forum [5].
- Complete the additional-notification-fields-for-searchlight blueprint.
- We need to make progress on landing the counting quotas changes early 
so we can shake out any bugs introduced by that complicated change.
- Close on the plan for moving claims to the scheduler, discuss it with 
operators at the Forum [6], and make good progress on implementation by 
the end of the milestone.
- Get more of the versioned notifications work done.
- Now that the /traits API is available, we need to make progress on 
adding support for modeling shared storage pools in Placement.
- Have a multi-cell CI job running which tests the conductor fleet 
deployment model and API, including move (migrate) operations within a cell.
- Continue adding support for the new Cinder attachment APIs. We should 
have the code in place to create new-style attachments by the end of 
pike-2, and testing it with the grenade upgrade CI job. This is needed 
for supporting volume multi-attach.
- Get some of the PowerVM driver patches landed, at least through 
spawn/destroy, but ideally to the point of supporting a console.

Current focus:

- We have the summit coming up in less than three weeks. People are 
working on presentations and planning for the Forum sessions.
- With the recent loss of the OSIC developer resources, we're going to 
need to evaluate which efforts were owned by the OSIC team and figure 
out who can take over those blueprints. I'll be working on this and 
sending something to the mailing list to ask for volunteers.

Overall I think we made good progress in pike-1 and have the stage set 
for big changes to land in pike-2 if we can stay focused.

[1] 
https://www.openstack.org/summit/boston-2017/summit-schedule/events/18727/cellsv2-operatordevelopercommunity-coordination
[2] https://docs.openstack.org/developer/nova/sample_policy.html
[3] 
https://www.openstack.org/summit/boston-2017/summit-schedule/events/18738/using-cinder-for-nova-ephemeral-storage-backend
[4] https://review.openstack.org/#/c/438134/
[5] 
https://www.openstack.org/summit/boston-2017/summit-schedule/events/18739/using-searchlight-to-list-instances-across-cells-in-nova-api
[6] 
https://www.openstack.org/summit/boston-2017/summit-schedule/events/18723/moving-resource-claims-from-nova-compute-to-nova-scheduler

-- 

Thanks,

Matt



More information about the OpenStack-dev mailing list