[placement][ptg] pre-ptg checkpoint

Chris Dent cdent+os at anticdent.org
Thu Apr 18 12:11:20 UTC 2019


Since I'm going to be out of touch next week, I thought it best to
provide a checkpoint summarizing where things are with the virtual
pre-PTG process so that we go into the PTG in reasonable shape.

For reference, the pre-PTG has been a series of email threads
discussion topics raised in the two placement-related PTG
etherpads:

* https://etherpad.openstack.org/p/placement-ptg-train
* https://etherpad.openstack.org/p/ptg-train-xproj-nova-placement

Each thread is linked from its topic in the etherpad. Any thread
which reached some resolution is marked with a boldface "ACTION", often
with a link to a story or review associated with the next step.

Overall I think the process has been useful. We've figured out some
plans of action for resource provider partitioning and consumer types,
worked out better testing for osc-placement, and have some ideas on
better container manager and performance testing. These things are
reflected in worklists on storyboard [1, 2].

We've also managed to uncover some more ideas, challenges, and
benefits with nested resource providers.

We have a planned Saturday morning session at the PTG with Ironic,
Blazar and anyone else who is thinking about using placement outside
of nova.

Where we've made less progress is getting owners on the new features
[3] and reaching some resolution or plan with the several nested
provider related threads (aggregate spanning, subtree "affinity",
json body, NUMA in placement, request group mapping, resource trait
filters). We've shared some ideas and managed to dismiss some things
that won't work, but at least so far a solid plan is yet to emerge.

Part of the latter may be my fault. I'm compelled to fully
understand what's going on and for things to make as much sense to
as many people as possible and I believed we had the time and space
to have a wide-open discussion in that direction in the weeks prior
to the PTG. There's still next week, so perhaps that's still the
case, but maybe a different approach is required. If you've got
ideas, please share.

Please carry on with the threads if there's more to add. I'll
incorporate them early in the week before the PTG and extract some
kind of schedule for our time at the PTG. As I've said before there
may not be much opportunity, we've only got Friday and Saturday and
it is mostly consumed. Friday will mostly be taken up with Nova,
Saturday morning will be talking with Ironic and Blazar. By Saturday
afternoon we will probably be exhausted and I suspect the best use
of our time will be talking without agenda about making sure working
on Placement is pleasant and useful for everyone [4]. We don't need to
actively seek more work: We've already defined got a lot, especially
when we account for work in other projects.

Thanks.

[1] Cleanups: https://storyboard.openstack.org/#!/worklist/575
[2] Features: https://storyboard.openstack.org/#!/worklist/594

[3] However, thanks very much to Surya for stepping up to make a
spec for the consumer type feature:
https://storyboard.openstack.org/#!/story/2005473

[4] To that end, please make sure you have added your thoughts to
the retrospective etherpad: 
https://etherpad.openstack.org/p/placement-ptg-train-retrospective

-- 
Chris Dent                       ٩◔̯◔۶           https://anticdent.org/
freenode: cdent                                         tw: @anticdent


More information about the openstack-discuss mailing list