Here's resource provider and placement update 18-05. 18-04 was skipped
on account of illness.

# Most Important

Feature freeze has come and gone, RC1 is next week. This means that
finding bugs and, where relevant, reporting them with a tag of
'queens-rc-potential' is top priority.

The PTG is coming up at the end of this month. If you have topics for
discussion that are not already on the etherpad add them:

I wrote a blog post to gather some thinking (and links) about
preparing to extract placement from nova (or at least ease the path
when it does eventually happen):

It's probably time to start writing specs for some of the things we
know will be a big deal with placement in Rocky. Eric has started with
a spec that covers the ProviderTree work. Much of that work is already
done, but never had a spec in the first place:

I'm on the hook to create a spec for enabling generation handling when
associating aggregates. If there are others, getting them started
before the PTG can help to make the time at the PTG more effective.

# What's Changed

A limit is now passed to /allocation_candidates to ensure that we
don't cause out of memory errors in big empty clouds.

Traits expressed as 'required' in flavor extra specs are passed in
requests to placement and /allocation_candidates accepts the the
required parameter.

More, but not yet all, requests from nova to placement include the
global request id.

Some, but not all, of the ProviderTree functionality has merged.

The full stack of Alternate Hosts is now merged.

The ironic driver now manages traits.

At least some support for VGPU merged. Not clear what this means for
end users.

# Help Wanted

Testing, Testing, Testing.

There are a fair few unstarted bugs related to placement that could do
with some attention. Here's a handy URL:

# Main Themes

We've not yet identified the new themes, other than to know that
Nested remains a big deal.

## Nested Resource Providers

The work to get nested providers represented in the
/allocation_candidates did not complete before feature freeze. It
remains in progresss at

There's been a lot of discussion in IRC about the sometimes differing
goals on how people want NRP to work. One example is at:

There's an email thread related to that discussion:

I think we'll be doing ourselves a favor if we can work to satisfy concrete
use cases and then generalize from that.

The related provider tree work is now under its own topic:

# Other

Plenty of these are bugs or fairly trivial and/or non-feature fixes.

* doc: mark the max microversions for queens

* [Placement] Invalid query parameter could lead to HTTP 500

* [placement] use simple FaultWrapper

* Ensure resource classes correctly

* Avoid inventory DELETE API (no conflict detection)

* Do not normalize allocation ratios

* Sending global request ids from nova to placement

* VGPU suppport

* Update resources once in update available resources
    (This ought, when it works, to help address some performance
    concerns with nova making too many requests to placement)

* spec: treat devices as generic resources
    This is a WIP and will need to move to Rocky

* Support aggregate affinity filters/weighers
    A rocky targeted improvement to affinity handling

* Move placement body samples in docs to own dir

* Improved functional test coverage for placement

* Functional tests for traits api

* annotate loadapp() (for placement wsgi app) as public

* Remove microversion fallback code from report client

* WIP: SchedulerReportClient.set_aggregates_for_provider
    This is likely for rocky as it depends on changing the api for
    aggregates handling on the placement side to accept and provide
    a generation

* Add functional test for two-cell scheduler behaviors
    (This is old and maybe out of date, but something we might like to

* Make API history doc consistent

* WIP: General policy sample file for placement

* Support relay RP for allocation candidates
   Bug fix for sharing with multiple providers

* Convert driver supported capabilities to compute node provider

# End

Usual caveats about missing things apply. One thing I'm curious to
know is: From an end user's perspective what does the queen's
placement work get ya?

Chris Dent                      (⊙_⊙')
freenode: cdent                                         tw: @anticdent
OpenStack Development Mailing List (not for usage questions)

Reply via email to