[openstack-dev] [nova] [placement] placement update 18-26

Matt Riedemann mriedemos at gmail.com
Fri Jun 29 16:25:09 UTC 2018


On 6/29/2018 8:03 AM, Chris Dent wrote:
> # Questions
> 
> As far as I can tell there was no discussion on last week's
> question, so here it is again:
> 
> In [IRC
> [last 
> week]](http://eavesdrop.openstack.org/irclogs/%23openstack-placement/%23openstack-placement.2018-06-21.log.html#t2018-06-21T13:21:14) 
> 
> we had an extensive discussion about being able to set custom
> resource classes on the resource provider representing a
> compute node, outside the virt driver. At the moment the virt driver
> will clobber it. Is this what we always want?
> 

We've always said the virt driver is the "owner" of resource classes for 
the compute node provider right? If something external wants to put 
custom inventory in that tree, they'd do so with a child provider (like 
neutron will do with bandwidth providers).

We have said that we should merge externally-defined traits with 
compute-defined traits, and I think that is OK.

> 
> * "How to deploy / model shared disk. Seems fairly straight-forward,
>    and we could even maybe create a multi-node ceph job that does
>    this - wouldn't that be awesome?!?!", says an enthusiastic Matt
>    Riedemann.
> 

Another thing with this is move operations don't really work with shared 
providers yet, there are TODOs in the conductor task code for when we 
move the allocations from the instance on the source host to the 
migration record - those don't deal with shared providers.

> 
> ## Nested providers in allocation candidates
> 
> As far as I can tell the main thing left here is to turn it on in a
> microversion. That code is at:
> 
> * <https://review.openstack.org/#/c/565487/>

Merged.

> 
> ## Consumer Generations
> 
> There are new patches in progress on this, related to the bugs that
> were discovered:
> 
> * <https://review.openstack.org/#/q/topic:bug/1778576>
> * <https://bugs.launchpad.net/nova/+bug/1778591>
> 
> There are a patches left on the consumer generation topic, some tidy
> ups, and some stuff related to healing allocations:
> 
> * <https://review.openstack.org/#/q/topic:bp/add-consumer-generation>
> 
> Is someone already working on code for making use of this in the
> resource tracker?
> 

In what way? The RT, except for I think the Ironic driver, shouldn't be 
dealing with allocations (PUTing them anyway).

> 
> * <https://review.openstack.org/#/q/topic:bp/placement-osc-plugin-rocky>
>    A huge pile of improvements to osc-placement
> 

Several of these are making progress now (getting review I mean), but as 
of <1 hour ago I need to redo something in the bottom change in the series.

> 
> * <https://review.openstack.org/#/c/561770/>
>    PCPU resource class
> 

I dropped the procedural -2 on this since the spec was never approved in 
time for Rocky.

> 
> * 
> <https://review.openstack.org/#/q/project:openstack/blazar+topic:bp/placement-api> 
> 
>    Blazar's work on using placement

Cool I was just looking at Blazar the other night for interest in the 
dedicated hosts feature request from the public cloud SIG and was 
wondering if they'd started integrating more with placement rather than 
the compute APIs (and notifications). Seems like a good long-term 
strategy on their part.

-- 

Thanks,

Matt



More information about the OpenStack-dev mailing list