Open Stack

Hi Matt,

Thanks for looking into this,


On Wed, Dec 5, 2018 at 10:27 PM Matt Riedemann <mriedemos at gmail.com> wrote:

> Belmiro/Surya,
>
> I'm trying to follow up on something Belmiro mentioned at the summit
> before I forget about it.
>
> CERN sets this value low:
>
>
> https://docs.openstack.org/nova/latest/configuration/config.html#scheduler.max_placement_results
>
> And as a result, when disabling nova-computes during maintenance, you
> can fail during scheduling because placement only returns resource
> providers for disabled computes.
>
> I believe Dan and I kicked around some ideas on how we could deal with
> this, like either via a periodic in the compute service or when the
> compute service is disabled in the API, we would set the 'reserved'
> inventory value equal to the total to take those computes out of
> scheduling.


Just read the discussion on the channel and saw there were a couple of
approaches proposed like traits and neg-aggregates in addition to the above
two.


> I think Belmiro said this is what CERN is doing today as a
> workaround?
>
>
As far as I know we don't have it in PROD, I will let Belmiro confirm this
anyways



> For the latter solution, I don't know if we'd proxy that change directly
> from nova-api to placement, or make an RPC cast/call to nova-compute to
> do it, but that's an implementation detail.
>
> I mostly just want to make sure we get a bug reported for this so we
> don't lose track of it. Can one of you open a bug with your scenario and
> current workaround?
>
>
We have already filed a bug for this:
https://bugs.launchpad.net/nova/+bug/1805984. Will add the workaround we
have into the description.

------------
Regards,
Surya.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20181206/f3a9e154/attachment-0001.html>