Open Stack

Hi all,

===
tl;dr: Now that we agree on waiting for the split prereqs to be done, we
debate on if ResourceTracker should be part of the scheduler code and
consequently Scheduler should expose ResourceTracker APIs so that Nova
wouldn't own compute nodes resources. I'm proposing to first come with
RT as Nova resource in Juno and move ResourceTracker in Scheduler for K,
so we at least merge some patches by Juno.
===

Some debates occured recently about the scheduler split, so I think it's
important to loop back with you all to see where we are and what are the
discussions.
Again, feel free to express your opinions, they are welcome.

Le 03/07/2014 19:53, Sylvain Bauza a écrit :
> Hi,
>
> ==
> tl; dr: A decision has been made to split out the scheduler to a
> separate project not on a feature parity basis with nova-scheduler, your
> comments are welcome.
> ==
> <snip/>

So, based on the feedback consensus, we agreed to wait for all the
preparation work to be made before doing the split.

That said, now that we're focusing back on the split prereqs, some good
talks are coming in about what the Scheduler should express as APIs and
what would be consequently owned by the scheduler (and later Gantt).

During the Summit, we expressed the boundary in between Scheduler and
Nova as the current Scheduler namespace in Nova and where the
ResourceTracker would take use of a Scheduler client for updating stats
to the Scheduler [1]. In this model, compute nodes would stay as a Nova
resource and Scheduler (and later Gantt) would have its own
representation of the resources it has to manage.

The implementation of this model is https://review.openstack.org/82778
still under review. Note also the spec has been validated 3 months ago
[2], so this was planned to delivered for Juno.


Based on a review we got on the implementation patch [3], there is now
an interesting discussion about if Nova should still keep its own
representation of the resources (and consequently should manage all
claims for resource usage). The idea of it would be to move the
ResourceTracker to the Scheduler namespace and consequently the
Scheduler should expose Tracking APIs for claiming/unclaiming usage, so
the Compute Manager (and no longer the ResourceTracker) would make use
of a Scheduler client.

A PoC for seeing the ResourceTracker proxified by a Scheduler client can
be found here https://review.openstack.org/105747
As we're now beyond the Nova Specs proposal deadline, that would mean
that any spec which would be created would see its implementation merged
by K at least.

Don't blame me if I'm trying to be pragmatic : at a first glance, I
appreciate the idea of having the Tracker moved to the Scheduler, but
I'm seeing this as another possible step for the split, not the first
one. So my proposal is to say :
 1. Move on the existing validated spec and review the existing
implementation https://review.openstack.org/82778
 2. Create a spec targeted for K to see what would be a world where the
tracking of resources is done by the Scheduler
 3. If that above spec gets validated, do the necessary work in K to
move the line
 3(bis). If that above spec gets refused, we are still on the good path
for splitting the scheduler


Thoughts ?

-Sylvain



[1] https://etherpad.openstack.org/p/juno-nova-gantt-apis
[2] https://review.openstack.org/82133
[3] https://review.openstack.org/#/c/82778/42

> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev