Open Stack

Hi all,

Unfortunately I couldn't make yesterday's meeting but just following through the notes this item jumped out at me:

* Live Migration w/ Placement rule enforcement - Owner: Nicole Reineke - 2 weeks for first draft review - Target 1/6/15

I believe I commented in the pad from the session in Paris that it wasn't clear to me exactly what the goals are here, and specifically whether existing proposals (some of them implemented) have been taken into account. I'll try outline what exists today, and what's already proposed and perhaps someone more familiar with the discussion in the context of the WTE group can fill me in:

* https://blueprints.launchpad.net/nova/+spec/live-migration-scheduling
  - Implemented: Grizzly
  - Description: Makes target host parameter to live migrate optional, if it is not provided the scheduler is used to choose a host.
* https://blueprints.launchpad.net/nova/+spec/find-host-and-evacuate-instance
  - Implemented: Juno
  - Description: Makes target host parameter to evacuate optional, if it is not provided the scheduler is used to choose a host.

What's of course missing here is that scheduler hints are not persisted so the decisions made by the scheduler when re-placing the instance will not necessarily be ideal. The blueprint intended to address this is:

* https://blueprints.launchpad.net/nova/+spec/persist-scheduler-hints

With regards to instance/server groups as used for anti-affinity in particular this is somewhat "special" in that the original hint is not required as the server group information is recorded in the database. As a result of the fix to https://bugs.launchpad.net/nova/+bug/1379451 the group membership is reflected accurately when re-scheduling placement of the guest. This doesn't help significantly for affinity groups, where re-inforcement of the policy using the scheduler actually *prevents* migration/evacuation (correctly) as there is currently no mechanism to migrate the entire group in one hit.

Given the above the question from me is what of the intended so scope of the new proposal over and above these? In particular it would seem to me like working out how to accelerate the proposal regarding persisting scheduler hints would provide the most immediate bang for buck in this area?

Thanks,

Steve