14:01:26 <jaypipes> #startmeeting scheduler
14:01:27 <openstack> Meeting started Mon May 14 14:01:26 2018 UTC and is due to finish in 60 minutes.  The chair is jaypipes. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:01:28 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:01:31 <openstack> The meeting name has been set to 'scheduler'
14:01:36 <efried> ō/
14:01:38 <takashin> o/
14:01:39 <alex_xu> o/
14:01:43 <tssurya> o/
14:01:59 <jaypipes> good morning/evening everyone
14:02:14 <edleafe> \o
14:02:52 <jaypipes> #topic specs needing review
14:03:05 <cdent> o/
14:03:32 <jaypipes> #link https://review.openstack.org/555081 "cpu resource tracking" could use a final once over
14:03:40 <sundar> o/
14:04:02 <melwitt> o/
14:04:06 <efried> jaypipes: I answered a q from tetsuro just now, pls verify that it gels with your intent.
14:04:14 <efried> (on that spec ^)
14:04:26 <jaypipes> #link https://review.openstack.org/#/c/565565/ "update add-consumer-generation to focus on API"
14:04:31 <jaypipes> efried: k, will do.
14:04:42 <gibi> o/
14:05:40 <jaypipes> besides those two, are there any other specs we are targeting for Rocky?
14:06:29 <jaypipes> bauzas: you have two specs, one on NUMA topology and one on multiple GPU types.
14:07:01 <jaypipes> bauzas: both have a number of requests in the review comments
14:07:29 <bauzas> jaypipes: yup, I know
14:07:44 <bauzas> jaypipes: for the multiple GPU types, I provided a change
14:08:01 <bauzas> jaypipes: maybe given it's simple, we should abandon the spec ?
14:08:18 <jaypipes> #link https://review.openstack.org/#/c/552924/ (NUMA topology with RPs)
14:08:29 <jaypipes> #link https://review.openstack.org/#/c/557065/ (Multiple GPU types)
14:09:23 <bauzas> looks like we have a consensus for https://review.openstack.org/#/c/557065/
14:09:45 <bauzas> https://review.openstack.org/#/c/564214/ is the implementation change
14:10:11 <jaypipes> bauzas: I don't see that consensus...
14:10:42 <bauzas> jaypipes: using the way of cinder
14:11:18 <bauzas> anyway, if that's needed, I'll provide a new spec version
14:12:02 <jaypipes> bauzas: sorry, I absolutely think a new spec version is needed. there are lots of unanswered comments and typos.
14:12:13 <bauzas> okay
14:13:58 <jaypipes> ok, are there any more blueprints/specs people want to discuss? sundar, how's your cyborg/nova stuff going?
14:14:48 <sundar> jaypipes: Most of us want to wait for nRP stuff, rather than start with a plan B and switch over.
14:14:58 <efried> ++
14:15:01 <sundar> SO, we are currently proceeding mostly on drivers etc.
14:15:27 <jaypipes> sundar: ok, so you are prototyping an update_provider_tree() implementation then?
14:16:41 <alex_xu> I remember sundar has os-acc spec up also, but that isn't scope of scheduler team.
14:16:52 <sundar> jaypipes: I don;t think much work is going on the Nova interaction, to tell you the truth. It is mostly drivers, devstack etc. Also, I would feel more comfortable if os-acc spec gets some reviews, because that is the other half of the Nova/Cyborg interaction
14:17:00 <sundar> :)
14:17:14 <alex_xu> sundar: ++
14:17:16 <jaypipes> sundar: link?
14:18:03 <sundar> #link https://review.openstack.org/#/c/566798/ os-acc spec, Cyborg/Nova/Glance interaction in compute node
14:18:24 <jaypipes> sundar: got it. will review later today. thanks.
14:18:52 <sundar> jaypipes: Thanks! Appreciate it. :)
14:19:11 <jaypipes> #topic open discussion
14:19:29 <jaypipes> alright folks, anything in particular people want to chat about, or should we get back to work?
14:19:34 <efried> Is tetsuro around?
14:19:39 <cdent> I wanted to remind people of two forum sessions that are placement related
14:19:49 <tetsuro> here @efried
14:19:49 <cdent> #link extraction: https://etherpad.openstack.org/p/YVR-placement-extraction
14:20:03 <cdent> #link cinder and placement: https://etherpad.openstack.org/p/YVR-cinder-placement
14:20:20 <cdent> please add to those etherpads so we have some substantial to talk about
14:21:02 <efried> tetsuro: I wanted us to sync up on the state of the nrp-in-alloc-cands work.  At this point it looks like there's two separate (and conflicting, git-wise) series in play.  Are you planning to rebase them into a single series?
14:21:08 * bauzas needs to go to the school
14:21:20 <jaypipes> cdent: ack, will try (though I won't be in YVR)
14:21:38 <tetsuro> yes, I reworked nested alloc cands work in a better way creating a new branch from https://review.openstack.org/#/c/567113/10
14:22:00 <cdent> we're going to use mystic powers to channel you jaypipes
14:22:42 <efried> tetsuro: Right, so are you going to rebase the other patches (we at least need the one that proposes the new microversion and actually exposes this from the API) on top of the new series?
14:22:48 <bauzas> tetsuro: will review your series ASAP
14:23:14 <tetsuro> efried and everyone: Right. I'm going to rebase some of old patches on top of it.
14:23:17 <bauzas> (at least once I'm done with children and slides writing for YVR)
14:23:24 <tetsuro> tomorrow.
14:23:27 * bauzas disappears
14:23:41 <efried> tetsuro: Cool, just making sure that was in the works.  Thanks.
14:23:43 <tetsuro> the first 3 patches are already ready for review from the bottle
14:24:10 <tetsuro> bottle/bottom
14:24:24 <efried> All: IMO this is the most important thing we've got going right now, so please put some review focus here if you can.
14:24:45 <efried> Everything relies on it.
14:25:40 <efried> that's it for me
14:25:54 <melwitt> will that series
14:26:02 <melwitt> "finish" NRP?
14:26:13 <efried> melwitt: Yes, insofar as that's possible.
14:26:21 <efried> melwitt: It's the last thing stopping us from using nrp from nova.
14:26:23 <melwitt> I've been wondering in general where/how that could be considered "complete"
14:26:27 <jaypipes> melwitt: NRP is like hotel california.
14:26:27 <melwitt> I see, thank you
14:26:34 <melwitt> jaypipes: lol
14:26:51 <efried> well, and the nova side of granular: https://review.openstack.org/#/c/515811/
14:27:06 <melwitt> ack
14:27:41 <efried> (...from nova, or neutron, or cyborg, or...)
14:28:02 <jaypipes> ok, any other topics?
14:28:55 <jaypipes> ok, thanks all. ending meeting.
14:28:57 <jaypipes> #endmeeting