15:00:32 <bauzas> #startmeeting nova_scheduler
15:00:33 <openstack> Meeting started Tue Apr 21 15:00:32 2015 UTC and is due to finish in 60 minutes.  The chair is bauzas. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:34 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:00:37 <openstack> The meeting name has been set to 'nova_scheduler'
15:00:54 <bauzas> folks, who's here to talk about the sched efforts ?
15:01:00 <edleafe> o/
15:01:14 <bauzas> n0ano can't make the call today, so I'm chairing it
15:01:17 <Sukhdev> amotoki: don't know what happened - was typing bye - still sleepy I guess :-):-)
15:01:52 <bauzas> Sukhdev: FYI, we started a new meeting
15:01:57 <amotoki> Sukhdev: :-)  I think you just typed "tab" or something too first..
15:02:06 <amotoki> sorry for botherig.
15:02:13 <Sukhdev> bauzas: sorry
15:03:48 <bauzas> edleafe: sounds like I will have to offer beers at Vancouver again
15:04:17 <bauzas> so, the next one raising hand is giving a free beer for Vancouver
15:04:33 <bauzas> given
15:05:20 <not_edleafe> o/
15:05:24 <egon> I'll raise my hand, but I was just here to watch.
15:06:36 <bauzas> edleafe: eh :)
15:06:50 <bauzas> edleafe: burned by IRC
15:06:51 <bauzas> (17:05:18) edleafe est désormais connu sous le nom de not_edleafe
15:07:09 <edleafe> oh, that was definitely not me!
15:07:14 <bauzas> ah ah
15:07:25 <bauzas> ok, let's start, it should be quick
15:07:34 <bauzas> #topic open discussion
15:07:36 <alex_xu> o/
15:08:01 <bauzas> so, no precise agenda today as I was afk for 2 days because I ran a meetup
15:08:30 <bauzas> that said, FYI, I just presented our efforts at the meetup and I gave good feedback :)
15:08:49 <edleafe> bauzas: good to hear!
15:08:50 <bauzas> most of our work is being silently supported, so thanks guys
15:09:31 <bauzas> anything you want to discuss ? Vancouver maybe ?
15:09:31 <egon> I was looking at scheduler options in the tree. It looks like we're down to a filter scheduler, and an ironic scheduler?
15:09:47 <bauzas> egon: nah, Ironic still uses the FilterScheduler
15:10:06 <bauzas> egon: Ironic just has a different HostManager for counting resources, that's it
15:10:15 <egon> bauzas: got it. thanks!
15:10:37 <bauzas> so, guys, maybe a few pointers
15:10:39 <egon> So, the trick if you want some new rule/behavior is to add someting in a filter.
15:10:49 <bauzas> https://etherpad.openstack.org/p/liberty-nova-summit-ideas
15:10:56 <bauzas> egon: yup
15:11:22 <bauzas> so, L83 and follows are the BPs I'm targeting for L
15:11:23 <edleafe> egon: also the caching scheduler
15:11:49 <bauzas> edleafe: and the ChanceScheduler too :)
15:12:26 <edleafe> heh
15:12:29 <bauzas> egon: http://sbauza.github.io/2015/04/20/#/
15:12:41 <bauzas> egon: you'll find some materials about how works the scheduler
15:13:01 <egon> bauzas: perfect
15:13:01 * bauzas didn't want to tease, but that's it
15:13:35 <bauzas> egon: you'll be missing the presenter's comments, but I guess it's self-explanatory
15:13:57 <bauzas> so, back to our Liberty planning
15:14:24 <bauzas> alex_xu: edleafe: feel free to amend some points in the etherpad if you want
15:14:35 <alex_xu> bauzas: ok
15:14:53 <bauzas> nothing really new but https://blueprints.launchpad.net/nova/+spec/shared-state-scheduler
15:15:10 <bauzas> I'm working on a POC before writing the spec
15:16:20 <alex_xu> cool, will read it
15:16:30 <bauzas> okay, I assume there is no real other question nor point
15:16:36 <bauzas> edleafe: ?
15:16:46 <edleafe> bauzas: in general, will this be a scheduler-owmed db backing the shared state?
15:16:57 <bauzas> edleafe: I would prefer call it a datastore
15:17:15 <bauzas> edleafe: but that would be driver-driven
15:17:36 <bauzas> edleafe: so yeah basically the POC would be using the regular SQLA wiring
15:17:56 <edleafe> bauzas: so that removes the need to constantly query the nova db?
15:18:08 <bauzas> edleafe: no, that's orthogonal
15:18:39 <egon> Reducing race conditions would be great.
15:18:59 <egon> We see double-scheduling of compute nodes when certain requests are made close together.
15:19:16 <bauzas> edleafe: I mean, populating the datastore would still be needed, but yeah it would just make sure that updating HostState would be something transactional
15:19:17 <edleafe> bauzas: ok, I'll wait for the spec then
15:19:50 <bauzas> edleafe: wait for the POC rather
15:21:19 <bauzas> okay, anything else ?
15:21:28 <edleafe> not from me
15:22:51 <bauzas> ok, crickets
15:23:03 <bauzas> I guess we can safely go back to our various concerns
15:23:05 <bauzas> ty all
15:23:07 <bauzas> #endmeeting