15:00:30 <n0ano> #startmeeting gantt
15:00:30 <openstack> Meeting started Tue Oct  7 15:00:30 2014 UTC and is due to finish in 60 minutes.  The chair is n0ano. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:31 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:00:33 <openstack> The meeting name has been set to 'gantt'
15:00:39 <n0ano> anyone here to talk about the scheduler?
15:01:03 <edleafe> o/
15:01:38 <PaulMurray> o/
15:01:59 <bauzas> \o
15:02:47 <n0ano> OK, let's get started
15:02:52 <n0ano> #topic forklift status
15:03:28 <n0ano> I reviews patch 110043 but I need to get on 119807, bauzas how are those coming?
15:04:00 <bauzas> n0ano: one sec, finding which changes you're talking :)
15:04:22 <n0ano> they're your patches, I thought they were burned in your brain :-)
15:04:38 <bauzas> n0ano: my changelist is becoming huge so I'm getting lost
15:04:48 <bauzas> so, indeed
15:04:54 <bauzas> https://review.openstack.org/110043
15:05:08 <bauzas> #link https://review.openstack.org/110043 setup_instance_groups is ready for reviews
15:05:15 <bauzas> jaypipes already gave a +2 to it
15:05:20 <bauzas> jaypipes: around ?
15:05:45 <bauzas> so that one is quite a no-brainer
15:06:06 <bauzas> I hope to get some support, and ideally folks are welcome for reviewingi t
15:06:13 <jaypipes> bauzas: yes, sorry, still working on blueprints...
15:06:25 <bauzas> jaypipes: no worries, was talking about https://review.openstack.org/110043
15:06:31 <jaypipes> see https://etherpad.openstack.org/p/kilo-nova-priorities for commentary
15:06:45 <n0ano> it's gotten positive reviews but I don't see a +2 on it
15:06:50 <bauzas> jaypipes: you already gave a +2, ideally it would be good if you could re-review it
15:07:06 <jaypipes> bauzas: yes, will re-review shortly.
15:07:12 <bauzas> n0ano: that's due to a change needed for caching the list of fitlers
15:07:26 <bauzas> now, the big piece
15:07:36 <n0ano> aah, a prior version got the +2, nevermind
15:07:36 <bauzas> https://review.openstack.org/119807
15:07:41 <jaypipes> fitlers == very aggressive, dominating filters.
15:07:46 <bauzas> :)
15:08:00 <bauzas> so https://review.openstack.org/119807 will be abandoned soon
15:08:36 <jaypipes> bauzas: ok, +2 from me on that one
15:08:54 <bauzas> so, I just abandoned 119807
15:09:01 <bauzas> jaypipes: cool thanls
15:09:22 <bauzas> now, the corresponding patch is being split into smaller chunks here
15:09:37 <bauzas> https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bug/1357491,n,z
15:10:00 <bauzas> ideally, we need to create a blueprint for this bug
15:10:08 <bauzas> I don't think we need a spec
15:10:21 <n0ano> bauzas, so it'll basically be the same changes just broken out into smaller pieces, right?
15:10:24 <bauzas> the last bits are missing from the series
15:10:28 <bauzas> n0ano: right
15:10:38 <bauzas> n0ano: it still misses 2 or 3 last patches
15:10:53 <bauzas> that's my on-going work
15:11:29 <bauzas> so reviews are welcome too
15:11:55 <jaypipes> bauzas: will review later this afternoon, for sure.
15:11:58 <n0ano> given that you have these patches tied to a bug we shouldn't need a separate BP for this work
15:12:55 <bauzas> n0ano: well, that's a tech debt bug, so a blueprint is kinda accepted
15:13:25 <n0ano> seems a little silly since the BP will just be a rewrite of what's in the bug but if that's the mechanism required
15:13:47 <bauzas> n0ano: that's the thing, a bugfix is typically smaller than that
15:14:12 <bauzas> n0ano: tracking patch series is FWIW better with blueprints
15:14:32 <bauzas> anyway
15:14:39 <bauzas> we also have kind of a no-brainer too
15:15:06 <bauzas> https://review.openstack.org/117042 ComputeNode creation at init stage
15:15:18 <bauzas> that one is currently marked as WIP because of a merge issue
15:15:26 <jaypipes> bauzas: that ^^ is another "bug" I filed :)
15:15:33 <bauzas> I'll try to find some time to fix it
15:15:51 <bauzas> anyway all the bits are coming in
15:16:05 <bauzas> and another one from PaulMurray should also help us
15:16:08 <bauzas> PaulMurray ?
15:16:24 <PaulMurray> bauzas, which one are you thinking of
15:16:36 <bauzas> the supported_instances thing
15:16:44 <PaulMurray> bauzas, nearly there
15:16:52 <bauzas> that's the last piece of work for using the ComputeNode object
15:17:11 <PaulMurray> bauzas, there's a bit of pci I think?
15:17:39 <bauzas> PaulMurray: really ? I was thinking that PCI was ready for this
15:17:59 <PaulMurray> bauzas, let me find the patches, hold on
15:18:07 <bauzas> #link https://review.openstack.org/125091 Add supported_instances to ComputeNode
15:18:22 <bauzas> that's the one worked by PaulMurray $
15:18:25 <bauzas> ^
15:18:37 <PaulMurray> https://review.openstack.org/#/c/110739/
15:18:58 <PaulMurray> https://review.openstack.org/#/c/76053
15:19:27 <PaulMurray> I haven't been in touch - this is last I think
15:19:58 <n0ano> I can ping Yunhong about those, he's on holiday this week
15:20:07 <bauzas> n0ano: sure would be nice
15:20:08 <PaulMurray> n0ano, please do
15:20:14 <bauzas> n0ano: we need to cover those
15:20:39 <bauzas> man, the ComputeNode object will become horrible
15:20:47 <n0ano> they weren't reject per se, just missed the deadline so we can do them in Kilo
15:20:57 <bauzas> because of the backwards compatibility and all these version bumps
15:21:07 <bauzas> n0ano: +1
15:21:28 <PaulMurray> bauzas, never mind the version bumps - it doesn't really matter
15:21:33 <n0ano> the real world (things like backwards compat) forces a lot of warts, just have to deal with it
15:21:49 <bauzas> PaulMurray: well, think about the lovely rebases you'll have to gezt
15:22:02 <bauzas> PaulMurray: I'm having 4 version bumps in my patch series :)
15:22:17 <bauzas> anyway
15:22:38 <n0ano> so, bottom line, I'm seeing about a half dozen patches we need to get done before we can even consider the split
15:22:53 <bauzas> we can also think about providing a middleground support for objects
15:23:07 <bauzas> n0ano: yeah, that ComputeNode object is kinda messy now
15:23:22 <bauzas> but we can also figure out what bits are missing and workaround them
15:23:40 <bauzas> ie. send to the sched an object plus extra dicts
15:23:49 <n0ano> I'm more concerned about the interfaces to the ComputeNode than the object itself
15:23:50 <bauzas> s/dicts/dict or kwarfs
15:23:57 <bauzas> kwargs
15:24:18 <bauzas> n0ano: the interfaces are there already
15:24:36 <bauzas> n0ano: ie. select_destinations() and update_resource_stats()
15:24:44 <n0ano> as long as the interfaces don't change because we've changed the object, that's my concern
15:25:11 <bauzas> n0ano: here you will pass objects instead of dicts but you won't change the interface
15:25:45 <n0ano> bauzas, I'm hoping you're right (I just worry a lot)
15:25:46 <bauzas> I mean, request_spec and filter_properties will become objectified in order to be sent thru select_destinations
15:26:10 <bauzas> and stats will be also objectified in order to be sent thru update_resource_stats
15:26:37 <bauzas> n0ano: see jaypipes's long monologue in the projects priorities etherpad :)
15:26:48 <jaypipes> heh
15:27:01 <bauzas> https://etherpad.openstack.org/p/kilo-nova-priorities
15:27:19 <bauzas> jaypipes: seriously, you beat me up on the length
15:27:20 <bauzas> :)
15:27:36 <jaypipes> bauzas: oh, it will be quite a bit longer shortly.
15:27:46 <bauzas> awesome readability
15:27:47 <jaypipes> bauzas: was working on dropping a bomb on the TC: https://review.openstack.org/#/c/126582
15:28:01 <jaypipes> bauzas: now back to blueprints...
15:28:04 <bauzas> can we perhaps move on and discuss on the Summit talks ?
15:28:18 <n0ano> bauzas, took the words out of my mouth
15:28:24 <n0ano> #topic kilo summit
15:28:47 <n0ano> I saw ttx's proposed schedule and note that gantt doesn't have any specific sessions
15:29:19 <n0ano> do we have to lobby for that or is the assumption that gantt will be discussed in one of the nova sessions?
15:29:26 <bauzas> n0ano: we're not officially an official program
15:29:27 <ttx> n0ano: it's proposed in a cross-project workshop, and as an "other project"
15:29:44 <jaypipes> n0ano: it should go in the cross-project sessions on monday and tuesday, IMO
15:29:50 <ttx> I would place my bet on "cross-project workshop"
15:30:01 <jaypipes> or possibly in the Other Projects track, though I'd prefer not to do that
15:30:02 <ttx> jaypipes: design sumit starts on Tuesday fwiw :)
15:30:12 <jaypipes> ttx: oh, sorry, duh
15:30:36 <n0ano> I was hoping for a cross-project session on tues and a gantt specific slot on wed or thurs
15:30:42 <bauzas> ttx: yeah I mass-filed that one by crossposting to both etherpads
15:31:36 <bauzas> n0ano: we should get a slot by the Nova team I expect
15:32:01 <bauzas> n0ano: this, plus a cross-project session or an other-projects sessions, or both
15:32:20 <n0ano> well, I think the critical one is the cross-project, I'd like to get concensus from other projects on what they want from a scheduler
15:32:29 <bauzas> n0ano: plus I was thinking of doing kinda BoF thing like the NFV team did in ATL
15:32:39 <bauzas> n0ano: +1
15:32:49 <bauzas> n0ano: hence the bet from ttx
15:33:31 <bauzas> n0ano: if we get enough people in the cross-project session, we could raise voice for a BoF event
15:33:38 <bauzas> about Gantt and others
15:33:58 <bauzas> jaypipes: your thoughts on that ?
15:34:13 <bauzas> as a TC member, take your hat
15:34:18 <n0ano> If we don't get a gantt specific on wed/thurs then a BoF would work, I don't think we need both
15:34:37 <bauzas> n0ano: wed/thur are project-specific
15:34:48 <n0ano> bauzas, exactly
15:34:59 <bauzas> n0ano: so the discussion will be around how we do the split
15:35:15 <bauzas> n0ano: not about how we integrate with other projects
15:35:21 <n0ano> we need 2 session, one on cross project issues and one on gantt specific (mainly the split)
15:35:43 <bauzas> I'm personally convinced that we will feed a 3rd session needed :)
15:35:53 <bauzas> feel (dammit!)à
15:36:05 <n0ano> bauzas, what would the 3rd session focus on?
15:36:22 <jaypipes> I'm interested in the refactoring scheduler interfaces sessions
15:36:42 <jaypipes> but not so much on the gantt-specific session -- just because I know exactly what it will devolve into.
15:36:59 <bauzas> jaypipes: k thanks :)
15:37:30 <jaypipes> it will just be a whine-fest about how much NFV and other teams want to tinker with the scheduler to place VMs closer to storage or network devices.
15:37:32 <bauzas> n0ano: I dunno I just remember the 40-min session in ATL and how time passed
15:37:47 <jaypipes> and frankly, I already know what folks want to do with a split-out-scheduler
15:38:07 <bauzas> jaypipes: eh, we'll be in France, so it could be a wine fest
15:38:12 <jaypipes> and I don't mind those things, but think the refactoring is more important.
15:38:18 <n0ano> jaypipes, my worry is the details, know what you want to do is different from how to achieve it
15:39:22 <bauzas> anyway, time is moving fast, maybe we can talk about this next week ?
15:39:35 <n0ano> my real worry is NFV & Neutron, I think I understand what Cinder/Containers/Ironic need, I'm worried that Neutron will be wildly different
15:39:37 <bauzas> I would like to discuss about bps and the draft from jay
15:39:44 <jaypipes> I promise I will have all the refactoring BPs related to RT and scheduler done by then.
15:39:50 <jaypipes> promise.
15:40:01 <bauzas> jaypipes: I can help you on this
15:40:03 <n0ano> but yes, we can move on, let's just keep an eye on the summit scheduling
15:40:07 <bauzas> jaypipes: that's in my agenda
15:40:10 <PaulMurray> jaypipes, is there an early version?
15:40:17 <n0ano> #topic Kilo BPs
15:40:21 <jaypipes> PaulMurray: of the blueprints?
15:40:31 <PaulMurray> jaypipes, of anything
15:40:39 <PaulMurray> jaypipes, to do with RT
15:40:45 <bauzas> https://etherpad.openstack.org/p/kilo-nova-priorities is a good starting point
15:41:02 <jaypipes> PaulMurray: https://etherpad.openstack.org/p/kilo-nova-priorities
15:41:16 <PaulMurray> jaypipes, yes, saw that
15:42:27 <n0ano> jaypipes, so you are planning on creating a Kilo BP about the RT changes, not just an etherpad rant?
15:42:37 <jaypipes> n0ano: yes.
15:42:42 <jaypipes> n0ano: they are underway.
15:42:44 * n0ano means rant in the positive way
15:42:52 <n0ano> jaypipes, excellent
15:42:58 <jaypipes> n0ano: I just wanted to make sure the refactoring sessions get on the summit schedule
15:43:06 <n0ano> jaypipes, +1
15:43:37 <bauzas> jaypipes: agreed
15:44:56 <n0ano> so I'm seeing only 2 BPs for Kilo, RT refactoring and the split, those are the two biggies
15:45:13 <bauzas> n0ano: we need to be more precise
15:45:33 <bauzas> jaypipes: could we just share what you're about to send before submitting it ?
15:45:45 <bauzas> n0ano: here, we need to consider
15:45:59 <bauzas> A/ ComputeNode object tech debt effort
15:46:02 <jaypipes> n0ano: I have 5 blueprints for Kilo involving refactoring.
15:46:26 <bauzas> B/ request_spec and filter_properties objectification
15:46:32 <n0ano> jaypipes, all related to RT or about different areas
15:46:50 <bauzas> C/ stats objectification
15:46:51 <jaypipes> n0ano: about RT, instance group and scheduler interfaces
15:46:56 <bauzas> D/ use of Claims within RT
15:47:12 <bauzas> oops, s/RT/scheduler
15:47:24 <jaypipes> bauzas: yes, and I even have B/ split into 2 blueprints...
15:47:36 <bauzas> jaypipes: I really like your proposal of sending a claim to the compute with a RPC abort method
15:47:52 <n0ano> what are the realistic odds that we can get ~10 different BPs approved and integrated in the Kilo timeframe
15:48:38 <bauzas> n0ano: I turn the question into what are the realistic odds if we don't do that ? :(
15:49:00 <jaypipes> n0ano: IMO, it's more realistic to get 10 slimly-defined BPs approved than 3 widely-defined BPs approved.
15:49:07 <jaypipes> n0ano: which is why I'm taking this tack.
15:49:10 <bauzas> jaypipes: +1
15:49:30 <n0ano> we'll see (see above, I worry a lot)
15:49:54 <bauzas> n0ano: whatever the effort could be, we'll still have to figure out how to get core support
15:50:14 <bauzas> and that's not really related to our coding bandwidth
15:50:19 <jaypipes> n0ano: worrywart.
15:50:39 <n0ano> jaypipes, somebody's has to be the Nanny :-)
15:50:50 <jaypipes> bauzas: ftr, you will have ndipanov_gone and myself as core support for all of these refactoring BPs...
15:51:06 <jaypipes> bauzas: and likely danpd's support on many of them.
15:51:25 <bauzas> jaypipes: good to hear
15:51:29 <n0ano> that will be good
15:52:02 <n0ano> OK, I think we know what we're doing here
15:52:08 <n0ano> #topic opens
15:52:15 <bauzas> jaypipes: I'm again emphasizing my wishes for seeing the bps before they're proposed, if possible :)
15:52:47 <edleafe> (Re-)Introducing myself: I worked on scheduler years ago, and am looking to get back into things with gantt.
15:52:54 <edleafe> I'm looking for some guidance as to where I can start contributing now.
15:53:01 <bauzas> jaypipes: because I'll probably be the worker ant behind that
15:53:16 <jaypipes> bauzas: I would like to be a worker ant as well :)
15:53:26 <n0ano> edleafe, you can start by reviewing the patches we talked about early today, that would be a good way to get your feet wet
15:53:36 <jaypipes> bauzas: but yes, will put each on etherpad for you to comment/review/change before proposing.
15:53:44 <edleafe> n0ano: will do
15:53:45 <bauzas> jaypipes: cool
15:53:55 <bauzas> edleafe: yeah, take time in reviewing all the scheduler changes
15:54:10 <bauzas> edleafe: try to get knowledge of the non-verbal things
15:54:19 <n0ano> edleafe, as we create the BPs we talked about some tasks will come up that you can take
15:54:35 <edleafe> bauzas: any existing reference material?
15:54:40 <bauzas> edleafe: Gerrit :)
15:54:46 <edleafe> bauzas: :)
15:54:54 <edleafe> Just didn't want to re-invent any wheels
15:54:58 <n0ano> edleafe, documentation - you silly person :-)
15:55:09 <edleafe> Yeah - what a concept
15:55:14 <edleafe> ;-)
15:55:27 <bauzas> edleafe: you having time ?
15:55:31 <n0ano> google openstack+scheduler and you'll find some stuff
15:55:40 <bauzas> n0ano: man, this info is pretty old
15:55:49 <edleafe> bauzas: Yes - I'm now tasked full time on OpenStack dev work
15:56:02 <bauzas> n0ano: I googled once in a time, and all the internal docs are quite old (2012 or so)
15:56:05 <n0ano> bauzas, I was going to caution about that but yes, it's old but gives an idea
15:56:07 <edleafe> And scheduler is where I worked most in the past
15:56:26 <bauzas> edleafe: well, atm, scheduler is not the real problem
15:56:32 <edleafe> I did the google approach, and yeah, things were old
15:56:42 <bauzas> edleafe: you probably understood that we need to work on cleaning up the interfaces
15:56:48 <edleafe> bauzas: exactly
15:57:05 <bauzas> edleafe: so, not really coding on the scheduler itself, but more likely on what we send to it
15:57:18 <n0ano> and those I/Fs are rather poorly documented unfortunately
15:57:33 <bauzas> edleafe: TBH, the FilterScheduler is kinda robust, so we won't reinvent the wheel
15:57:47 <edleafe> bauzas: how to separate it, and how to define the interfaces - that's what I'm looking at
15:57:52 <bauzas> n0ano: nope, the best way is to pdb them :)
15:58:17 <n0ano> bauzas, ignoring proposals like holistic scheduling which kind wants to replace the guts of the scheduler
15:58:27 <n0ano> s/kind/kind of
15:58:45 <n0ano> anyway, we're approaching the top of the hour and I have to run
15:58:46 <bauzas> edleafe: https://review.openstack.org/117042
15:58:52 <edleafe> n0ano: With a clean interface, they could do that
15:59:06 <bauzas> edleafe: you can maybe begin on that one
15:59:19 <edleafe> bauzas: sure. I
15:59:22 <edleafe> ugh
15:59:23 <bauzas> edleafe: all the stuff is already there, I'm just having an issue
15:59:26 <n0ano> edleafe, you can do it right now, the filter scheduler is pluggable, but do you really want to
15:59:30 <edleafe> I'll probably ping you with questions
15:59:35 <bauzas> edleafe: sure of course
15:59:42 <n0ano> edleafe, feel free, always willing to help
15:59:44 <edleafe> bauzas: just let me know if I get annoying ;-)
15:59:49 <bauzas> eh
16:00:00 <n0ano> tnx everyone, talk to you next week
16:00:03 <n0ano> #endmeeting