15:00:10 <n0ano> #startmeeting gantt
15:00:11 <openstack> Meeting started Tue Apr 28 15:00:10 2015 UTC and is due to finish in 60 minutes.  The chair is n0ano. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:00:15 <openstack> The meeting name has been set to 'gantt'
15:00:18 <n0ano> Anyone here to talk about the scheduler?
15:00:33 <alex_xu> o/
15:00:48 <PaulMurray> o/
15:02:05 <n0ano> let's get started (others may appear later)...
15:02:07 <edleafe> o/
15:02:11 * bauzas waves late (again)
15:02:19 <n0ano> #topic Liberty specs
15:02:48 <n0ano> you may have noticed I created a tracking wiki for our Liberty specs - https://wiki.openstack.org/wiki/Gantt/liberty
15:03:14 <n0ano> if everyone could review it an make sure I've included everything that would be good.
15:03:53 <n0ano> to talk specifics for a minute, PaulMurray I believe your spec has already been re-approved for Liberty - right?
15:04:08 <bauzas> n0ano: I began to draw something about that
15:04:16 <PaulMurray> n0ano, right
15:04:32 <bauzas> n0ano: https://etherpad.openstack.org/p/liberty-nova-summit-ideas L89
15:04:42 <bauzas> n0ano: you can possibly take those
15:04:52 <bauzas> n0ano: and I'm writing one last BP
15:04:59 <n0ano> PaulMurray, cool (I got that right :-), especically since it's half implemented that should go through
15:05:01 <bauzas> n0ano: which will require a spec
15:05:25 <PaulMurray> bauzas, johnthetubaguy told me it did not need a spec
15:05:43 <PaulMurray> bauzas, so he approved the bp
15:05:51 <bauzas> PaulMurray: I was not referring to your BP :)
15:06:01 <PaulMurray> oh - what are you referring to ?
15:06:05 <bauzas> sec
15:06:47 <johnthetubaguy> PaulMurray: oh, maybe I still need to approve it for liberty? I may have confused myself on that one
15:06:51 <bauzas> PaulMurray: n0ano: https://blueprints.launchpad.net/nova/+spec/check-destination-on-migrations
15:07:21 <n0ano> johnthetubaguy, hate to comment on spcs vs. blueprints, I still think there's confusion on that
15:07:39 <bauzas> n0ano: I can help you on that
15:07:50 <johnthetubaguy> n0ano: if you need clarification on any specific thing, just ask me in IRC, let me find the doc for you...
15:07:51 <bauzas> n0ano: if you have quesitons
15:08:19 <n0ano> I'm not confused, I think other people are
15:08:21 <johnthetubaguy> n0ano: this is the agreed process, with examples: http://docs.openstack.org/developer/nova/devref/kilo.blueprints.html#when-is-a-blueprint-needed
15:08:25 <PaulMurray> johnthetubaguy, you approved the bp with target liberty-1 on grounds that its all up for review already
15:08:27 <bauzas> but basically, something trivial doesn't need a spec, but some trivial thing needing having an operator visibility deserves a spec
15:08:35 <bauzas> like API changes or DB migrations
15:09:19 <n0ano> we're in the situation where the spec was approved for Kilo but all it's patches didn't make it...
15:09:34 <n0ano> now we have active BPs and a spec that needs to be re-approved for Liberty
15:09:36 <johnthetubaguy> n0ano: if there are issues, do raise them in the nova-meeting in the open discussion, and we can try clarify and update the docs if needed, my advice is to "ask questions"
15:10:08 <bauzas> n0ano: it can be fast-approved as a spec
15:10:22 <johnthetubaguy> n0ano: emailed on the ML, same as last release, spec is only approved for one release, the BP will be unapproved to show that
15:10:32 <bauzas> n0ano: I had my 2 specs very quickly approved for Liberty since they were already approved for some previous cycles
15:10:52 <johnthetubaguy> n0ano: the process is email me or ping me if there is confusion, and I will sort it out for you
15:10:59 <n0ano> bauzas, that's what I was hoping, in fact, I was hoping that all out Kilo specs that didn't get implemented could be quickly re-approved
15:11:09 <n0ano> s/out/our
15:11:21 <johnthetubaguy> n0ano: its on the ML, just resubmit with the tag PreviouslyApproved: kilo
15:11:27 <johnthetubaguy> and ping me if we are slow
15:11:28 <bauzas> n0ano: at least the ones I reproposed were re-approved
15:11:46 <bauzas> n0ano: I didn't resubmit for jaypipes or PaulMurray's ones
15:12:20 <n0ano> bauzas, then the only ones left are PaulMurray (we just talked about it) and jay's for object models (he should be on top of that)
15:12:30 <PaulMurray> johnthetubaguy, bauzas so I should resubmit the spec then?
15:12:45 <bauzas> n0ano: yup, I already discussed with jay about that and he accepted to do so
15:12:49 <johnthetubaguy> PaulMurray: almost certainly yes, ping me after with a link, and I can confirm
15:12:52 <bauzas> n0ano: but we can probably hassle him :)
15:13:01 <PaulMurray> johnthetubaguy, ok
15:13:16 <johnthetubaguy> more details here: http://specs.openstack.org/openstack/nova-specs/readme.html
15:13:20 <n0ano> little bit of process foo needed but I think we're sorted now
15:13:22 <johnthetubaguy> they might need updating a bit
15:14:14 <johnthetubaguy> ideas on how to make this all clearer are very welcome
15:14:49 <n0ano> well, it would be clearer to me if we dropped specs and went back to just blueprints but I don't think I want to fight that battle
15:16:12 <n0ano> in summary, bauzas has re-submitted his specs and needs to submit a new one, PaulMurray will re-submit his and hopefully jay is on top of re-submitting his
15:16:58 <bauzas> n0ano: I actually need to submit two :D
15:17:33 <bauzas> https://blueprints.launchpad.net/nova/+spec/check-destination-on-migrations and https://blueprints.launchpad.net/nova/+spec/shared-state-scheduler
15:17:33 <n0ano> bauzas, that works also, I just want to keep on top of the work we're doing so nothing drops through the cracks
15:17:41 <bauzas> n0ano: agreed
15:18:12 <n0ano> bauzas, I'll add those two to the wiki, unfortunately since they're new the approval process might take a little longer
15:18:19 <bauzas> n0ano: fair point
15:18:32 <bauzas> n0ano: but new is always better, right ? :)
15:18:54 <n0ano> bauzas, don't get me started :-)
15:19:27 <n0ano> I think we're good on this subject for now, let's move on...
15:19:34 <bauzas> yeah
15:19:39 <n0ano> #topic Liberty summit ideas
15:19:56 <bauzas> so, I guess most of you guys missed that
15:20:24 <bauzas> https://etherpad.openstack.org/p/liberty-nova-summit-ideas L270 and below
15:21:16 <n0ano> looks like Wed. 9-10:30 is the scheduler session
15:21:50 <bauzas> n0ano: looks like the whole morning, you mean ?
15:22:06 <edleafe> yeah, the whole morning
15:22:19 <edleafe> with cells and RT
15:22:22 <bauzas> n0ano: because, and that's my 2nd point, cells V2 people need us
15:22:37 <n0ano> I guess the question is what about cross project, wed. morning looks like it just Nova focused
15:22:51 <bauzas> n0ano: cross-project stuff is on *Tuesday*
15:23:32 <n0ano> I see that on the schedule but it's not broken out, just all day
15:23:55 <bauzas> n0ano: because cross-project proposals are not managed by the nova team
15:24:35 <bauzas> n0ano: tl;dr tues is for cross-project talks, wed for nova internals, thurs for operator-related nova stuff and friday is free beer event
15:24:44 <n0ano> I hope they are managed by someone, who do we need to lobby to make sure the scheduler is covered
15:25:04 <bauzas> n0ano: for the cross-project related stuff ?
15:25:09 <n0ano> bauzas, yes
15:25:20 <bauzas> n0ano: well, I actually discussed with a neutron guy and he wants to do QoS
15:25:24 <edleafe> ttx might be a good starting point
15:25:37 <edleafe> n0ano: ^^
15:25:42 <bauzas> n0ano: but I don't know exactly if we need to produce a cross-project show
15:26:01 <n0ano> edleafe, yeah, good point, I'll ask him what the plans are
15:26:04 <bauzas> n0ano: in particular as it's quite unclear if we shape the scheduler out of nova, or if we leverage it inline
15:26:38 <n0ano> bauzas, without spcifics about what to discuss at x-project I'm afraid nothing will be decided
15:27:01 <bauzas> n0ano: the main problem is that we have little to show up
15:27:18 <bauzas> n0ano: like saying "eh, that's how it will work"
15:27:32 <bauzas> even if we have a plan
15:28:31 <n0ano> bauzas, I'm more interested in  asking Cinder, Neutron, Containers - what do you need to be measured and how do you want things scheduled - e.g. what are your requirements
15:29:00 <edleafe> n0ano: the cross-project meeting is later today
15:29:09 <bauzas> n0ano: that's worth asking
15:29:16 <bauzas> edleafe: it has been cancelled
15:29:23 <bauzas> for today I mean
15:29:52 <edleafe> bauzas: ah - hadn't seen that
15:29:57 <edleafe> it's still on my scheudler
15:30:08 <n0ano> I'll start with email to ttx and go from there, I can attend a x-project IRC next week (today would be prolematic)(=
15:30:10 <bauzas> you have a CrossProjectFilter ?
15:30:39 <edleafe> n0ano: that goes to what I've been pushing: defining what a resource is, for anything that might be scheduled
15:30:48 <edleafe> i.e., how to represent resources abstractly
15:30:56 <n0ano> edleafe, +1
15:31:09 <edleafe> the current design is completely compute-centric
15:31:23 <n0ano> edleafe, you think :-)
15:31:26 <bauzas> edleafe: that's just the whole idea of https://blueprints.launchpad.net/nova/+spec/resource-objects
15:31:33 <PaulMurray> edleafe, I'll bend your ear on that one
15:32:17 <PaulMurray> edleafe, (that means tell you everything I want - not hurt you in any way )
15:32:26 <bauzas> edleafe: without ^, I think we would face some hard time, in particular given the optimistic design of the scheduler
15:33:06 <bauzas> edleafe: because it implies a check for claiming a resource that the project needs to implement
15:33:24 <edleafe> bauzas: yes, resource objects are a great idea
15:33:39 <bauzas> edleafe: resource objects are just one flip side of the problem
15:33:50 <bauzas> edleafe: they are how to describe a resource
15:33:52 <edleafe> we just need to make sure that they can adequately represent all the resources we may be scheduling
15:33:57 <bauzas> edleafe: the other side is how to consume it
15:34:19 <bauzas> ajo: around ?
15:34:37 <edleafe> PaulMurray: yeah, I know the phrase :)
15:34:43 <bauzas> ajo was having a good example of what neutron guys want to achieve
15:35:34 <edleafe> claiming will always be racy with two sources of data
15:36:21 <n0ano> edleafe, which two sources are you thinking of?
15:36:31 <bauzas> edleafe: that's what I'm beginning to think that the scheduler is definitely nova-centric
15:36:43 <bauzas> edleafe: because at the end, the user wants to boot a VM
15:37:21 <edleafe> n0ano: nova db and scheduler db/in-memory
15:37:31 <bauzas> edleafe: but it could only be one way to claim, either you ask for a VM (then Nova), or for a volume (then Cinder)
15:37:36 <n0ano> bauzas, or allocate storage (Cinder) or assign a network (Neutron) - it's not all about VMs
15:37:47 <bauzas> n0ano: it depends on your usecases
15:38:00 <bauzas> n0ano: if you're VM-centric or not
15:38:06 <bauzas> n0ano: but yeah
15:38:15 <edleafe> bauzas: assuming it's all about VMs would not be good
15:38:23 <n0ano> edleafe, I assume that ultimately the source of truth will be in-memory, not DB, and ther'll only be one soruce
15:38:37 <bauzas> edleafe: that's what I want to achieve with shared-state-scheduler, my big baby I want to grow up :)
15:38:45 * n0ano needs to learn to type today, hopefully everyone can work out what I'm saying
15:39:05 <edleafe> n0ano: no worries; we ignore what you write :)
15:39:18 <bauzas> n0ano: edleafe: I would recommend you to read https://blueprints.launchpad.net/nova/+spec/shared-state-scheduler and the associated etherpads
15:39:22 <bauzas> etherpad even
15:39:24 <n0ano> edleafe, touche :-)
15:39:35 <edleafe> bauzas: yes, I've read them
15:39:58 <lxsli> o/ sorry I'm late
15:39:59 <bauzas> edleafe: so there is actually no matter of a Nova DB or whatever
15:40:11 <n0ano> lxsli, NP, we just gave all the ARs to you :-)
15:40:14 <bauzas> edleafe: we have a datastore that needs to be updated
15:40:27 <edleafe> bauzas: by nova :)
15:40:43 <edleafe> (or any other scheduled service)
15:40:51 <lxsli> :)
15:40:58 <bauzas> edleafe: by the scheduler rather
15:41:08 <n0ano> edleafe, hopefully by the resource manager (may or may not be part of Nova)
15:41:32 <bauzas> anyway, we're diverting
15:42:10 <n0ano> bauzas, true, this is kind of summit discussion but we can probably move on...
15:42:18 <n0ano> #topic opens
15:42:23 <n0ano> Anything new for today?
15:42:34 <PaulMurray> out of interest
15:43:04 <PaulMurray> has anyone come across anything to do with something like "rack aware scheduling" etc. recently
15:43:36 <PaulMurray> or any other kind of infrastructure or HA awareness
15:43:51 <n0ano> PaulMurray, I think people inside Intel are thinking about that, I don't have any more details but I can find out
15:44:09 <bauzas> PaulMurray: affinity filters can help that, but nothing really physical yet
15:44:10 <n0ano> PaulMurray, we're certainly look into HA issues
15:44:13 <PaulMurray> n0ano, would be good - I ask because I have some requirements in that area currently
15:44:33 <PaulMurray> would be good to discuss with intersted parties
15:44:36 <bauzas> n0ano: the HA story is not provided by the scheduler
15:44:45 <n0ano> PaulMurray, are you more intersted in physicality or HA
15:44:48 <bauzas> the Nova community is very clear on that
15:45:18 <n0ano> bauzas, I wouldn't say `provided by` but it can have an impact
15:45:26 <PaulMurray> not necessarily HA - more about placement relative to other things
15:45:33 <PaulMurray> the requirements come form things like hadoop
15:46:00 <bauzas> PaulMurray: eh, that reminds me something starting with Solver and ending with Scheduler
15:46:31 <bauzas> PaulMurray: but to be clear, nothing but affinity filters and servergroups now
15:46:32 <n0ano> PaulMurray, I can check, I think we have people looking into that area just not in my group
15:47:05 <PaulMurray> bauzas, that is on the right track, but it doesn't really have to be that complicated
15:47:22 <PaulMurray> n0ano, thanks
15:47:57 <n0ano> Anything else?
15:48:38 <n0ano> I'm hearing crickets so...
15:48:40 <bauzas> PaulMurray: if it's all about HA http://blog.russellbryant.net/2015/04/08/implementation-of-pacemaker-managed-openstack-vm-recovery/
15:49:42 <n0ano> after being startled for a moment the crickets came back
15:50:02 <n0ano> OK, tnx everyone and we'll talk next week
15:50:10 <bauzas> cool bye
15:50:12 <n0ano> #endmeeting