15:00:48 <bauzas> #startmeeting nova_scheduler
15:00:48 <openstack> Meeting started Tue Jun 23 15:00:48 2015 UTC and is due to finish in 60 minutes.  The chair is bauzas. Information about MeetBot at http://wiki.debian.org/MeetBot.
15:00:50 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
15:00:52 <openstack> The meeting name has been set to 'nova_scheduler'
15:01:05 <bauzas> morning folks
15:01:16 <bauzas> (or pick the one you are)
15:01:34 <edleafe> good $time_of_day to you
15:02:13 <edleafe> we should just adopt UGT: http://www.total-knowledge.com/~ilya/mips/ugt.html
15:02:17 <bauzas> waiting the crowd
15:02:46 <bauzas> edleafe: interesting :)
15:02:57 <edleafe> alex_xu_ won't be on - he has a new daughter today
15:03:01 <bauzas> \o/
15:04:08 <edleafe> we're going to have a long wait if you want a crowd :)
15:04:44 <bauzas> edleafe: well, I should say that Justin Bieber is there
15:05:07 <lxsli> o/
15:05:10 <edleafe> then I'm outta here
15:05:35 <NotJustinBieber> edleafe: do you prefer then ?
15:05:50 <edleafe> much better
15:05:57 <NotJustinBieber> (btw. this nick is registered. so awesome)
15:06:06 <lxsli> I couldn't get xi :(
15:06:21 <NotJustinBieber> okay enough kidding, we're 3
15:06:51 <nihilifer> hello everyone
15:06:58 <bauzas> nihilifer: hi
15:07:04 <bauzas> so, let's start then
15:07:12 <bauzas> #topic Specs process
15:07:34 <bauzas> so, we agreed on putting our review requests in https://etherpad.openstack.org/p/liberty-nova-priorities-tracking
15:07:53 <bauzas> I just stroked the specs which were merged
15:08:24 <bauzas> so, we're in a good shape for working
15:08:54 <bauzas> at least the two main ones (request-spec-object and resource-objects) are approved for Liberty, thanks to jaypipes and lxsli
15:09:26 <edleafe> Still waiting on NoValidHost reporting
15:09:27 <bauzas> so, there are 2 specs requests waiting for approval
15:09:40 <bauzas> https://review.openstack.org/187739
15:09:41 <edleafe> seems like people wanted it to solve every problem instead of just one :)
15:09:53 <bauzas> edleafe: yeah, I saw that
15:10:25 <bauzas> edleafe: I'm going to do another pass by today if I can, so if I'm happy, I could put that one as "given +1 from the subteam"
15:10:45 <bauzas> edleafe: to be clear, I would like to see that spec worked for Liberty
15:10:52 <lxsli> edleafe: My preferred response to such is "that's a great idea you are welcome to propose in a follow-on spec"
15:11:09 <edleafe> bauzas: that actually happened
15:11:09 <bauzas> edleafe: I think you scoped the spec correctly
15:11:27 <edleafe> https://review.openstack.org/194204
15:11:32 * jaypipes on a hangout...
15:11:52 <bauzas> edleafe: I was having some concerns about the scalability on that, so I need a few more time for thinking about that
15:12:19 <edleafe> bauzas: Sure - it went through several scenarios before I settled on this one
15:12:22 <bauzas> edleafe: I got your reply on the IO concern, I'm now still a little concerned about the memory usage
15:12:27 <edleafe> bauzas: most were highly un-scalable
15:12:52 <bauzas> edleafe: agreed, hence why I need some time to review your spec - because I don't want to argue against something previously having a consensus
15:12:55 <edleafe> bauzas: exactly. To reduce memory, I'd have to log every step
15:13:11 <edleafe> bauzas: then that would clutter the logs uselessly
15:13:15 <bauzas> edleafe: okay, lemme give some time to think about it
15:13:29 <edleafe> of course
15:13:34 <bauzas> edleafe: the real question behind that is "does that honestly need a spec ?'
15:14:00 <edleafe> Well, since there are concerns about the effect of the implementation, I'd say 'yes'
15:14:07 <bauzas> edleafe: you're not changing the APIs (neither REST or RPC), you're not persisting something, you're not adding something in a dict etc.
15:14:42 <bauzas> edleafe: okay, I would propose you to put the BP as questionable for the next nova meeting if it deserves a spec
15:15:03 <bauzas> edleafe: because I've seen more controversial BPs getting approved without a spec
15:15:13 <edleafe> bauzas: yeah, I got no feedback on the BP itself
15:15:16 <bauzas> edleafe: I mean, we can still argue about the implementation
15:15:23 <edleafe> so I wrote the spec
15:15:38 <bauzas> edleafe: and discuss on technical details, but that can be done directly in the implementation
15:15:56 <bauzas> edleafe: having no specs doesn't necessarly mean it's trivial :)
15:16:04 <edleafe> well, part of the idea of a spec is to argue *before* time is spent on an implementation
15:16:34 <bauzas> edleafe: agreed, but like I said, I can see a general consensus on the design while some discussion happens on the implementatioon
15:16:53 <bauzas> edleafe: I mean, I saw nobody complaining about logging stuff - rather about detailsd
15:17:12 <edleafe> bauzas: yeah, it's not a clearly defined border between "needs spec" and "no spec required"
15:17:14 <bauzas> edleafe: so maybe it would be worth asking the question in a nova mereting
15:17:19 <nihilifer> about the "does that honestly need a spec?" question
15:17:35 <bauzas> edleafe: because we do have a topic for that
15:17:43 <nihilifer> some time ago I "resurrected" an old spec, not approved for Juno
15:17:48 <bauzas> edleafe: https://wiki.openstack.org/wiki/Meetings/Nova#Agenda_for_next_meeting
15:18:05 <bauzas> nihilifer: yup ?
15:18:13 <nihilifer> https://review.openstack.org/#/c/181054/
15:18:24 <nihilifer> I had some doubts whether it needs spec or not
15:18:29 <nihilifer> since it's a filter
15:18:40 <nihilifer> but it also requires new conf option
15:18:45 <bauzas> nihilifer: see my last comment :)
15:18:56 <nihilifer> oh, it's you :)
15:19:07 <nihilifer> that's what I wanted to ask about
15:19:22 <bauzas> nihilifer: so, there are some examples
15:20:12 <bauzas> nihilifer: see http://docs.openstack.org/developer/nova/devref/kilo.blueprints.html#when-is-a-blueprint-needed
15:20:24 <bauzas> nihilifer: given your proposal, I can see it as a self-contained change
15:21:15 <nihilifer> bauzas: ok, then I will just begin work on it
15:21:30 <nihilifer> bauzas: thanks for clarification
15:21:44 <bauzas> nihilifer: the same as for edleafe, I would recommend you to officially ask in https://wiki.openstack.org/wiki/Meetings/Nova#Agenda_for_next_meeting
15:22:02 <bauzas> nihilifer: but I'm clearly pro-BP only
15:22:21 <bauzas> nihilifer: in case of a needed spec, just resurrect it
15:22:35 <lxsli> my 2p is I'd like a "trivial specs" process, so in cases like this where someone has taken the trouble to write a spec, we can fast-track it rather than abandoning the spec
15:22:37 <edleafe> bauzas: I prefer the spec discussions myself
15:22:43 <bauzas> nihilifer: hence the formal question 'do I need a spec' during the nova meeting
15:22:48 <lxsli> that's not really a subject for the scheduler meeting though
15:23:17 <bauzas> lxsli: my idea on that is that we can't hardly assume the current spec backlog
15:23:44 <bauzas> lxsli: so, we probably indeed need some way to ask for drafting out a design doc if needed
15:23:50 <bauzas> but only if needed
15:23:57 <lxsli> bauzas: things "too simple to need a spec" naturally make for easy, quick spec reviews though, no?
15:23:59 <bauzas> edleafe: see the examples on the link I provided
15:24:12 <lxsli> bauzas: and this process may reveal it wasn't quite as simple as was initially thought
15:24:20 <bauzas> edleafe: there are 2 examples of filters changes not requiring a sepc
15:24:37 <edleafe> bauzas: agreed
15:24:38 <bauzas> lxsli: agreed, that's an on-going improvment process
15:24:47 <lxsli> I may be biased because I find writing in English quite easy ^^;
15:25:17 <bauzas> lxsli: so I will officially ask johnthetubaguy to ask for specs written in French only
15:25:20 <bauzas> enough kidding
15:25:49 <bauzas> last spec is https://review.openstack.org/#/c/179224/
15:25:51 <bauzas> #link https://review.openstack.org/#/c/179224/
15:26:05 <bauzas> still needs some approval, because there is a REST API change
15:26:15 <bauzas> (plus a behavioural change)
15:26:38 <bauzas> so I chatted with jaypipes, let's see if he will obviously kill my idea or support it :)
15:27:04 <bauzas> any specs to mention before we move on ?
15:27:05 <lxsli> doesn't seem obviously crazy
15:27:10 <lxsli> yes
15:27:23 <lxsli> https://review.openstack.org/#/c/84906
15:27:32 <bauzas> lxsli: feel free to review the spec, I could mark it as 'given +1 from a subteam member' accordingly :)
15:27:34 <edleafe> lxsli: emphasis on 'obviously' :)
15:27:37 <lxsli> this got merged for Juno but afaik nothing has happened with it
15:28:00 <lxsli> bauzas: dansmith assigned me a tonne of work to do with no-more-soft-delete, I've been wading through that :(
15:28:14 <bauzas> lxsli: no excuses, you gotta review !
15:28:16 <dansmith> lxsli: assigned? :P
15:28:23 <dansmith> lxsli: it's much appreciated though
15:28:24 <lxsli> dansmith: suggested ^^
15:28:34 <bauzas> lxsli: don't you work 24x7 ?
15:28:52 <lxsli> I am strenuously avoiding that!
15:29:00 <bauzas> anyway, so back to 84906
15:29:28 <bauzas> lxsli: that spec would require a refresh for Liberty anyway, so there are litterally very little chances we could work on it for this cycle
15:29:39 <lxsli> yeah, does it seem plausible for Mita ?
15:29:48 <bauzas> lxsli: also, I think some work has to be done before that one
15:30:08 <bauzas> lxsli: don't say the name :p
15:30:23 <bauzas> lxsli: so, even Muppet, I dunno
15:30:24 <lxsli> Sorry, Muppet :)
15:31:01 <bauzas> lxsli: ndipanov is planning to work on something about claims within the scheduler, and also fix the missing gaps with claims for Libertyy
15:31:15 <bauzas> lxsli: sec, giving you the specs
15:31:22 <lxsli> bauzas: awesome, thanks
15:31:31 <bauzas> lxsli: so I would definitely prefer ndipanov's work be done before we persist those
15:31:48 <bauzas> lxsli: and honestly, I would also prefer resource-objects to be landed *before* that
15:31:54 <lxsli> bauzas: that's cool, just getting a sense of where we are
15:31:59 <lxsli> yep definitely
15:32:09 <lxsli> I still need to look at John's scheduler-evolution thing :(
15:32:13 <bauzas> lxsli: so the specs are
15:32:14 <bauzas> https://review.openstack.org/#/c/191226
15:32:31 <bauzas> https://review.openstack.org/#/c/193576/
15:32:41 <bauzas> lxsli: I still need to provide an update too
15:32:49 <bauzas> lxsli: I basically promised it to johnthetubaguy
15:32:59 <edleafe> bauzas: also https://review.openstack.org/187739
15:33:37 <bauzas> edleafe: your point on that one? thought we discussed about it before ?
15:34:01 <edleafe> bauzas: I thought you were listing outstanding specs needing review
15:34:02 <lxsli> edleafe: that's a no-valid-host spec, did you paste the wrong one?
15:34:03 <bauzas> edleafe: oh, you mean the specs being worked for Liberty ? yeah, we're tracking them
15:34:37 <bauzas> edleafe: no worries, I'm trying to keep the etherpad of doom up-to-date
15:34:44 <bauzas> ok, moving on ?
15:35:07 <bauzas> 1
15:35:08 <bauzas> 2
15:35:10 <bauzas> ...
15:35:11 <bauzas> 3 ?
15:35:18 <bauzas> #topic Open discussion
15:35:39 <bauzas> oh sec
15:35:40 <bauzas> #undo
15:35:41 <openstack> Removing item from minutes: <ircmeeting.items.Topic object at 0x9668350>
15:36:10 <bauzas> #topic new meeting time
15:36:18 <bauzas> edleafe: you had an action for this
15:36:40 <edleafe> jaypipes said that Monday might be a little better than Tue-Wed
15:37:01 <edleafe> I'm wondering if it's worth doing another Doodle to see if that would work for everyone
15:37:04 <ndipanov> persist what bauzas ?
15:37:30 <bauzas> edleafe: fair
15:37:34 <lxsli> edleafe: Monday is fine for me
15:37:55 <edleafe> ok, then I'll send out the doodle message later today
15:38:15 <edleafe> #action edleafe to create Doodle for Monday meeting options
15:38:16 <bauzas> ndipanov: https://review.openstack.org/#/c/84906 given by lxsli
15:38:23 <bauzas> #action edleafe to create Doodle for Monday meeting options
15:38:49 <bauzas> ok, moving on
15:38:53 <bauzas> #topic open discussion
15:39:19 <bauzas> I have nothing to say but that I'm incrementally updating the request-spec-object BP
15:39:37 <bauzas> reviews are welcome if you're fed up by reviewing specs
15:39:40 <xyhuang> hey guys may i have 1 more request for review...
15:39:42 <xyhuang> https://review.openstack.org/#/c/193635
15:39:54 <bauzas> https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bp/request-spec-object,n,z
15:40:50 <bauzas> xyhuang: strange, I thought I said it needs a spec
15:41:09 <bauzas> xyhuang: because it changes the interface between nova and the scheduler
15:41:37 <xyhuang> thanks bauzas, will do a spec
15:42:00 <bauzas> xyhuang: that's MHO, probably someone could say something else
15:42:18 <lxsli> +1 to spec
15:42:20 <edleafe> bauzas: I would tend to agree
15:42:58 <bauzas> xyhuang: that said, I wrote a spec for amending the future RequestSpec object by passing a destination host, so yeah I would love a spec still
15:43:14 <bauzas> xyhuang: do you know about http://specs.openstack.org/openstack/nova-specs/specs/liberty/approved/request-spec-object.html ?
15:43:38 <xyhuang> bauzas: yes
15:43:48 <lxsli> xyhuang: have you seen this? https://review.openstack.org/#/c/184534/
15:44:11 <xyhuang> oh probably no...
15:44:25 <lxsli> xyhuang: I've only just seen your patch but it may make sense to provide the network info as resource objects
15:44:50 <bauzas> lxsli: hence a spec, because I wonder if that's a compute resource or a request information
15:45:00 <lxsli> xyhuang: there's already an implementation patch, for reference: https://review.openstack.org/#/c/128992/
15:45:02 <bauzas> lxsli: if we claim about it, then you're right
15:45:38 <xyhuang> lxsli: thanks will check it out
15:45:45 <lxsli> xyhuang: very welcome :)
15:46:11 <bauzas> xyhuang: so yes, I think you should consider how to provide that information/resource to the scheduler
15:46:23 <bauzas> hence the specd
15:46:24 <bauzas> spec
15:46:33 <xyhuang> i see
15:46:39 <xyhuang> so a spec will be done
15:46:45 <bauzas> xyhuang: that said, do you know that there is a spec freeze deadline by Thur ?
15:47:10 <xyhuang> not aware of that, but will do it asap
15:47:39 <bauzas> xyhuang: as you would probably require either resource-objects or request-spec-object to be implemented, I think it's difficult to merge the implementation by Liberty but rather the M cycle
15:48:20 <lxsli> Yes I agree - it's very unlikely it can be done in L
15:49:10 <xyhuang> would it be possible to add it based on the current arch of request-spec? as this may be a small change
15:49:42 <bauzas> xyhuang: that's the design discussion which needs to be done
15:50:06 <bauzas> xyhuang: I mean, is it something that nova-compute is reporting to the scheduler, or is that only a hint given by the user ?
15:50:13 <lxsli> xyhuang: we want to be more strict about what can be in request-spec to aid Nova back-compatibility and stability
15:50:53 <xyhuang> ok
15:51:08 <lxsli> xyhuang: so adding extra stuff to it will make that more difficult, I expect core team will prefer to wait until it's tidier
15:51:43 <xyhuang> ok i see, anyway i will write up the spec and we can discuss later in more details maybe
15:51:51 <bauzas> xyhuang: cool
15:52:24 <xyhuang> thanks
15:52:59 <bauzas> okay, any other question/concern before we call that meeting done ?
15:53:18 <edleafe> nope
15:53:25 <edleafe> at least not from me :)
15:54:12 <lxsli> we seem to have moved to #nova anyway :)
15:55:06 <edleafe> lxsli: we never left :)
15:55:41 <bauzas> I'm still there
15:55:57 <bauzas> okay, saying bye bye to the meeting ?
15:56:01 <bauzas> #endmeeting