13:59:43 #startmeeting nova-scheduler 13:59:44 Meeting started Mon Jan 18 13:59:43 2016 UTC and is due to finish in 60 minutes. The chair is n0ano. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:59:45 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:59:47 The meeting name has been set to 'nova_scheduler' 13:59:48 o/ 13:59:57 anyone here to talk abou the scheduler? 14:00:10 \o 14:00:14 i'm here but happy to pass if there's nothing going (other than pointing at some reviews that need some eyes) 14:00:16 * n0ano cdent is too quick too early in the morning 14:00:35 cdent, that's was kind of my idea 14:00:53 let's see who all joins 14:00:55 It's 2 in the afternoon for me, but that's still early for me 14:01:14 edleafe, isn't it a holiday for you, take a break :-) 14:01:33 n0ano: not a work holiday, unfortunately 14:02:06 how odd, I thought MLK was pretty universal these days, oh well 14:03:14 oh méan 14:03:26 I was waiting in #openstack-meeting :) 14:03:35 sooo 14:03:38 * bauzas waves 14:03:56 it's holiday, yeah, but not a paid day off 14:04:02 bauzas, NP, I've tried to kick people off #openstack-meeting in the past :-) 14:04:14 is there an US holiday those days ? 14:04:23 I can see some folks out 14:04:39 bauzas, yep, Martin Luther King day, it's an official holiday for me, not for edleafe 14:05:01 ok 14:05:06 anyway, let's get started 14:05:23 I like cdent idea, let's just have some quick opens and go, so 14:05:27 #topic opens 14:06:10 gerrit ill? 14:06:47 cdent, reall? I got emails yesterday, if you think it is broke you should bring it up on #openstack-nova 14:07:04 No, just now it was giving me 502s 14:07:10 several times 14:07:13 but I guess it was a blip 14:07:31 I wanted to point at this: https://review.openstack.org/#/c/262561/ a documentation tuneup that's been sitting idle for a while 14:08:28 gerrit wfm 14:08:39 cdent: I got blipped, too, but then it started working again 14:08:51 working now, hopefully OK 14:09:05 * cdent shrugs, c'est le gerrit 14:09:18 cdent, looks like you need a couple of +2's on that 14:09:35 cdent: cosmetic change only, right? 14:09:57 pretty much 14:10:04 tried to tune some grammar 14:10:05 okay, I'll look at 14:10:13 and added links 14:10:25 okay, because me is a French guy, so you know about my grammar 14:11:52 bauzas, s/me is/I am/ :-) I see your point 14:11:52 Is jaypipes around or is he properly holidaying? It appears holidaying (no tab complete). 14:12:24 I'm wondering what the situation is on the resource-* specs. I've got https://review.openstack.org/#/c/263496/ this (and its relations) pending, but the spec is still out for review 14:13:49 cdent, jay is always iffy, looks like he's not here today 14:14:37 cdent, you might want to send him an email to see where he is on the spec 14:14:53 cdent: since the universe exploded with many specs, I reiterate my will to get the specs + patches in https://etherpad.openstack.org/p/mitaka-nova-priorities-tracking 14:15:08 I poked him late last week he said that he'd have something at least by today 14:15:30 he also said he'd put stuff on there but I guess that didn't happen. I'll do it. 14:15:53 cdent, then a polite reminder to him seems appropriate 14:16:12 * cdent nods 14:16:37 AOB (Any Other Business)? 14:17:16 not really 14:17:35 Is 'grep -i schedule tempest' a good way to find scheduler related tests in tempest? 14:18:23 not really IMJO 14:18:39 is there a good way? 14:18:44 cdent: you could look at the scenarios rather 14:18:58 cdent: what do you want to know actually 14:18:59 ? 14:19:55 I'm trying to do a simple analysis of the extent of scheduler coverage in tempest. I already know that it is pretty weak, but I'm just digging around for a bit more data. And given that tempest is generally inscrutable, I just wondered if there was a way to improve my filtering (so I don't have to read everything). 14:21:49 cdent: so, you should rather see which filters are checked 14:22:03 cdent: by seeing which tests are modifying the CONF opts 14:22:51 cdent: since by default a very small portion of filters are enabled, it means that if no tempest tests are modifying that flag, you could guess that it's mostly because they don't use them 14:23:21 * cdent nods 14:23:48 cdent: then, once you admit that 99% of tests (I leave my opinion to be accepting something I dunno) are using the default filters, you could just look at whether the boot requests use hints 14:23:56 or flavor metadata 14:24:00 or image extra specs 14:24:20 by grepping over 'hints' or the other terms 14:24:44 The implication of all that, though, is there's not much in the way of explicit scheduler testing, rather a lot of implicit stuff as a result of various instance boots 14:24:51 and then, you'll see that 99% of the time, tempest doesn't care about providing more than just 'eh, scheduler, boot me a request' 14:24:56 yes 14:25:09 cdent: exactly 14:25:22 speak of the devil 14:25:24 cdent: that's my main worries here, all of the scheduler implications are implicit 14:25:32 so, lemme now tell you a story 14:25:51 of someone modifying the scheduler and not having some explicit tempest tests 14:26:01 you know how that guy can functionnally test that ? 14:26:18 he runs the caching scheduler unittests + he runs the cinder functional tests 14:26:34 that's the only coverage we have 14:26:45 :( 14:26:57 to be clear, just play locally and crash the scheduler 14:27:12 you'll see that only the caching scheduler should notice that 14:27:30 not speaking of a full crash 14:27:36 just something bad in the filters 14:28:35 Well, my hope here is to gain some understanding of existing stuff so I can make it better (and be prepared for the summit session on the topic next week) 14:29:25 cdent: sure, and that's a very valid point 14:30:11 cdent: I'm just telling you that we have 3 levels of coverage : 2 in-tree (unit + functional) and one in Tempest and all of them are not veryfying the scheduler but implicitly 14:30:20 yes 14:30:23 I've figured that out 14:30:39 by saying "implicitly", I mean "run the default filters with a default query and hope that the scheduler will return you a tuple" 14:31:02 s/hope/assert 14:31:25 * cdent nods 14:31:39 one hessitates to say that at least that's something and testing can always be made better 14:31:50 * cdent nods at that too 14:32:12 anyway, anything else? 14:32:21 not from me 14:33:29 crickets from me 14:33:33 note that, barring arguements from anyone, I'll cancel next week's meeting, it should be a travel day for the mid-cycle for most 14:33:40 ++ 14:33:41 +1 14:34:13 OK, tnx everyone, some will talk in person next week, the rest of us will be here in two weeks. 14:34:19 #endmeeting