13:00:36 #startmeeting nova api 13:00:37 Meeting started Wed Apr 19 13:00:36 2017 UTC and is due to finish in 60 minutes. The chair is alex_xu. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:00:39 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:00:42 The meeting name has been set to 'nova_api' 13:00:47 who is here today? 13:00:57 o/ 13:01:29 gmann: let us wait two minutes for johnthetubaguy and sdague 13:01:34 sure 13:01:43 o/ 13:01:48 o/ 13:02:42 o/ 13:02:57 #topic priorities 13:03:14 o/ 13:03:31 johnthetubaguy: from the last nova weekly meeting, I saw there are some note about more problem found for policy 13:04:17 ah, yes 13:04:35 I should have added comments in the spec, but... well things got in the way 13:04:57 basically, I think we need more consensus on the general direction around policy 13:05:08 there is a google hangout scheduled for later today 13:05:17 in the keystone policy meeting slot 13:05:45 ok, cool 13:05:54 I think sdague and I got close to the problem, its very related to the concept of "global" ness 13:06:03 like project observer vs global observer 13:06:09 and how that would fit into the world 13:06:26 the current status quo is massively confusing 13:06:30 yeh 13:06:48 is_admin_project is the main point of contention 13:06:49 I honestly feel like its a few more hours of discussion to unpack the whole thing 13:06:55 sdague: +1 13:07:36 honestly, I know folks glazed over at the PTG in a fog of confusion, we need to get through that 13:07:36 ok, looking for good thing come out 13:08:08 yeah, I think we have folks interested and thinking about the problems, so hoping we can straighten something out 13:08:43 sdague: I forget actually, did we say we keep anything from those two specs, I think probably not? 13:09:09 johnthetubaguy: I don't know 13:09:11 obviously, we finish off the policy docs 13:09:19 honestly, I feel like we were starting to get somewhere, then folks had to go 13:09:24 and we can still remove those still policy rules 13:09:42 sdague: yeah, that was "the OSIC" meeting 13:09:47 ah 13:09:55 yeh, I can understand why that was important 13:11:44 ok, we just need to wait for new spec come out, i guess? 13:12:19 yeh 13:12:35 +1 13:12:42 I honestly think glossary and diagrams of the current state of the world would make more sense then a new spec in a lot of ways 13:12:54 yeah, I think that the bit we need 13:13:01 the glazing over is because the current reality is pretty hard to understand 13:13:01 the landscape 13:13:20 turns out we all don't have the same context here, including lots of the keystone folks 13:13:30 the only reason I can follow half of it is because I learned a chunk for the policy in code bits 13:13:30 well, we all have different bits of the context 13:13:55 yeah 13:14:15 anyways, more ground work needed around that stuff 13:15:18 you guys' words make me looking forward to see the new spec a lot, looking for a surprise :) 13:15:28 heh, maybe! 13:15:43 heh 13:15:53 #topic open 13:16:11 #link https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bp/api-no-more-extensions-pike 13:16:20 johnthetubaguy: looking fo one more +2 ^ 13:16:35 * cdent loves that ^ stuff 13:16:42 cdent: :) 13:16:48 I will take a look at that after the meeting 13:17:00 been going through the +2 back log past few days 13:17:07 johnthetubaguy: thanks 13:17:10 yeh, getting the /servers one landed would be great 13:17:10 got stuck in a few holes along the way, digging out now! 13:17:52 try to add a test to ensure the microversions are sequential https://review.openstack.org/#/c/458004/, the test failed, but approved the test works... 13:18:44 interesting 13:18:56 but as sdague said, there is no way to test the microversion sequential on json-schema, which can found bug like this https://review.openstack.org/457577 13:19:43 ah, interesting, I spotted something similar in placement the other day, so it brought this back to mind 13:20:02 alex_xu was good enough to provide a similar thing for placement 13:20:08 sweet 13:20:10 s/thing/test/ 13:20:10 johnthetubaguy: yea, I have a patch for placement also https://review.openstack.org/458049 13:20:21 yeh, I honestly think if we did things a little different in how we executed the wrappers it might make it easier to verify 13:20:21 nice, i did not about that bug but yea checks for such are great to have 13:20:56 for instance some kind of __versions on the function 13:21:20 honestly, I wondered about the explicit routes, should versions go in there? 13:21:30 johnthetubaguy: how? 13:21:49 I love watching you guys be so honest all the time ;) 13:22:04 like GET / -> version range -> function, rather than GET / -> function 13:22:16 cdent: heh 13:22:42 johnthetubaguy: emm...sounds good 13:23:16 johnthetubaguy: but you can't trust that route list, we have a lot of in-code version check 13:23:17 or just put the schema ranges in https://review.openstack.org/#/c/445864/14/nova/api/openstack/compute/routes.py@122 13:23:48 sdague: yeah in there is what I was thinking, how I am less sure 13:23:51 anyway, this is a brainstorming exercise for later 13:23:55 yeah 13:24:04 the point is the interface is currently easy enough to misuse that we have bugs 13:24:12 ++ 13:24:13 agreed its too easy to screw it up 13:24:13 yea, having at router level makes less error prone 13:24:22 and we need to climb the hierarchy of - http://sweng.the-davies.net/Home/rustys-api-design-manifesto 13:24:43 sdague: oh, nice 13:25:17 * alex_xu adds that to readlist 13:25:52 sdague: nice one. thanks for sharing. 13:26:20 * alex_xu always expect sdague's sharing 13:26:25 yeh, no worries, I heard it in a rusty keynote a long time ago, and it always stuck with me 13:26:44 :) 13:26:49 ok, next topic? 13:26:58 I think I only ever saw the tl;dr of that before 13:27:00 anything people want to bring up? 13:27:18 i start fixed in the nova-api under wsgi stuff 13:27:18 I guess spec freeze has passed 13:27:25 cdent: cool 13:27:31 at https://review.openstack.org/#/c/457283/ 13:27:33 ah, sweet 13:27:48 still a bit roughly hewn, suggestions welcome 13:27:59 there's associated devstack changes in progress too: https://review.openstack.org/#/c/457715/ 13:28:09 sdague did a pile of nice work to make uwsgi simple in devstack 13:28:19 and we're thinking may as well make nova-api default to that 13:28:29 one question is: should the metadata server too? 13:28:41 oh... that is a harder question 13:28:59 I guess thats often deployed on the nova-compute node, with a neutron proxy in front of it 13:29:00 keeping in mind that this is for devstack only 13:29:18 I think the md server should run the same way 13:29:24 johnthetubaguy: even in the neutron case? 13:29:29 its tempting to say yes, just not sure what the deployers think 13:29:35 I thought in the neutron case, just the proxy was there 13:29:52 it was nova-net where you either put it local, or did the multicast route thing 13:30:03 does it matter what deployers think, with regard to devstack? part of the point of moving to wsgi-only is so that deployers have a) choices, b) must make a choice 13:30:12 c) have choices which are sane 13:30:17 cdent: I think it's fine to do for devstack 13:30:30 me too, just checking the pulse of others 13:30:45 cdent: does it require the md server as a dedicate uwsgi process? or do they run in the same one? 13:31:32 I think it would be better separate, more debuggable and inspectactable 13:31:45 the mod-wsgi setup (which didn't really work anyway) was separate 13:31:51 sdague: so I had that all wrong in my head, I know what you mean now 13:32:09 and presumably for multi-node separate metadata server is a good thing? 13:32:22 cdent: I like separate too, so everything is the "same" in some sense 13:32:27 * cdent nods 13:32:38 we can kill some more silly config variables that way 13:33:05 johnthetubaguy: true 13:33:23 * johnthetubaguy makes bashing config option noises 13:33:35 honestly, it's probably worth getting this all working enough to deprecate the eventlet way of running nova-api for the release 13:33:42 +1 13:33:47 \o/ 13:34:08 \o/ for removing something again 13:34:22 well, it will be a couple cycles before it can be code deleted 13:34:34 we'll call the the r cycle: removals 13:34:37 * alex_xu sad now 13:34:40 but it does simplifiy 13:34:51 cdent: nice 13:35:09 (and the s cycle: simplify) 13:35:10 we did hard and fast transition on a few things in this last cycle, I think for this we need to give people a little more time 13:35:35 cdent: time to start adding locations on google maps near summit venues 13:35:48 yeah, this one needs time 13:35:55 * cdent makes a todo item 13:36:17 well, the had time for the other ones, just they didn't notice 13:37:44 sounds, should ask next topic? 13:38:02 yep 13:38:09 +1 13:38:23 I don't really have a topic 13:38:25 johnthetubaguy: sounds like you said half words about freeze 13:38:34 I was just going to check we have no specs hanging in mid flight 13:38:38 that are really important 13:38:44 I think the answer is no 13:38:57 i think no 13:39:13 i probably needs to check how much spec approved for api 13:39:28 s/needs/need/ 13:40:18 #link https://review.openstack.org/457181 13:40:28 the patch about deprecation is up ^ 13:40:53 that is all i have 13:41:03 anything more want to bring up? 13:41:50 ok, I guess no 13:42:05 thanks all, let us close the meeting 13:42:13 #endmeeting