20:00:37 #startmeeting arch_wg 20:00:38 Meeting started Thu Jan 12 20:00:37 2017 UTC and is due to finish in 60 minutes. The chair is SpamapS. Information about MeetBot at http://wiki.debian.org/MeetBot. 20:00:39 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 20:00:41 The meeting name has been set to 'arch_wg' 20:00:56 Courtesy ping for nikhil, harlowja, dstanek, kragniz, auggy, rockyg, rocky_g, kgiusti 20:01:01 hello hello 20:01:12 #link https://wiki.openstack.org/wiki/Meetings/Arch-WG#Agenda 20:01:18 ttx: thanks for coming! 20:02:05 I was hoping more would o/ given that we're kicking off 2017 :) 20:02:25 o/ 20:02:32 ah 20:02:35 yo 20:02:37 huzzah, we've 4 20:02:53 #topic previous meeting action items 20:03:06 #link http://eavesdrop.openstack.org/meetings/arch_wg/2016/arch_wg.2016-12-15-20.03.html 20:03:08 even with that "we're gonna gossip about nova" invite 20:03:10 FWIW I pushed https://review.openstack.org/#/c/419402/ so that arch_wg changes are mentioned on channel 20:03:12 * cdent is disappoint 20:03:28 * SpamapS SpamapS email proposed arch-wg repo process to openstack-dev 20:03:33 ttx: SWEET thanks 20:03:50 cdent: right? I figured we'd at least get some haters. :) 20:04:07 we need "container" in every title 20:04:12 I did in fact email openstack-dev and I think at this point we're more in the "let's try the process out" phase. 20:04:18 ttx: *ding ding* 20:04:29 * cdent drinks 20:04:30 * SpamapS SpamapS write up nova-compute rant as a proposal to arch-wg repo 20:04:41 I did that, and it's in the repo now yay 20:04:50 got some nice responses 20:05:13 Indeed, I think with the two proposals we have we'll be kept quite busy through Q1 20:05:14 SpamapS there is a nova-compute rant? 20:05:16 where where 20:05:16 * SpamapS Rockyg to write up implementation bleed-through thoughts and submit to arch-wg repo 20:05:34 harlowja: http://git.openstack.org/cgit/openstack/arch-wg/tree/proposals/nova-compute-api.rst 20:05:45 kk, thx 20:05:48 I don't see a rockyg, so we'll have to carry that one 20:06:16 #action Rockyg to write up implementation bleed-through thoughts and submit to arch-wg repo (carried from December 2016) 20:06:44 and finally 20:06:46 * SpamapS SpamapS send message to openstack-dev cancelling further 2016 meetings. 20:06:48 did that one. 20:06:51 woot 20:06:52 ha 20:06:59 #topic Proposal Process Review 20:07:24 So, what do people think? Do we have a well defined process that is working? Any tweaks needed? 20:07:55 I've been wondering how to move forward on base services and here is what I came up with 20:07:58 I feel like we're still in the alpha test phase so it's a bit of a moving target anyway 20:08:22 I added a "proposed implementation" to the proposal so that people know what that would trigger in practice 20:08:23 I think the "how to make a proposal for discussion" part it is pretty clear (because we've done it) but happens next less so (because we haven't) 20:08:28 then will star ta ML thread on it 20:08:39 i'd like my bikeshed red 20:08:42 lol 20:08:48 * harlowja runs away 20:08:51 and after that we can amend and approve or throwaway 20:09:07 since that one is pretty basic we can use it to test the process 20:09:30 the only thing that is bikesheddable would be the narrow/wide approach 20:09:40 but arguably not directly ties to the proposal 20:09:44 tied* 20:10:31 (a bit orthogonal) 20:11:19 so, the way I've envisioned it, the proposal is just for "do we have a basic idea of what needs to be understood, and enough people who think they could work on it?" 20:11:43 Once we agree on that, we rename it into active, and we start to add action items and assigned people. 20:12:03 ah, ok. I guess we should move it to active first then 20:12:12 Frankly, it's a 1.1 person job 20:12:19 0.1 being +2A on reviews 20:12:29 Hah yeah 20:12:43 well I think the actions will be some openstack specs, and maybe a community goal. 20:12:45 the main work being to push the thing through the TC washing machine 20:13:01 in that case it's just a definition 20:13:27 but before we go deep on that (which I htink we should!) .. does that process sound good? ANd if we do that, do we want to try and land all supporting documentation in not-arch-wg specs, or do we also want to publish our own findings? 20:13:32 we just name the thing that already exists 20:13:57 process sounds ok. I just think I skipped a step 20:14:12 I think for most important things that we want to do the critical piece is publishing why the status quo is not good enough. Without that, nothing will happen. 20:14:40 cdent: yes, we probably need some template of the answer the proposal must answer 20:14:56 including why the status quo is not satisfactory 20:15:16 "why the change" 20:15:28 sadly I think that's also the part that will be most contentious and subject to mystical thinking 20:15:32 Right, so the basic sequence I want to see is "Raise a topic (base-services), agree it's important enough to track, document the current state of things, including problem statements, and then propose specs for addressing said problems, once specs are complete, proposal is considered "done" 20:16:19 and the idea is that we keep checking in on status and raising awareness and finding resources to help complete each stage. 20:16:23 do you imagine a timetable on those things? 20:16:42 sort of three cycles, or something like that? 20:16:44 2 months 20:16:47 5 years 20:16:47 etc 20:16:51 :) 20:16:56 SpamapS: while the proposal can be free-form, I think active topics will have to follow some templates, which we'll have t odefine 20:16:58 I think it's going to be like any other EPIC story.. the pieces will have time tables, but the whole thing will be murky until late in the story. 20:17:29 ttx: +1 for that. Do you want to define it by morphing base-services into an active workstream? 20:17:49 (workstream sounds really weak.. what do we call them? We could still keep them as "proposals".. RFC?) 20:18:01 SpamapS: my fear is that base-srevices is a bit of a corner case. A siple one, but still not a classic example 20:18:06 simple* 20:18:21 ttx: we can both take a stab at it, you with base-services and me with nova-compute-api 20:18:34 ack, let's come up with somethig more detailed and then regroup 20:18:36 then try and converge on a template from what we learn trying to structure those 20:18:43 exactly 20:19:12 #action ttx and SpamapS to transform their proposals into active {somethings} 20:19:23 with that... 20:19:31 #topic Proposals for work 20:19:39 * Base Services - ttx 20:20:03 Waiting for it to be active to move to next step 20:20:26 unless people want more details 20:20:41 #info base-services is approved for active work 20:20:45 woohoo 20:20:45 how about that? 20:21:01 well, we need to move it physically to active/ 20:21:04 since you already have an action to convert it, we won't assign more redundant items 20:21:07 but I can propose that 20:21:12 but typically I think here we'd assign somebody to do the conversion 20:21:32 Yeah I think approved things can sit in proposal until the assigned person/team/etc. converts it to active 20:21:48 #action ttx to move base-services to active and complete the details 20:22:04 git kinda sucks for rename tracking though.. so we might want to do a rename-only commit just to keep the breadcrumbs. 20:22:17 will do 20:22:21 kk 20:22:29 * nova-compute-api - SpamapS 20:22:29 what I meant by "move" 20:22:34 ttx: indeed 20:22:58 hang on I'm going to just squawk in #openstack-dev about the fact that we're talking about this now 20:23:19 so this one represents more work, not sure one person can rnu wit hit 20:23:24 run with it 20:23:47 Agreed 20:23:55 while we are waiting, small reminder that the Arch WG day at the PTG will be on the Tuesday 20:24:10 The interest level on this one tells me it is worth it to at _least_ go through the act of documenting how it works now. 20:24:38 ttx: thanks. I just got my travel approval today, so I'll book soon. :) 20:24:41 * cdent is sad to miss the ptg 20:24:43 the way I see it, we can discuss the active proposals, then pump the backlog 20:25:20 cdent: miss by choice or necessity ? 20:25:23 ttx: I also think we should get matching Architecture-COP hats and ticket people for discussing deep new ideas in the hallways. (The ticket will be an invite to this meeting) ;) 20:25:36 ttx funding/timing 20:25:53 ok nobody's biting I think. :) 20:26:32 so this proposal.. the way I see it, the first action would be to go through all of the ways that compute hosts _currently_ interact with the rest of OpenStack, and document them with diagrams and basic descriptions. 20:26:37 SpamapS: the appropriate way is to drop the bomb on the keynote stage 20:26:48 I think I can play a fun role in this discussion: a regular nova contributor and a guy who thinks nova-compute should be its own repo 20:27:02 cdent: YES you are my best friend now. 20:27:33 the first person thinks: when in the universe will anyone ever find the time to do this 20:27:47 the second person thinks: more better faster sooner yes please now 20:28:00 so, once that documentation is completed, I would expect some problems to be readily apparent, including race conditions and circular dependencies in the system that make it hard to refactor anything. 20:30:27 I recall from BCN that some of the detractors suggested that this was all a lot harder than anyone was imagining so the documentation will definitely be interesting 20:30:47 Well yeah, we could just get 50% done and then give up. ;) 20:30:56 Though that, IMO, _is the problem_ 20:31:05 that it's so twisty and fraught with danger 20:31:17 I'm perfectly comfortable with this taking half a year to document. 20:31:38 Because I'd like to include interviewing folks from each project that has agents on the compute node 20:31:49 that's a good idea 20:31:52 so, Neutron, Cinder, drivers, Ironic, etc. 20:32:36 so I'll take the action to convert it into an active {thing} and in so doing I'll have this action item to begin with, and many more will spring from it: 20:32:37 btw, apparently ceilometer has given up on interacting with nova for its polling and now is going to do its own thing. I haven't got any details on that though, just heard about it couple days ago 20:33:12 (re "agents on the compute node") 20:33:26 * SpamapS enumerate every agent that runs on compute hosts. 20:33:43 cdent: good point about ceilometer 20:33:45 ok 20:34:05 #info nova-compute-api is approved for a move to active status 20:34:07 Agreed? 20:34:13 I should learn the bot 20:34:17 I think there's an #agree 20:34:37 there is 20:34:42 #action SpamapS to move nova-compute-api to active status and add structure 20:35:02 \o/ 20:35:10 #topic Open Discussion 20:35:33 One thing I also wonder about is how people feel about my email where I admonish people to please start these discussions on the mailing list. 20:35:51 a general question (maybe resolves into a proposal) but something i don't quite understand, is that if openstack has more private cloud usage now (vs public clouds), why would most of those private clouds want the complexity that neutron offers 20:35:58 Re: PTG, we could set up an etherpad, although the content will likely be pretty simple 20:36:07 My thinking there is that a 1 hour meeting every two weeks isn't going to be enough to fully discuss some of these things, and review comments are too pragmatic to allow free thought. 20:36:36 harlowja: Funny, I'd have put Neutron in the "things only private clouds would want" category. :) 20:36:37 SpamapS: could do some ad-hoc meetings for specific active topics 20:36:42 I agree that more email discussion would be grand, but there seem to be forces working against that 20:36:59 SpamapS maybe it goes into the question of 'who would want this in private or public' 20:36:59 the lack of free thought email has been my number one complaint about the openstack community from when I started (and remained so) 20:36:59 lol 20:37:09 SpamapS: I see the regular meeting as a checkpoint for progress, not where the work gets done so much 20:37:26 cdent: I believe the summits have contributed to that somewhat. 20:37:42 each group working on each active problem should schedule their own time on #openstack-architecture to make progress 20:38:02 I'll meet all day tomorrow with my 1.1-person team on base-services :) 20:38:06 ttx: Agreed, and that's a good idea, as a topic grows, we would definitely be served by saying "if you're working on this effort go to just this meeting" 20:38:13 SpamapS cdent i can ask the "who would want the complexity of neutron on the ML" though i'm not sure it will be helpful, lol 20:38:28 i don't mind getting flamed, lol 20:38:31 harlowja: no you SHOULD ask it 20:38:39 I actually think flames are part of the free thought 20:38:49 Neutron was conceived of a _long_ time ago. 20:38:58 With a different set of market forces. 20:38:59 harlowja: I think cloud like you have and want it is _much_ different from cloud how telcos want it? 20:39:11 It's entirely possible nobody will stand up and say they love it. 20:39:18 "I actually think flames are part of the free thought" <- /me saves that for later 20:39:28 fair nuff 20:39:37 i shall put on my flame suit and send something, lol 20:39:47 fwiw infra has seen a lot of resistsance around the idea that compute instances should just get IPs and talk to the internet 20:40:09 clarkb: well, that's like, the extreme conservative stance 20:40:11 so there are definitely people out there that want to micromanage every little networking detail, hopefully they chime in should you raise the question 20:40:56 Well I think most will be in that NFV space. 20:41:01 no 20:41:03 thats my point 20:41:13 infra isn't trying to do anything NFV we just want compute hosts with networking 20:41:21 and that is suprisingly difficult to accomplish 20:41:26 ya infra is like a good example imho 20:41:30 Right, I mean, most who want to micro-manage their L2's will be in NFV. 20:41:40 As in "not the kind of thing infra does at all" 20:41:55 SpamapS: right, but when we go to a cloud interested in contributing we say this is what we want l3 that works 20:42:00 they run away screaming half the time 20:42:12 because what they want to give are managed l2s with little holes punched 20:42:22 and we see that often outside of nfv 20:42:33 (just trying to point out its not an nfv specific thing) 20:43:06 Yeah, enterprises are probably equally weird for equally confusing compliance reasons. :-P 20:43:06 and hopefully they will talk about it publicly if asked 20:43:49 I also think Neutron isn't actually that complex, ml2 just conflates with Neutron and you get a double whammy of moderate complexity. 20:44:35 ya, i guess i just don't understand who would want to manage there own networks, there own subnets, there own ports ... 20:44:42 their 20:44:44 theirrrrrr 20:44:46 sorry 20:44:48 ther 20:44:48 twitch 20:44:49 lol 20:44:53 thur? ;) 20:44:55 le theire 20:44:56 thar 20:45:26 le lol 20:45:32 I think most of it boils down to legacy methods of isolation. 20:45:41 i mean, idk, maybe i'm just not normal person and most normal people want to manage all that, lol 20:46:14 By making the chatter between two things isolated at L2, they can believe that the traffic is "safe". 20:46:45 It's the soft-squishy-core of security model, but it has some merit over other legacy methods. 20:47:10 harlowja: well think about pre-cloud architecture of apps. 20:47:15 I think it's a bit of compromise with legacy networking teams. 20:47:32 "You can do cloud but I want l2 isolation" 20:47:42 or $feature 20:47:45 It was pretty common to have a public IPv4 on your load balancers, a separate L2 for frontend app servers, and yet another separate L2 for database servers and file storage. 20:47:47 SpamapS do i have to think pre-cloud, lol 20:48:36 so while we know you should secure all of those end-to-end, folks forklifting their apps out of datacenters onto cloud don't necessarily want to do both at the same time. 20:49:05 so should we start to advocate for all those things that really modernize peoples clouds ? 20:49:18 (which is why I see it as a primarily private cloud thing, because public clouds can be a little more pushy and homogenous) 20:49:24 and start to say ya this project will help u get there, but reallyyyy its a stop-gap 20:49:33 and ...?? is the better way going forward 20:49:42 harlowja: no, I think we should just accept Neutron and make it better. 20:50:30 but why not push for `secure all of those end-to-end` (for example) 20:50:31 and in 10 years when everybody has clouded everything, we can finally factor out L2 isolation. 20:50:44 the point is the cloud won't help you do that 20:51:18 You have an address, and it's reachable, and you shouldn't trust the source address alone of anything that reaches you. You should demand cryptographic proof. 20:52:13 and usually that's happening up in layer 7.. or _sometimes_ at layer 3. 20:52:28 anyway 20:52:30 fun topic 20:52:34 harlowja: worth a ML thread for sure 20:52:37 :-p 20:52:39 if you so desire 20:53:00 22.2% desire 20:53:04 it would be fun to watch if nothing else 20:53:07 :) 20:53:08 I'll give you a dollar if you do it 20:53:11 just to re-state ttx's point: Tuesday a the PTG is Arch-WG day.. please come join us for the fu-n. ;) 20:53:15 i aim to please cdent 20:53:22 a dollar 20:53:24 dooooo it 20:53:27 oh man 20:53:28 you can get a drink with that in ATL 20:53:35 peer pressure 20:53:50 Ok, see you all back here in two weeks 20:53:54 #endmeeting