Friday, 2018-07-06

*** edmondsw has joined #openstack-tc00:07
*** edmondsw has quit IRC00:12
*** annabelleB has quit IRC00:25
*** ricolin has joined #openstack-tc00:32
*** ianychoi_ is now known as ianychoi01:16
*** edmondsw has joined #openstack-tc01:55
*** edmondsw has quit IRC02:00
*** edmondsw has joined #openstack-tc03:43
*** edmondsw has quit IRC03:48
*** diablo_rojo has quit IRC04:02
*** flwang1 has quit IRC05:03
*** edmondsw has joined #openstack-tc05:31
*** edmondsw has quit IRC05:36
*** edmondsw has joined #openstack-tc07:20
*** edmondsw has quit IRC07:24
*** tosky has joined #openstack-tc07:36
*** jpich has joined #openstack-tc07:54
*** dtantsur|afk is now known as dtantsur08:57
*** edmondsw has joined #openstack-tc09:08
*** edmondsw has quit IRC09:12
*** cdent has joined #openstack-tc09:28
*** dtantsur is now known as dtantsur|brb09:55
*** aprice has quit IRC10:10
*** aprice has joined #openstack-tc10:10
*** ricolin has quit IRC10:24
*** srwilkers has quit IRC10:34
*** srwilkers has joined #openstack-tc10:34
*** edmondsw has joined #openstack-tc10:56
*** edmondsw has quit IRC11:00
*** cdent has quit IRC11:34
*** cdent has joined #openstack-tc11:38
*** edmondsw has joined #openstack-tc11:44
*** cdent has quit IRC11:48
*** cdent has joined #openstack-tc11:52
*** rosmaita has joined #openstack-tc12:00
mnaserHi everyone !12:56
smcginnisGood morning13:03
*** mriedem has joined #openstack-tc13:18
*** dtantsur|brb is now known as dtantsur13:29
*** hongbin has joined #openstack-tc14:08
*** annabelleB has joined #openstack-tc14:09
*** edmondsw has quit IRC14:23
*** edmondsw has joined #openstack-tc14:24
*** edmondsw has quit IRC14:27
*** edmondsw has joined #openstack-tc14:32
*** annabelleB has quit IRC14:38
*** purplerbot has quit IRC14:49
*** hongbin has quit IRC14:49
*** hongbin has joined #openstack-tc14:49
*** purplerbot has joined #openstack-tc14:49
*** annabelleB has joined #openstack-tc14:58
*** openstackgerrit has quit IRC15:19
mriedemhere you go tc people
mriedeman example of (1) what is openstack and (2) why is nova so mean15:21
mriedemas a callback to a discussion a couple of weeks ago15:21
mriedemjaypipes: zaneb: ^ specifically related to your conversation in the ML15:21
cdent_so mean_15:25
zanebmriedem: good example. so how can we make those requests go away, other than repeatedly telling people to go away?15:25
smcginnismriedem: It seems like that particular one should be redirected to Horizon.15:29
*** openstackgerrit has joined #openstack-tc15:30
openstackgerritChris Dent proposed openstack/project-team-guide master: Emphasize value of mailing lists
openstackgerritChris Dent proposed openstack/project-team-guide master: Replace out of date reference to cross project meeting
mriedemsmcginnis: or osc15:32
mriedemor something15:32
mriedemeveryting in openlab15:33
mriedemzaneb: idk15:33
mriedemzaneb: either always tell them no, or just bend over and do it15:33
mriedemor redirect to some other orchestration thingy that does it for them already15:33
zanebmriedem: I was thinking along the lines of a clear statement of what Nova is and is not, and how it fits in with other tools (e.g. orchestration)15:34
mriedemthat's what i linked into the spec review15:35
mriedemeven though we clearly bend that rule at times15:36
mriedemwith things like the get me a network feature in mitaka15:36
smcginnisWow, somehow I read that as "get me a network failure" at first. :D15:37
openstackgerritChris Dent proposed openstack/project-team-guide master: Emphasize value of mailing lists
openstackgerritChris Dent proposed openstack/project-team-guide master: Replace out of date reference to cross project meeting
zanebthat doc actually looks pretty good15:37
dansmithfwiw, we've been trying to say no to more orch like this for years, hence writing it down in that document15:45
*** mriedem is now known as mriedem_away15:45
dansmithnetwork ports have a lot more attributes like this, which we're kinda opening ourselves up to if we start doing more of this15:46
dansmithwe (openstack) needs to decide if the other projects are really intended to be used like we've been saying (create a volume, pass it to nova) or if nova is really going to be *the* api for everything up to a point15:46
zanebI'm sure a large part of the problem is that when you open the door to this (which happened years ago and can't be fixed) everyone is going to want to pile on their minor little enhancement15:47
dansmithbecause if we do this for volumes, I imagine people are going to ask why they have to go talk to neutron themselves...all they want is a port but with a few extra attributes....maaaan.15:47
smcginnisThere was also the proposal to have some other "openstack api" layer over the top of everything so it's not nova's api and cinder's api and neutron's api - consumers would just interact with "the API" and it would hide all the lower layer stuff.15:49
dansmitheither we're a bunch of related but distinct services, or we're not.15:49
clarkbpart of the complication with nova and neutron was that nova net "just worked" (tm)15:49
zanebsmcginnis: if we do that we definitely need to call it 'Corona'15:49
clarkband you lost that functionality as a user when clouds upgraded to neutron15:50
zanebclarkb: was that the same with Cinder vs. nova-volume too? (that part was before my time)15:51
dansmithclarkb: yeah, I'm not saying splitting services and apis was the right thing to do, I'm saying we need to decide at this point15:51
clarkbzaneb: I think that boot from volume was a feature added after the split so no feature parity concerns with that (but I'm not a fan of the boot from volume stuff so not sure if that was there early on)15:52
clarkbdansmith: ya I think it does help inform how we got to where we are today though. And deciding how to move forward is good idea15:52
cdentttx, smcginnis : a) thanks, b) that was too easy16:14
smcginniscdent: Thanks for getting that added/updated. As usually, docs don't ever get enough attention.16:15
openstackgerritMerged openstack/project-team-guide master: Emphasize value of mailing lists
openstackgerritMerged openstack/project-team-guide master: Replace out of date reference to cross project meeting
dhellmannmriedem_away , dansmith : thanks for those links. I agree that continuing to reiterate the message that we don't want to cram lots of orchestration features into nova may help. Maybe someone would even build those into a layer like oaktree (or a new thing as smcginnis & zaneb mention)16:26
dansmithI don't think it will actually help, but... yeah :)16:27
dhellmannmaybe describing what that orchestration layer might look like would encourage someone to start it and relieve some of the pressure16:27
dhellmanndansmith : well, stand strong in the mean time :-)16:27
dhellmannthe game is over and it's time for lunch, bbiab16:27
zanebdhellmann: we actually had to build that orchestration layer inside the OS::Nova::Server resource in Heat and it looks ****in' horrible16:28
*** annabelleB has quit IRC16:33
*** dtantsur is now known as dtantsur|afk16:37
*** ricolin has joined #openstack-tc16:39
*** jpich has quit IRC16:50
zanebI don't even wanna know how many thousand words that email I just sent was16:59
dhellmannzaneb : I wonder if extracting that stuff into a separate service of some sort makes sense. I haven't looked at the internal architecture of heat in ages.17:00
dhellmannmaybe heat just morphs into being the orchestration layer17:01
zanebdhellmann: maybe. I believe Rackspace did a bunch of stuff where they created Heat templates for common use cases and presented it to customers in the dashboard as if it was an API17:01
zanebdown to the point of hiding them from 'stack list' so the users wouldn't know17:03
dhellmannI wouldn't go that far, but having some pre-populated things makes sense17:03
*** diablo_rojo has joined #openstack-tc17:04
zanebto your other suggestion, I don't think splitting it into a separate service would help with Heat's "inner platform effect" problem. the issue is that an orchestration API needs to have access to everything that is going on. so whatever stealth orchestration Nova/Cinder/Neutron do under the covers we have to deal with/reimplement to handle updates correctly17:05
zanebe.g. we had to reimplement get-me-a-network to support get-me-a-network with updates17:05
zanebwe reimplement the thing of automagically creating a port so we can keep track of what ports there are17:06
zaneblike I said, horrible17:06
zanebdhellmann: it's one API call in neutron, so not as bad as it sounds, but we are in fact calling it from Heat and not passing it to Nova17:07
dhellmannI guess there's a good bit of difference between the need for something that manages a resources through it's lifecycle and a thing that orchestrates setting up some resources that are then left to be managed by something else17:07
*** smcginnis is now known as smcginnis_afk17:08
zanebyep, exactly17:08
clarkbright what we need is something that understands what useful aggregates of primitive resources look like17:10
clarkbthat a user facing orchestration tool could talk to17:10
*** mriedem_away is now known as mriedem17:10
dhellmannthat sounds a lot like a set of default stacks in heat17:10
zanebit really does, yeah17:10
dhellmannthe tricky part is describing that in a way that we can use for the interop program17:11
cdentzaneb: I just finished those however many thousands of words and had an interesting (to me) reaction: a) I enjoyed reading that, b) i have no idea what to do now17:11
cdentI think (b) is the curse of our current days17:12
clarkbfrom the network api perspective routers, subnets and networks are basicaully useless as individual resources17:12
clarkbas a user you need all three "wired" together to do anything useful17:12
zanebcdent: have a beer? it's Friday.17:12
* dhellmann slips a link to into the conversation17:12
clarkbI think the api you want is "Give me useable networking" and not "lookup a heat stack published by cloud by default"17:13
dhellmannright, we have some very low level APIs in some places and some higher level APIs in others and we're not consistent across services17:13
cdentdhellmann: yeah, it's a balance. the amount of color zaneb provided by using that many words is important to me. I have what feels like a higher fidelity context.17:13
cdentzaneb: soon on the beer17:14
dhellmannas annoying as neutron's low-level-ness can be, maybe that's actually the right approach17:14
zanebdhellmann: 256 pages :)17:14
dhellmanncdent : yeah, I haven't read the email17:14
cdentzaneb: in fact I think it's going to be fish and chips and beer on the beach17:15
zanebI might try turning it into a blog post. interleaving responses with jaypipes probably contributed a lot to the length17:15
cdentthat's a good idea17:15
zanebcdent: that's cruel when you know I'm stuck in a country without decent fish & chips17:15
dhellmannzaneb : does heat have a notion of stacks owned by one project but available publicly?17:17
dhellmannor is everything separated by tenants?17:17
zanebeverything is separated by tenant17:18
zanebunless you are a super-admin, then you can list everyone's stacks17:18
zanebbut you have to turn that on in policy.json17:18
cdenti've often wonder about shareable/reusable stacks17:18
dhellmannthat doesn't help with providing default stacks that are just part of the installation of heat itself17:19
cdentzaneb: some people like these fish and chips, say they are very good, but I don't like them. They are cooked in beef dripping, which is on trend, but I think makes for a brick in the belly17:19
zanebI think we need shareable *templates*, not stacks17:19
zanebyou can always throw those in Swift and make it public17:19
zanebGlare was going to make that easier, but... :(17:20
dhellmannI guess my idea is that after installing openstack there would be a thing ready to be used, and the user wouldn't have to do anything to import templates or whatever17:20
dhellmannthere are probably some implementation details of heat that I'm not familiar with that are leading me to use imprecise wording17:20
zanebso a stack is what you get after instantiating a template17:21
zanebso what you want is some sort of standardised template repo17:21
dhellmannmaybe? it still feels like that requires an extra step17:22
dhellmannI want to install heat and then call a heat API to instantiate a *server* in a way that requires orchestration17:22
dhellmannI don't want to have to give the cloud instructions for what that looks like17:22
dhellmannif we're going to move orchestration out of nova, it needs to be baked into something else17:22
dhellmanneven if that can be extended, a certain amount of minimal functionality needs to just be there17:23
clarkbdhellmann: ya that is what I meant earlire the api needs to be make me a network/VM/high level useable resource. Not lookup the 10 templates to put together to do that17:23
zanebI could see a thin wrapper service that emulates the current Nova API, assembles a Heat template to do all the behind-the-scenes bits, and then instantiates it in Heat17:23
dhellmannwhen we had that problem with routers at Dreamhost, we hooked into the account creation process to create some defaults in neutron for the user when they signed up. so I guess that's an option17:23
clarkbthe implementation behind the scenes might run those templates for you17:23
dhellmannclarkb : right17:23
dhellmannzaneb : I'm not even worried with emulation. Nova's existing API wouldn't go away for some time, so adding a new one doesn't necessarily need to bring any "baggage" with it17:24
dimslike kubevirt?
dhellmannI don't know, does that do orchestration?17:25
dimskubevirt talks to local libvirt17:25
dhellmannthat feels a little lower-level than what we're talking about, then17:26
dimsthe orchestrator is kubernetes though. for example volumes can be mounted -
dimsthis probably helps -
dhellmannI don't really want to throw everything we've done away. I want to consider how thinking differently about what responsibilities we place on each service makes things easier.17:28
clarkbthats also more of a nova net model aiui, you get what is prebuilt for you at the networking layer17:30
dhellmannSo saying that we want to stop adding orchestration to nova means we need somewhere else to do it. Not wanting every user to have to implement that means we want it in a reusable service of some sort. Heat is already doing a lot of orchestration work. Can we use that, somehow, and still provide an installation that has a useful minimal set of features without the user having to understand how to import templates to make17:30
dhellmann the cloud do something for them.17:30
dhellmannif not in heat, then is there another place that makes sense? or do we need a new place? and if we make a new place, how does heat also change as a result of that?17:30
zanebdhellmann: I believe that yes, you could write a pretty small service to do that. I don't believe it belongs in Heat, because it doesn't need access to anything other than Heat's public API17:31
dhellmanncdent : I'm finally getting around to reading
dhellmannusing oslo.config as a global is an anti-pattern, fwiw17:31
dhellmannit's just a very pervasive one17:31
dhellmannzaneb : sure. I guess your comments about having to re-implement existing orchestration features made me think we should look at it from the other way.17:32
*** ricolin has quit IRC17:35
zanebdhellmann: it's a good idea. we already reimplemented them once in Heat, so why not reuse them rather than do a third reimplementation17:35
dhellmannon the other hand, maybe we'd get it right the third time? ;-)17:35
cdentdhellmann: I can't remember if I wrote on that the context in which those notes were made: I was sitting outside with no electricity and noodling. It was actually pretty much random. So that oslo.config thing was a sort of blip of memory rather than a OMG I HATE OSLO_CONFIG17:36
zanebcdent: I can't stand all this talk of fish & chips. I'm going to get BBQ for lunch.17:36
cdent3rd times a charm17:36
cdentzaneb: good choice, that's something I can't get17:36
cdent(at least not _good_)17:36
dhellmanncdent : yeah, we have a global configopts instance mostly to avoid breaking the projects that had that before oslo.config was extracted from the incubator, but none of the other oslo libraries use it so you have to pass a conf object around anyway17:37
dhellmannthere's a whole thing about that in the oslo.config docs somewhere17:37
* dhellmann steps out to run an errand17:38
cdentdhellmann: if you have some more respones to that thing, I'd love to hear them (or see them written down somewhere). jaypipes you still planning to write something?17:41
*** e0ne has joined #openstack-tc17:44
dimsso fyi to folks, the next thing for k8s<->openstack will an implementation/provider for the "cluster-api"
dimswill be a new repo with initial code from
cdentdims: so this is clusters of k8s clusters?17:48
cdentso much going on17:49
cdentI think I better call it a day before my brain melts17:51
* cdent waves17:51
*** cdent has quit IRC17:51
*** e0ne has quit IRC17:56
jrollI feel the same as cdent about that email17:58
jrollzaneb: thanks for sending that17:58
jrollit very much solidified (for me) where you're coming from with many of the things you've been saying for years :)17:59
*** e0ne has joined #openstack-tc18:04
notmynamezaneb: IMO the key part of your email (holy wall of text, batman!) is the response to the "[nova's competition is] the compute provisioning parts of [public cliuds]"18:17
notmynamezaneb: I agree with your summary there. ("working at cross-purposes")18:19
*** e0ne has quit IRC18:24
fungizaneb: not to echo what others have said, but i really appreciated the porcelain/plumbing analogy ;)18:33
fungiand we actually have decent fish and chips in the part of the country where i reside, but that's mostly thanks to the ready availability of fresh fish _and_ fresh beer from the brewery who is in the back of the pub using it in their batter18:34
fungithough i ended up getting it on a bun instead18:35
clarkbfungi: I think zaneb is in the same part of the country as you18:35
fungi(i do recall getting a fair amount of terrible fish and chips when i lived in the mountains)18:36
fungi((on the other hand, bbq in the mountains was stellar))18:38
fungiclarkb: in the same state, but about 500mi inland and up into a mountain chain18:40
fungibasically where i grew up18:40
fungibefore i fled to find civilization, and then fled again to get away from civilization18:41
openstackgerritDoug Hellmann proposed openstack/governance master: write up the python3-first goal
openstackgerritMerged openstack/governance master: Add os-acc repository to the governance repository
openstackgerritMerged openstack/governance master: Remove release-tools from Release Management
openstackgerritDoug Hellmann proposed openstack/governance master: add
openstackgerritDoug Hellmann proposed openstack/governance master: write up the python3-first goal
zanebjroll: me too :)19:37
*** flwang1 has joined #openstack-tc19:39
zanebfungi: you can get decent pub fish & chips anywhere there is an Irish pub (so, anywhere), but nasty, cheap, greasy fish & chip shop fish & chips just doesn't exist in this country and I miss it so much19:45
openstackgerritDoug Hellmann proposed openstack/governance master: write up the python3-first goal
fungizaneb: oh, yes that really oily dense stuff19:46
fungii concur. i've only ever received it in certain parts of the british isles though i expect it has permeated other parts of the former commonwealth19:47
zanebindeed it has :)19:47
fungiironically, we have no irish pubs here, which is a strikingly odd situation. also no italian, greek or middle-eastern restaurants20:02
fungiunless i want to hit up one of the towns an hour or so away on the mainland that is20:03
zanebfungi: which island do you live on?20:04
fungidepends on who you ask, but the usgs labells it as bodie island20:05
fungikill devil hills is the township20:06
fungibasically the strip north of hatteras and pea islands, east of roanoke island20:06
zaneboh cool, yeah I have been there a couple of times20:08
fungiit's technically a peninsula in present day since the inlets to the north have shoaled over and so it's physically connected to the southeast corner of virginia, but that's not accessible by vehicle20:08
zanebwe have a friend whose family is from Ocracoke20:09
*** edmondsw_ has joined #openstack-tc20:09
fungilove that place. pain to get to but they've just opened a slightly faster pedestrian-only ferry as an alternative to the automotive ferry now20:10
fungidrops you off in the marina right in town as opposed to 10 miles away at the point20:10
zanebOne time I went to her wedding on Ocracoke on a Saturday night and then left on the 5am ferry to drive to Raleigh to get to Tokyo on Monday night for the Summit20:11
*** edmondsw has quit IRC20:11
zanebluckily the plane was empty and I had 3 full seats to lie down and sleep on20:12
fungithat helps. i rarely experience that on flights in more recent years20:12
fungiseems like i'm always on the sardine express any more20:13
zanebthat's the only time it's ever happened to me20:13
fungioh, also ocracoke has a brewery on the island as of less than a year ago. i haven't been down that way since it opened but intend to once the tourists recede20:15
fungigood opportunity to finally try out that fancy pedestrian ferry too20:19
zanebit leaves from Hatteras?20:20
fungiyep, same terminal ferry on the hatteras side, but the ocracoke side is where the swan quarter ferry lets off20:21
fungibasically the marina right in the middle of town20:21
zanebthat seems like a much better system :)20:22
fungiit's technically more distance on the water if you plot a dead straight course, but since i expect it rides higher with less weight it can clear some of the shallows to the sides of the inlet without swinging quite as far out into the sound, so probably actually less actual distance20:23
fungithe travel times are 15 minutes faster at any rate20:24
fungiplus you can just leave your car in hatteras, and it's not as if they charge to park there (not so far at least)20:25
fungii think they're just trying to curb (no pun intended) as much of the vehicular traffic on ocracoke as they can20:25
jaypipeszaneb: I specifically called out the mountains of North Carolina for a reason. :)20:33
fungii'm going to be up there a couple times over the next month. my favorite caribbean restaurant (outside the actual caribbean) is there. oh, and family20:33
fungistrange that i can get better caribbean food in the mountains than down here at the beach20:34
zanebfungi: is it Salsa's? It's Salsa's isn't it.20:35
* fungi laughs20:35
funginine mile20:35
fungithe one in montford. that new haywood rd location isn't rubbish but it lacks atmosphere20:35
fungigranted, i haven't been in the montford location since the renovations this year, so who knows20:36
zanebI should go to the one in Montford. only been to Haywood Rd and it did indeed lack atmosphere20:36
jaypipesnotmyname: I'm a little confused... if you look at Swift, would you ever think to say "you know, what we need to add to Swift is a filesystem layer so that our object storage system can look more familiar to application developers who want a filesystem"? I view the requests to have Nova be more than a low-level compute provisioning system in much the same way, which is why I push back hard on the scope of Nova expanding further than it already20:37
notmynamejaypipes: you mean like ;-)20:38
fungijaypipes: i've at times wanted posix-like filesystem features out of swift, albeit for entirely misguided reasons20:38
jaypipesnotmyname: that is your company's product.20:38
clarkbI think jaypipes is saying you wouldn't put it in swift itself20:38
jaypipesclarkb: bingo.20:38
notmynameit's an open source project that our company has written20:39
clarkbbuilding such a tool is fine, but out of scope for the low level thing20:39
jaypipesnotmyname: yes, I know. and there's nothing wrong with that at all.20:39
jaypipesnotmyname: I'm not saying "product" as if it's a bad thing. :)20:39
jaypipesnotmyname: I'm just pointing out that there's a reason you didn't just tack all that functionality into Swift itself.20:39
jaypipesnotmyname: because it would violate a sense of single responsibility of Swift's purpose.20:40
notmynameok. yes there are reasons (technical and otherwise) why proxyfs isn't in the swift codebase20:40
notmynamesorry, I've context shifted from whatever you were replying to me about, and I don't remember. what's the context? :-)20:40
jaypipesnotmyname: you were saying you agreed completely with zaneb's response to me about "[nova's competition is] the compute provisioning parts of [public cliuds]"20:41
zanebjaypipes: I look at Swift and think "why is Storlets a thing that is tied to Swift, instead of being a service that listens to a cross-project OpenStack event bus?" We could have invented Lambda first if we hadn't been all stuck in our silos :/20:41
notmynamejaypipes: ah, right. I agree with his statement that different groups in openstack have been building towards different goals all the while saying we're building the same thing. that leads to working at cross purposes20:42
notmynameand I found that to be the crux of the argument/response20:43
* fungi thought lisp invented lambda functions20:43
zanebjaypipes: the point of the email isn't asking for more porcelain in Nova. Nobody _here_ wants that (although there are plenty of people who misguidedly want that, Matt pointed out some this morning)20:44
notmynamejaypipes: your recent brief reply to zaneb about rancher is interesting. you point out that people refered to "openstack" and not "nova". but zaneb [I thought] was saying more generally that the two are largely equated and have been for years20:50
notmynamewhen reading these "what is openstack" email threads when they pop up every year, my opinion is that participants often use "openstack" and "nova" (or "compute provisioning") rather fluidly. or at least with a "openstack's primary purpose is compute provisioning" perspective20:53
notmynameI think that's a common perspective both inside openstack and in the wider tech community.20:54
*** zaneb has quit IRC21:00
*** diablo_rojo has quit IRC21:14
*** diablo_rojo has joined #openstack-tc21:23
*** rosmaita has quit IRC21:40
*** edmondsw_ has quit IRC21:53
*** mriedem has quit IRC22:12
*** hongbin has quit IRC22:25
*** edmondsw has joined #openstack-tc22:49
*** edmondsw has quit IRC22:54

Generated by 2.15.3 by Marius Gedminas - find it at!