OCI Weekly Discussion, 13 May 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OCI Weekly Discussion - 13 May 2020

Description

OCI Weekly developer's call from Wednesday, 13 May 2020.

Notes: https://hackmd.io/El8Dd2xrTlCaCG59ns5cwg#May-13-2020

A

Yeah feel free to address and the attendees from Mather I don't see Steve quite yet, and he's got an artifacts topic. Hey folks, hey welcome.

A

I see there's some Alexa topics, but given what time I know it is in Australia I assume we could discuss it, but I have a hard time believing he'd be here. Did.

B

He add those yeah. It's like this tutor.

C

B

C

Did a lot of people here he.

D

D

Taking the day off so in the spirit of taking a day off, I'm gonna, let it be.

C

Okay, this was his issue.

D

A

Alexi here no I was just mentioning. I would be surprised given the hour. It is in Australia.

B

If we can move that discussion to a different time, because I know Frew way wanted to attend as well and it's fruit, I mean he's in a similar time. Time constraint.

E

If Josh gets here, I know he's a good second format, because apparently Matt was asking those questions because Josh had asked them which ironically then basically came back to Josh yes yeah, so I did ask I, basically brought up in the helm. Call. Last week, I've been getting questions about, basically all the helm, OCI supporters behind an experimental flag, so people are kind of like and you're like antsy to have that not experimental and so I just brought up the question.

E

Why and they're basically when's the right time to lift that and Matt Farina was the one you had like the biggest opinion on it and then went to documented in those tickets and I sort of agree, but at the same time I think Jimmy said something last week that was great, it was like you know it doesn't change the fact that we're all using registries, even though distribution spec isn't one now so I'm wondering if we can, just as far as artifacts go just put a mini paragraph and distribution spec before it hits one.

E

Oh that just says you know this is a general purpose. Api that can work for anything. However, if you are looking for best practices, come see this repo, and that would point to artifacts.

C

I kind of got caught in it on both sides. Honestly, at one point, I was like look. This only affects a certain number of people, I'm looking to see if Chris's here, because Chris and I were on a slack conversation with Matt on this, and it was you know, one point is like look: there's only affects a couple people there Zach only you know, but several teams that are gonna create artifacts, eventually there'll, be lots more well, then by then we'll have more clarity.

C

If the members of this TOB and the maintainer Zimbabwe are all telling you like it's good like what really is the the blocker that you need, because this is a circle. Customers registry is like the ones we run here or obviously, probably more likely to open, but there's other ones. They were only gonna open it when there's enough artifacts to want to do it.

C

They'd only give me enough artifacts to do it if we can get them to take things out of experimental, there's an interesting bug with terraform that terraform can actually deploy helm, charts because that's things under experimental and then they're not willing to do the work also, so we want to break the loop. What's really blocking you doesn't seem like anything's blocking you. At the same token, there is some credibility to wait. It's really a handshake promise.

C

There's no handshakes these days, they're just virtual promises that we're not going to change anything even in conversations amongst Eric and me on these various calls. I can see we're out of sync on a couple of things only because things have changed over time. As we've learned, the Ayana process helped clarify some things in others.

C

So if we don't have clarity amongst ourselves, it is kind of hard for somebody externally to commit code that has a large public promise to a large community that we, the standard process, is something is versioned and you don't make breaking changes within a version is the implied promise so I'm kind of of the opinion. Is there a downside to saying that is aspect? There is a version.

C

It's you know it's dependent on these other versions, but there is a version to it and by having a version it says: okay, the media type form that we've identified that gets registered with Diana. If you want to guarantee your ownership of it, but the format of it isn't going to change and other other attributes that we've talked about in that are facts. Bec are more fixed.

B

Is that the only thing those is the format for media types, because I mean the ones that the artifact creators like I, guess from my perspective? It's it's a set of guidelines and we give suggestions for how these things should be laid out. But if it changes it doesn't break anybody like. If we change the guidelines, it doesn't mean that everything that was done before it is now incompatible. That's why I just don't quite understand what the what the compatibility issues are and I'm not saying that versioning is bad.

B

uh My my perspective was more about like like what is a 1.0 mean, like some things like guidelines are better just seen as kind of snapshots at specific dates like this is what the guidelines were for. You know this period of time, and then they changed on this at this time, not necessarily like this is 1.0. This is 1.1. Oh, we changed the format. Now it's 2.0 like I, guess I just have a hard time wrapping my head around it. No.

C

It's fair, there's a couple of things in here that are subtle: the bigger ones that will be more impactful or not yet, and but let's talk about what's in right now, right, it is the media types, the format of applications /or, it's actually referencing the ANA definitions. Then there is whether you use a config or not. There is some stuff in there that actually has evolved with some clarity that wasn't there before you actually can push an empty config and that's perfectly valid.

C

We just need a unique media type on that config because that's how the equivalent of the file extension there's some stuff in there around hey here's the guidance and it is guidance on how you split layer types, we've also put some stuff in there. That was the feedback again from this group on the extension of the layer, type and so forth. The piece that's not in there yet and why I wouldn't really declare this released.

C

One, oh is the well known stuff, although it won't be the quote: exact clearinghouse, it won't be an official Authority by us accepting a PR into the artifacts back for the helm. Definition that doesn't declare uniqueness what we've now deferred that to Ayane, but when somebody does want to tell other registries, hey here's what my logo looks like here's, the localized text rings, here's what else is in there what tools are associated with it? There was a whole bunch of things in that JSON file.

C

That is the next PR that I'll make I purposely wanted to keep those separate because it was too much intermixed, but that stuff will actually have specific, schema specific details to it, and we could say: maybe that's when we'll do a one. Oh and I was that's. Why I deferred in the PR? If you remember, I changed it from spec to artifacts authors, yeah,.

E

But like going back to what Derek said, what does it actually mean to be one out? Does this need versioning at all? It.

C

Will when you get those schemas, those schemas will have schemas that are unique to what it takes to publish one.

E

Okay, so those schemas are gonna, be more than just like an image like assets. It'll.

C

Be like JSON there's lots of different JSON formats that are heard and manifest for pets. I did at one point called it a publishers manifest and that definitely confuse people. So, while Mike and I had a long conversation with that, I didn't realize the the history of it so I renamed it.

C

In fact, I went just pulling it out because that'll be another PR that'll be specifically for publishing, well-known artifacts, so it I think it'll be much more clear later when they're, when it's needs to be version, because I I'm sure that those schemas will evolve. I'm sure we'll learn something and we'll need want to change them. It's not as clear now and until that's making an issue of this that I'm not trying to say Matt personally, but it's you know until one of our partners. That's trying to use this as trying to say hey.

C

How do I have some stability around this I was trying to defer it well.

B

I guess I guess I just want like a clearer picture of like because when, when I hear stability, I hear somebody's implementing flow of something and they're worried about that flow. Changing, except to be honest, like I, don't understand what this flow is. That's why I was trying to understand like okay he's talking about stability, which means he has something very specific in his mind and I don't have any idea. It wasn't clear to me exactly what that's referring to like. Yes, if you have an API, you have specific media types or schemas.

B

Then stability is important if you start writing code against that, but none of that stuff exists so I just can. I was just confused by that I guess. If.

E

I can summarize his thoughts I think it's I, don't want to speak for him, but I think it's mostly like small things like if we decide that the media type config media type is not the place that that's going to happen, that you're gonna keep of different artifacts.

E

If we've published things with him today next month, are we going to be able to download them? And you know, distribution going. One, oh I think is a separate concern, but it's more about. Are things going to change.

C

Think about how long we debated is the config object, the right thing to actually say this: what uniquely defines an artifact that, yes, we've finally narrowed, that we've published it? It is but right now it's a guidance. If we decide next week, we're gonna use some annotation or use something completely else, that kind of leaves on a tough spot because they have a released product helm that has imitation. That is stuck on a certain way. Well,.

E

It's almost like kind of to turn the thing back at mad. Well, helmets had this feature flag and the support for the longest of anything, basically maybe excluding like some of the scene AB stuff, but they should be the ones with the most feedback for us saying yeah if it's working for them or not right. So it's it's kind of like a bye deck bi-directional conversation. If we want to try to make some more guarantees, we're gonna need some feedback from from the home community determining whether that actually.

F

There's another.

E

Side to that, though, to registries not accepting home charts so like people cannot really use it right now, unless you're on a CR or I guess today, Harbor put out home support, but, like it's I, don't know, people are really excited about it and I. It's people really wanted and I think most people don't care with what part of the JSON it's on and that type of thing it's.

C

E

C

It is just if helm just does it, it will break through cuz. Now it's Bradley did it works and it'll drive its own demand, so I feel like I agree with most of folks here and like the current state of where things are I, really don't see what the blocker is and I put that into that.

C

A couple of times that said I think it will become more clear that we think we will wind up arguing that we do want a version on it and if it's we're blocking this circle, do we deal with it now and don't get into this argument anymore, because we're more than likely eventually do it anyway or do we just say you know, because at the end of the day, it's Josh that's doing all the development work at helm.

C

That, for this feature, so I think we're really trying to help him get this work in is the way I'm interpret. I was even on the call I'm just getting like these slack conversations. No.

E

It's it's not so much that and and other the community is actually coming in and doing OCI stuff now, but like did it's.

E

Can people when you download helm do this out of the box and today the answer is no because it's hidden and I don't know. There's there's a few other things like related to the helm, UX that I think make it experimental.

E

But I don't know it's tricky, I think I, think the answer to this is we're close to distribution. One! Oh.

E

If we can just put a little language in there that I don't know that like kind of signs off on the way helm is doing it. I don't know it's there's not really yet yeah I think Matt's looking for just direction from us and he wants to concrete guarantee from us which, in the strictest sense, distribution, spec isn't even 1.0 yet right. So we have some subtle changes that are gonna be made to the spec still.

E

So that was just kind of my counterpoint in that thread, but that's not to say that most of the registry is out here today, aren't paying their marketing team to go out and say they support OCI, even though there's no like finish specification for them to be compliant with, and most of them don't pass the compliance we'd. So things work today, even though not everyone is implementing the specifications to the strictest sense. So to that degree you could absolutely implement this stuff and start using it in depending on it, because people already are and.

C

I think if, as long as helm charts like hung charts in single a how much art is certainly a much broader surface area of people that are using it and since the singularity like there's a really deep community, but it's not as broad and I'm, not sure we're open, some of the other ones are yet and we saw they have seen apps just waiting for us to do the next round. But if we just got helm, unblocked and unblocked might just be like Josh Matt. What do use to you need that?

C

Can we just get this done, then the I think customers aren't going to come to AWS, Azure and others. Saying hey? Are you OCI compliant? Do they like hey? Do you support Hilton charts and that's what they care about? So we have the predominant number of people that are on this call or where we talk to on a regular basis. We've got the hard part's in place. I know, you know every pretty much. Everybody here is either on the cusp of getting the pieces done to get this working for them as well.

C

I guess I'm kind of putting back to you Josh like to some extent I at you, as the representative I I support. What he's asking for I, don't see it as a blocker. What.

B

I would like is kind of more specifics in terms of so I. Don't I still don't have a good understanding of where some of the failures are on different registries. Is it using kind of not sticking to the manifest types? Is that different registries are rejecting these different, manifest types Oh.

C

Doesn't work yes,.

B

Yeah, like like I, mean.

E

There's nothing wrong. There's.

B

Nothing wrong in the spec, but I know that registries are can be very firm about the actual content that shows up in the manifests and those media types. So yeah, that's that's kind of what I want understand from their perspective. What is the limiting factor? What what is it? What will it take for a registry and very specific terms, not just saying registries.

F

Warrants home or.

B

Doesn't support home but like when a manifest gets pushed up that has this type they're getting rejected correct so that we know exactly where to put the language in the spec, because yeah I mean I think we could have some general statement talking about it, but I think it'd be good. To put it very specifically in, like the manifest endpoint spec for endpoint.

C

The the right distribution spec specifically says you must support this, but it's open for others. What happened was all the registries had limited, limiting on what manifest types they supported? In fact, we were ready. We were confused with what we had an ACR for a while, and we wound up just opening it up. Joey wanted to keep it closed. We have this debate back and forth.

C

So now, I think we've got a good answer, so you're right to be this to be complying with the spec you actually and what we is the spec says you must support this.

C

You don't have to support others and we're gonna, say the same thing and artifacts, because there's two other mean documents I'm going to publish, there's one that says: here's how to publish a well-known artifact I'd, because those are the breadth of people that want to create these there's another document that will be for the people that are on this call and a couple others that are on different timezones. It says registry operators, here's the things that you would you would want to have to do.

C

You would do to enable it, but we're also gonna say that customers and registries may limit like I as a customer. I'm contoso I don't want anything, but this particular set of media types in this repo. That's this other repo and that'll be a perfectly valid thing so, but that'll be driven by the dev team, says I'm trying to push these things to a registry. Why isn't that open? Oh, let me go flip that and open it or that registry doesn't support it.

C

Hey I'll, open a ticket on clay, I'll pick on Jimmy for a second and they'll. You know add that up so the the pieces are just not there's no motivation yet to be completely compliant with the spec, because there's not enough artifacts that are complete to support this. If we can get a artifacts test in the conformance test, also by our Josh de Liske, how repo developer to then I think the pieces would be in place and they people just go. Oh yeah, that's all I need to do done. I.

E

Think the tricky thing, too, is that exact behavior on how you handle on, like a media type that you don't understand, was actually a change that had to be made on the image specification and not in the distributions back, because that actually surprised me when I open that pull request, that we actually.

C

Is not a split between the two years.

E

Yeah, and so that's probably, what is causing a lot of confusion is because this is now more important than it has been in the past, and hopefully this is a thing we can address when we're kind of reorganizing the spec, the distribution spec, because this is actually important now and we need to clarify this behavior yeah. The fact that it's in desperate places it was kind of strange. That's.

C

The factoring that I can open up the issue on refactoring, where the schemas are referenced versus. Basically, we've got distribution, we've got image spec and we have these two manifest schemas that happen to be living in the image. Spec yeah.

B

I, don't think us continuously screaming the image images. It can be anything it can be anything. It can be anything as enough because I feel like we're, but that approach isn't scaling. We tell one person at a time and then some like the next day. Somebody new comes up and says: oh shoot, I see the spec is only for images so clarifying that.

B

But actually this goes back to what you were saying about adding schemas to artifacts back to me, it almost doesn't make sense to do that unless we're going to try to fork off the index in manifests, expect from image spec into artifact, spec I know.

C

I say suppose: I say schemes that would be really careful, because this is the problem. I got myself until last time. There is a way for if I am the Marky Mark thing that the Josh did a while ago and I want to have a special artifact on then I have a special logo associated with it. Here's my localized strings. There is a JSON document that has a schema. That is nothing to do with a manifest. It looks nothing like a manifest. It looks like a JSON file format.

C

That's the thing that would be defined in this. It doesn't get pushed to a registry. It will figure out how it is. It's basically, just here's a way to define a set of things in code so that we can automate. These I was not trying to find yet another manifest index kind of schema. That would say.

B

Maybe maybe that's not such a bad idea, or for it to to go that way. What I'm saying is if, if artifacts were to be a speck, then the first thing we should do is split out the manifest and index back into it and then have the image. Spec just be the image artifact, which defines the image config as well as how to create an image artifacts.

C

F

C

Take it to them, it's really refactor I didn't get it cleaned up, I mean.

B

Yeah, it's a bigger change, but if that's the direction we have to go in, we could go in that direction. We don't we're not to do it overnight, but.

G

C

Hard to hear you Alexa you there my closer good good morning, by the way.

C

From yesterday, did you like wake up in the middle of your night to get on our call I.

G

Mean it's 7:00 in the morning: it's not that bad, but um the meaning link is actually wrong. It says it's 9:00 a.m. which is anyway yeah. Sorry. What does it say? Is that the this, the split between image, distribution, spec and an artifacts is a series of historical accidents. Slash you know, witnesses, slash politics and all the rest of it. So I would say: that's us trying to reemerge.

G

Some of those things is entirely natural and honestly, as where we should be, is how it should have it from the beginning whether that reemerging is either literally taking two projects together and pushing it together and merging them, or it means taking things that should be in a different project in moving them. So, for instance, when it comes to manifest I can definitely there is some slight complications about.

G

You know how you would implement an image spec tool if there is no manifest in the image spec, but on the other hand, I mean manifest is something which makes more sense in some senses to be in distribution. That's why Chris this is discussed would have to be taken with. Like you know, everyone sitting down shouting at each other, but uh I I think that I think that trying to undo some of these historical witnesses is definitely a good idea and I could be great.

G

It is confusing to a lot of people, especially when you say like Oh image. Spike was released and I feel like, oh, so how do I gone, but I was like well later know the distribution spike you and the distribution spectrum it's like. Well, how do you use it? Well, you know it is thought effects thing and it's reemerging. These things would be a good idea, because the ADA is.

E

um I feel like we've been going in circles for two years or something but maybe you're live.

A

E

I I think I have more information. I can talk to there's a the helm calls tomorrow. I can I can try to get like what Derrick's asking you know. What are the concrete things that they need from OC I try to go from there, but it's already at 4:30. So I don't know if we want to move.

C

On Alexis stuff, so what? Let's? Let's do some action, so that was a good conversation. You'll, take it up with the helm folks to see how they can get unblocked in the short term. We should queue up a conversation. It's obviously not going to be at Kubb con and in the same place, having fun drinking will be at home drinking. Let's queue up another caller, but how we want to refactor it we'll figure out how to get the Luxor mimosa first, so you can get up early in the morning.

C

It will shift times that sound like they're right we've been like 2 justice. Well, we have been talked about this for two years since we've been talking about this. It's been great. It's been this good evolution. Evolutions have weird things that grow out of them. It's been a really good place. You know, we've got Brian here from github.

C

Is there they've got some stuff going on there that'll be nice to leverage all this stuff as well, but there is a lack of clarity and that we I think the way Derrick said it made a lot of sense. We spend more time trying to explain to people how you tie these three things together. Then it would be baby just to clean up a couple of things that I think Josh had a PR where we're using the word artifact instead of image in the distribution spec and some that might help.

C

Forget it and we'll cue that up and hand off to Alexa.

C

I guess it'll actually.

G

Sure yeah so I have two topics: I'm gonna give them in reverse order, so that, because one is very, very minor, it's gonna make sure that it's mentioned. Is that so there is a Runcie issue which assignment we're dealing with with Iran, see I just thought: it'd be useful to everyone to know this. It turns out that OCI projects that actually been misusing semantic versioning when it comes to release candidates. So the- and this is like such a minor thing, but it's saying that we happen to hit because we're at least kit that's before use.

G

So, but basically the issue is: is that when you write the least candidates, super super minor thing is that the way that they will be written with, like the for semantic versioning, is like 1.0 rc3.

G

It turns out that this, actually that this rc3 string is actually compared lexicographically, even though it has a number in the way you would actually want to write. It is RC dot, three or RC dot, for which is actual trigger semver, um to treat the dot 3 as a numeric, which means that, for instance, RC 10, which of run C, which was released some time ago in January I, think, is actually treated to be a release before RC, and that is triggered as some issues with people who use, though modules.

G

So that's like super quick, something that people should be aware of a little working on the projects. It's um it's something which I definitely lost a little bit of a bit about this and yeah. We have to clean out try figure out a way to work around this because yeah and all the proposals whirring anyway and and one of the proposal is to just go past a version one, oh and then to go one over one aussie, but there that's um we'll figure something else, but yeah. That's what I wanted to say it is.

G

Some people were doing other specs and things just keep this in mind when you're doing releases because they fit us. Ok, though, luckily I mean luckily no of the specs are actually done, Aussie tens because that's ridiculous or it's not mainly alcohol. This happened. Okay, how's, the first topic, I'm sure that no one is going to come with around because mundane detail. The second thing was yeah OC, IV and setting up a working group.

G

So it's something which has been discussed a little bit, I mean obviously I've been I've, been championing some proposals. Other folks have been also Rosen things yeah there are. There are two things. The first thing is is that we should probably have at least some idea of how we would set up a working group, because, as far as the way, we don't really have those in our CI as well. As you know, who's interested who would like to have a chat about it. So yeah so, first of all does anyone has anyone set up?

G

A working group before I think is the first question I would ask. Is.

B

This facilitating.

H

G

Who you expect this is this is basically reworking the well. There are several things that would like to get done. One of them is replacing two archives with another format that has better better characteristics and then also at a loss and on the way a list before was discussing image deltas, and this tend to discussion of. Maybe we should have thought about how that should integrate into possible scary 2 as well, so.

H

I tried- and that was a maybe like at the end of last year, beginning of this year, a.

H

H

But I don't I do think that went anywhere. I, don't know what the mechanics of it on I understand that you have to submit a proposal somewhere, but only but I'm, not sure where so yeah it's big.

F

It's OC I, even big enough to need special working groups or are we in the call here? Basically, the working group.

E

F

E

I was basically in like that. It basically is a smaller working group, but we've just kind of co-opted the weekly meetings. I, don't think we should necessarily require creating a working group because that's a whole bunch of organizational overhead unless we find that these meetings are too full or like aren't representing all the people that actually care about what needs to be discussed.

C

If you use the notary v-2 stuff as an example like we, there was enough, we were using this call. Then we started doing some stuff. In slack, like we created a slack channel, we did have some separate meetings. We were reporting status back here.

C

We did create a repo because there is a different project, but in this case it's this project, so I think there's some stuff you could do but to jimmy's point it there's a lot of overhead.

C

I don't know if you really need, since most of the people are right here on this call anyway, so you probably have some other calls, maybe a little later of the day for you and others which I don't think many of us went back to, because you'll have much more detailed conversations with a focus set of people, but I think that generally works.

D

The link to the Charter into chat right now that it's silent on, like you know, working groups and all of that I think this is probably under developer committee, but it doesn't really seem like there's a lot of need to be able to put a lot of process around this. My miss hearing.

G

G

I mean yeah, we just say: hey: let's have a cool. You know a few hours later so that I don't sleep in then.

G

G

C

You see, we've been high on.

G

C

Agenda sometimes so feel free to dominate until other than the fact that the timezone might not work for you. The time might work a bit yeah.

I

I was just wondering like do. We need to have I feel like it's more technical discussion that would work better over email. First, you know.

A

Yeah I mean so just to follow up. You know. Clearly the OCI has no formalities around working groups, but the dev email list is almost always quiet and anyone's free to use it to start threads of discussion. It's just a google group, so you know archiving and all that and like Steve said I think you know it's clear that people who can just use common sense for you know if you want another call if you want to use a slack channel on the OCI slack team.

A

You know those are all open options where you know they're, not gonna, add a bunch of heavyweight process for people just to get together and have technical discussion. And you know this call is definitely free for use for Freddy.

H

Good luck just creating.

H

H

Over it, and if the comments get too large, then making a follow-up meeting.

A

Yep yeah, that's a perfect, perfect model as well.

A

G

I wasn't suggesting that we set up like like a whole set of an extra set of your same things. I was just yeah. Well, what is the recommended way of getting people together for stuff, like this cuz, at least for the specs, that I've worked with usually they're? The base document was imported from somewhere else. Right, so I mean runtimes. Backing image back were imported from from the world, and so it didn't really go through this kind of set up, but I guess I'm a for artifact, something like variable. So it was different.

G

So yeah I appreciate that well, yeah I'll just set up a shareholder in there. The thing with the manliest is that, while obviously I mean doing kind of mailing list, Silvan is like pretty standard. I, don't know if it's just that yeah I I find that technol discussions on on the that canoas devil is sort of like Peters out. Quite often, especially, can you two have discussion over like several months about about things but yeah also I'm chill doc. Man in that photo, but.

C

I know we've had debates around slack and so forth, but I think it's been working generally pretty well, so it unless there's objection to the people are actually in your core working group. I would suggest. Slack has worked out well for exactly the problems that email doesn't do well with any tract conversations and search and so forth, and that's.

I

Well, I, don't currently use flex on a daily basis, so it'd be great to avoid that I do, however, use IRC on a daily basis, but anyway, yeah I think we just I, don't know, there's like a lot of local discussion. That needs to happen right now. I think it's more. It's mostly just like. However, it is whether it's I don't know in a dock or on a mailing list or whatever I think that's sort of.

I

We need to write down what are people interested like the requirements once then decide on a set of requirements, and then we can figure out a design and blah blah blah.

E

Has there been a past any precedent for like gathering feedback like? Can we make polls where everybody says hey? What's the most painful thing that you found when implementing these things like who do you poll? Do you poll, the cult maintained errs, follow a different container run times like how do we actually get feedback on a lot of these things, I think is a pretty good general question across all the specs I.

G

Mean the way that it's always been done is sort of like making it have issue, but I think it wouldn't be a bad idea. If we said and I don't know, that's we're getting just people to clean it out or something is to say, hey we're gonna. Do a poll on this particular issue. Please participate if you're interested and then people who are really interested in are really invested. They can say: oh can I join the discussions for the actual respect over that's right, yeah.

A

I would just add that you know this is where the OSI is. You know has 40 some member companies, I mean we can Amy or Chris, can I'm sure get us communication to dev teams who have least expressed interest in the OCI and, like, like Alexa said you know, can tweet that out from the OCI channels or whatever. If we want to do something broader than just trying to figure out how to get the word out, what.

E

I think actually worked really well for the artifact stuff, and distribution in general is actually writing a blog post. So maybe, if you get like a fundamental a couple couple ideas like the ones you listed when I asked earlier kind of flesh those out a little bit and then run post about it and instead like this is how you get involved. I think that would probably get more more interest for sure.

D

Yeah sure with me, I have edit rights on the blog. So are you ready.

H

So another thing that I want to just point out is a lot of a lot of the unknowns that are surrounding this spec seems to be dependent on. You know whether thing actually would weather something that is no any idea would actually work or not so I'm thinking I'm to me.

H

You know the reason why I haven't like gone beyond that one time that I got Vincent and Steve together to try and figure out some stuff is because I'm still working on, like a proof of concept or something I'm still trying to learn and I would think that Alexa has the same issues, there's bound to be some time where there would be like radio silence from people who are involved.

H

I, don't know, that's how I feel I don't know if anyone else feels that way or if, if they don't feel that way, then instead are there any.

H

Are there any ways we can combat that or account for it.

G

And I think I think their winter little thing specs having implementations is important, is very important because it means we can. As you said, you can actually make sure the thing works properly. I think that when it when it comes to the the Lycos 82 proposals right now, all we really need is. We just need people to be speaking together and saying like having it written down in terms of what different people find important in in fixing what things they find that introspective deficient in and what features they look like and I.

G

Think that that's something which we can do without having long periods of silence, because long prison sounds in that context, means we stopped working on a physics Kant's, but yeah I. Think that once we have that written down, we start fleshing out some ideas, I'm sure that most of it we based on concepts and explaining. Oh, we found that this doesn't work like on.

G

This does look and then probably even comparing different proof concepts to getting a better idea of, because I don't know if we're gonna end up with like two or three different designs, that we have to pick the Queen. But if it happens having having prototypes with me, it would be good and that in that case you can end up waiting at least a couple weeks for to have like prototypes that actually like work and.

G

F

And we in the Loomis community have basically the same system in production today, so especially for this one I can provide a lot of feedback.

H

Well, I mean, for example, if you remember Alexa, you had you had the last time you were working on OCAP 2, or at least the last time. I heard it. The whole idea of how to include s poms was something you didn't know anything about or like you had.

H

No idea about, and I've been trying to figure out that part and I still do not really have consensus around how you would define the software components that you'll be putting inside the container image and how you would represent, how they connect other containers or other artifacts, that's still being flushed out.

H

So that's just an example of how long it takes for these things to come together. It's all giant web.

G

Yeah yeah I'm, no illusion this is gonna, take at least at least least a little while so I yeah, I think that if we, but but until it seems like like, I suppose, it's something which I mean during this caption will come up with, will end up coming.

F

G

On how much is going to be directly and better in the image? How many things might be? Necessarily, you know optional additions, you know, accessibility, these are all things. Are we gonna discuss as part of the discussion, yeah and I know that I know.

C

Sorry obviously, hopefully, a focus on oh well, a discussion Oh will we we should probably ship one. Oh.

G

There is a 1 lb respect that that did happen. Yeah yeah, run, run, see, is sort of that was stuff, I was about it, um so the thing about yeah and you noticed every two inches. It's really just a silly name, because it would do back with compatible anyway, because we're adding something extra rows and removing the old thing anyway. Lucy I mean one boy. One doesn't sound as sexy is the reason why I'm referring to those safety to actually.

C

Looks a there's some tangent of that, the stuff that Vincent has been working on until he like change jobs and got busy the extensibility stuff has a lot of potential. So I, don't know exactly what you're proposing I've seen a bunch of ideas.

C

It'd be interesting to know how much of that would be enabled through the extensibility model, and we drive that into the distribution spec and then you and a bunch of others are free to create other stuff that you don't need changes in the spec anymore, because the spec incorporated and extensibility most almost like your requirements. Could that drive the use case to validate this extensibility model design so.

G

Yes, though, I think that there is in theory we everything we could do with those heavy to, at least from from what I was doing, an imitation of the all the things that I was touching. They work or types within the respect, with things that in business well done to making your own media type I mean.

G

Actually the implementation that I have any mochi it makes use of extensibility supporting leaders are having a mochi which, basically like given a media type that isn't alerted a media type, use a different puzzle and, like that, all the code for doing that is all done. That way, so I think it, at least for the part that I have I, don't know how far we're gonna go with with once. We get everyone talking to each other and having having a broader discussion about what sort of thing to do.

G

I want, but assuming that at least the changes are in the same tangent like in the same trend, the extensibility model does work. The thing is: is that I think that, while they simply model does mean that arena, for instance, I mean Tigers mentioned before the baby, they don't even use terrifies. They use.

G

Excuse me, god, it's too early in the morning for this watch it that whatever.

G

You squash the fest for um four layers of layers or for the roof s and that's language speak sensibly. Model allows them to do because it's just two different media's I mean it's a bit dodgy, because the the layest stuff is a little bit. Let's take a little bit hairy in the in the image spec, but having changes like that is fine. The thing is, it's re. Having exams like that there are outside aspect is fine.

G

The thing is, is that for I think that if we want to get either wide adoption of something or if we want to actually have answers to some of the problems that people have with the image, SPECT I think it it deserves. We put in the interest like, even though technically it could be an extension that someone else wrote, but that's my view on it. Yeah I.

E

Don't suspect this will affect distribution very much at all, but yeah I definitely agree with that. You don't want to see fragmentation like you, don't want to make the spec so generic that nobody actually even knows what they're doing anymore and reading the spec is like. Basically, just noise.

E

C

Windows are in woman's test like I. Think the question would party this extensibility is how easy is it if, for a customer, to walk up and add some capability to a registry? If I think the problem will be if a bunch of these extensions require each of us to do coding that takes us time and prioritization and so forth and yeah? It becomes a very big tangled mess, because each rest could have different priorities: different customers and so forth.

C

If we can get to something like a kubernetes kind of CRT model where they just submit something and it just works, I don't know if that's even practical, but.

E

Mostly, what Alexis talking about are actually like the contents of the blobs that get uploaded at the registry, so this is not anything the registries care about really whatsoever. They don't need to know if it's squash, FS or anything else. They just know it's blob. They don't even need to know it's a tarball. To be honest, we do and for some validation in the media type and, like I'm sure certain registries are more restrictive than others right, but these are all things to run the image and not necessarily just to download it and upload it.

G

There is one exception, which is that the discussion of image deltas design is one. We will probably want to have a session with the distribution inspector balance, how they might going to support it or language level who they want to support it. But but the German bit, as you said, the core idea is- is just change just the place in the bulk of other globes, that the distribution spike really shouldn't care about. I.

B

Mean that could be one, that's good for discussions or the extensions because, like with deltas I mean it may make sense, I know Steven. They proposed this at some point that, if we're going to do Delta's, the best place to do deltas is in the registry itself, because trying to manage all of these pushing up all these Delta's just kind of floats the registry a little bit and makes puts quite a lot of burden on the on the builders to try to predict how clients are gonna, be pulling stuff.

G

Yeah I think there is anything a trade-off in terms of how house not the Reg needs to be and how much data has to keep from the discussion. Oh, what happened? You know that trade up discussion would happen if, once we all, we discuss the adding to the registry but abusing spec, but I would say that yeah I agree that it's not like, we should have like an entry in the image. Manifest saying here is the DIF blobs from these other versions.

G

I think that would be a mistake or a variety of reasons, but on the other hand, making it so that if we could make it to it, either users could say generate the last three deltas when they upload an image would be one option all while having the distribution spec. So the registry do it for you. Those both would be useful models like, for instance, the the the latter, the form of being be basically the lxd model, though it's actually done automatically. But basically you don't generate deltas for everything and you don't generate deltas dynamically.

G

You just generate a delta put the bar for the last three versions of the image so that when people pull it as long as they pull often enough, you get deltas. Otherwise, you get the whole thing that would be knit under get an example of like that's the way you can do it without having to make it so that the registry always generates deltas. So.

E

I'm not necessarily sure what exactly we're talking about, because a layer as it exists today is a tarball that has white out files or new versions of the file. That's basically a diff that basically is a delta and then with the mounting API is in the registry distribution like you, don't actually download all those things if you have permission to one of those blobs from another repository or it's another tag in the same repository it gets mounted and you don't end up pushing it. Pray.

G

This this would be this isn't fully from lay. This is for the face, lay of being changed, so this.

E

Would be like one tag referring to another tag, basically saying that it's just this extra set of layers on dice. It's.

B

Like if you update it for your base, you've been doin it for, for, like security reasons, right now, like.

F

B

A 200 megabyte you'd have to redownload all 200 megabytes, even though you know, 199 of those megabytes may actually be the same. But tar doesn't know that.

G

Yeah yeah that's mean, and the thing is is that in theory, the proposal I had, which was that we just make it so that the OCI image format is robin using opaque table gloves.

G

Instead, we use a tree, or rather, we basically convert all the files into individual blogs, or rather the chunks of all the files into do blobs, and then you have just like a single object that describes here is the set of all files and if you compress that the benefit of doing sand like that is that you can then say, you've n individual chunks, you only download the chunks that are changed.

G

The problem is, is that practically speaking, there is actually some evidence that even that system is not as ideal as having like binary gifts and so there's mentioning is. How should we just? How should we include binary gifts? If you want to include them but yeah? If we just had a different model for rather than tar archives, you can actually eliminate eliminate. The problem is duration,.

B

Yeah so yeah, that's why I was mentioning I think it's really important that we get this extensions and for the registry, because I mean I think what has really worked best for OCI is kind of you start off with some rough consensus through some group, then you gain some sort of user adoption and then OCI can standardize it, and we could certainly use like these kind of working groups to get a rough consensus and try to go out and get some user adoption.

B

But at that point like it's not really like an OCI standard.

E

B

So having something in the specification that allows for extensibility I think give some room for developing some of these features. I.

G

Mean design I think that just having some tested in a wild is definitely the way to go and making the spec before it's actually tested is putting a cop of the horse so I agree.

G

The only thing is is that- and this is again the chicken egg from a chicken chicken and egg problem, which is people not let's say refused adoption but they'll, so that we robot can place to adoption if it's not a standard, but you can't really standardize it unless it's been tested and then they get tested need to get people to adopt it, at least as an experimental thing, but I think that's how you can set up event if we had again if we make some use of some, obviously a blog post, saying that this is the process we're going to be using, and so therefore we asked people to, at the very least try the handed and invitations or in the bath too.

G

So we can get some better feedback in terms of how well the current design dog works. That's what I would suggest is.

F

A pre interesse previous back from Sun Microsystems, using this available so yeah. You can also pace with that. I.

B

Mean yeah you're gonna have to target at some point, I think we're saying getting getting the runtimes involved like like I know. This was something we wanted to do quite a while ago and can Tanner do you like I know a lot of this? A iv2 stuff is like Steffi price, with Stephen leicht like years ago that hasn't really gone anywhere. One of the reasons is yes. We're focused on stabilizing the client.

B

Another is a lot of those problems we tried to solve in like other ways that are more more fixed on the individual cluster level, rather than like kind of like time of first Pole. If that makes sense kind of things, you know what to do like peer to peer within a cluster rather than doing peer to peer across like the whole world or just having a different chunk mechanism, or something like that.

B

But that doesn't mean to say that, like we're, not interested in that, like I, think that's a great thing we could add in container D and then, if that's something, that's an option at those client levels and getting user adoption. I think is is pretty easy at that point, because then it's just telling people to go and able some flag on their kubernetes server, rather than trying to go out and and explain what it is to everybody.

E

Yeah I mean the introducing with that. Oh is our back, though right. How do you know who's supposed to have access to something as soon as you're sharing images across your kubernetes cluster? If you don't have some kind of global or back.

E

I'm also like questioned at what point doesn't make sense for some of the interfaces that are used for orchestration to be managed by something like OCI like. Why is CNI a CN CF project? Why isn't CNI a standard managed by OCI like it was originally developed for rocket, not even necessarily kubernetes so like it was for container runtime? So technically, it's not even necessarily for orchestration.

E

So I like there's a bunch of things and I feel like OC I could start to move towards in terms of like actually considering orchestration, because that is the most popular way the stuff is being consumed.

B

Or if we like, slow-moving api's takes Eri.

E

G

I was gonna say that regarding production, we had a cool the TV article yesterday discussing but working on making exact the exact terms of what production, what it actually means and what sort of projects we adopted. So I say you know. What's this space were working on at least agreeing on on works? What sort of projects should they accept? Those yeah? Because if you look at the Charter, the Charter is like it wasn't in 2015 like it. It refers to the wrong time and I mean technically. You know if we're gonna be.

G

If we're gonna be lawyers about it. Basically, every project off the image, spec technically actually shouldn't be a new CI. If you read the Charter, so I think it's language we're working on, because it's it's just outdated and no one had known status, sat down and updated, but yeah.

E

Yeah I was thinking that even with like what it would look like if we kind of did a rework right because work always guys working on the notary, v2 working group and it's like notary and tough- is a CN CF thing no see I thing, it's kind of like good who's. Actually in control of these things, who does actually make sense to own these things, yeah I think a lot of those needs to be thought about and discussed cross cross. Org I.

G

Mean speaking of Starkel witnesses, one of them would be the the how things I split. The few notes, the audience again CF, because it would miss, is language we are discussing anyway. Yeah still posted the piada, so there's a PR against the to be repo to explain, which is dumb people looking out to try to explain like what things should be should be including those projects but yeah, but I would say as opposed things in this discussion and the container world. It's all like I seriously started a lot of things.

G

What why things ended up with a did.

A

Yep absolutely.

A

So we're a little over time, so I assume to honor people's schedules will wrap up. We did not oh yeah, we missed an item. Josh. Is it okay? If we reset up for next ones, yeah.

E

So if you're concerned with the conformance stuff, just take a look at the link, I compiled real-world results from the major registries and saw a lot of discrepancies between what the spec says and what people are doing and there's an issue there that we're that's in the one that o milestone today.

E

That this is four so feel free. To comment on that on that github issue, we.

C

E

You for doing that, Josh Nora to Cass. Of course,.

C

So well, I have an engineer queued up to be able to catch up with the stuff that you've been working on. So apologies for the delay just been busy, but we do have that work schedule, yep, good or worse.

G

Great and I'll just say sorry, just to finish up the Tokyo saying over, say: yeah I'll send another man list with a link to a candy explaining what we just ghost people who learn the pole and then we can. They can go to great.

A

Yeah thanks Alexis all right well a good week we'll set up the agenda if anyone has something we missed or we want to come back around to and maybe I'll I'll give a quick update on T of the activities. Next Wednesday cuz I sort of meant to do that today, but we had plenty to discuss so talk to everyone next week. Thanks all.

G

Right packing thanks folks.