Cloud Native Computing Foundation TAG App Delivery / Bi-Weekly, 20 Sep 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: 2023-09-20 TAG General Meeting - kluctl, WG APIs

Description

TAG web site: https://tag-app-delivery.cncf.io/
TAG Slack channel: https://cloud-native.slack.com/archives/CL3SL0CP5
TAG git repo: https://github.com/cncf/tag-app-delivery
TAG meeting notes: https://docs.google.com/document/d/1OykvqvhSG4AxEdmDMXilrupsX2n1qCSJUWwTc3I7AOs/edit

A

uh So today uh is September 20th welcome everyone.

A

Yes, okay, I just wanted to restart the recording, because now we'll get it straight from here, so welcome to September 20th meeting everyone thanks for joining uh I posted the the chat, the the agenda in the chat. If you wanna check it out or add anything um yeah, we uh today I think the three main things that I'm tracking for the agenda today is talk a bit about kubecon plans.

A

um Alex is going to present on clue, cuddle, KLU cuddle, um some ideas that he's been developing there uh and then do. We have folks from the API extensions or enhancements group everyone here from that group.

A

Okay, so we were gonna, maybe talk about that, but if they're not here, we might have to defer that um we'd like to start, though, by letting folks that are new introduce themselves. uh You know, let us know your interest and basically we want to know how we can help you help us how we can help you contribute um and also just FYI. We do try to ask you know especially end users to come and tell us stories about app delivery at their organization. So.

B

If you want to do that.

A

um Here I'll open the floor for a minute or two.

C

So maybe I can start.

A

That'd be great thanks, so.

C

uh Yeah, this is mayur, so I'm from iro systems, uh which is India only based company. We are into Outsource uh product development for Enterprises and startups, so we are from I o flow team, the the open source project which we are working on yeah. We represent the I o flow platform team at iro.

A

Nice I'm just pasting that in so you mean the I o flow that submitted a Sandbox submission. Yes,.

C

A

Same oh sorry, I pasted the wrong thing. Let's paste on the one from Upstream.

A

Very cool um yeah, yeah I guess we didn't pencil you in for presentation yet, but it would be great for you know maybe next time to walk through what what you've got here and uh are you open to that.

C

A

All right, like um so I, think right now, the next uh the next meeting, although it says uh Kraken I'm, not sure if they're going to be presenting, then so we could pencil you in for that.

A

There's one I hadn't dead.

A

All right thanks, who else have we got here that wants to introduce.

A

um I don't like to call people out first because I, maybe you were here before and that's not fair, but also don't want to put pressure but yeah, let's jump in I'm pasting in here. Also you can put your names into in the dock, so we know who is here. Let me put my name in.

D

I'm not sure if I actually ever did an introduction, so I can I can do one real quick if you want yeah, go ahead, um yeah so uh I'm John from uh chromewear and uh uh my boss, Colin uh kind of told me about this and I jumped in kind of to observe and listen uh to the uh maturity model, uh work. That's being done and uh the nerd in me came out and I just kind of jumped into the conversation and so I've been working in that area so nice to meet you all.

A

C

A

Meet you yeah yeah nice to meet you yeah. We love Colin's work on the platform stuff.

A

um Actually, on that point, uh so we won't discuss the maturity model or the platforms or today, but uh a near diversion three or 0.3 is a is about to drop today literally I'm talking with Abby can I can I share with everyone in this call so she's like yeah, but we got to tell the.

B

Whole Ouija also.

A

B

A

End I'll share it.

A

um Yeah and I say I see yosuke. Thank you for joining us yosuke you're from the uh the API working group uh group.

E

So yeah that's correct and sorry I'm in a little bit of a noisy area, so um but yeah thanks for having me I want to talk about the aips. Maybe later this meeting.

A

Yeah, thank you for joining um um okay, so so moving on I guess the status updates. Unless I guess I'll pause for five more seconds. Anyone else wants to jump in and share a story or about themselves.

B

Have been off camera because I was trying to finish up from the last meeting, but um okay I'm Karina angel I'm, here again um I guess the announcement hasn't been made yet, but I mean um I'll, make the announcement, then um I'll be I'm. One of the new tech leads for the tag um so definitely excited to be working with all of you and I really like the work that I've already seen from yusufu's presentation on AIP, so I'm looking forward to hearing about that and then every all the other topics.

B

So all right, that's it.

A

Thank you, Karina. Thanks for coming off, mute and introducing and yeah that's about how it works in cncf, you uh you step up to do the work, and then you announce that you're doing the work and yeah. If you're here, then we're happy to have your support. So thank you um actually, okay, so dipping into status updates! Oh Studio, do you want to introduce yourself see? You came off a video Mew.

F

Oh yeah yeah! So sorry before a few minutes, uh you see you all were not Audible, so myself and I'm working with iro as a consultant. Okay, uh since uh last seven years so yeah.

C

So he he's from the same eye of floating he's with uh iof Team only.

A

Got it yeah thanks for joining.

A

You're also Joshua you're, with them yeah I wish I had penciled you into to present today and yeah, but I'm not sure if we're gonna have time so yeah not to be for next time.

A

Okay, so next on our agenda is status. Updates, I guess do you want to just maybe briefly uh inform everybody what's happening with the annual reviews, I. Think of the group here you're most in line with that. Maybe just take a minute.

B

Oops um yeah, so it was recent, as of actually yesterday for the annual reviews, they're all put on pause, um while the cncf staff figure out more automation. So it's not as heavy of a lift on each of the projects and then also um you know just rethinking the process.

B

um It doesn't have to be annual. If they're triggers that we can do um just keeping an eye on projects and then it triggers a review or if you're a Sandbox project and then go into incubation, then that's when a review is done. So those are things that are being thought through um and if you want to watch the TOC meeting from yesterday.

B

But if you have any questions about that and if you're wondering about your annual review, if you haven't gotten pinged by somebody yet then yours is probably just put on pause and you don't have to worry about it for right now.

F

A

Thanks yeah sounds great.

A

um Okay, next qcom plans, uh yeah I, know I, don't have a ton, I, don't know if the folks here are coming, but I'll just share broadly in case people are reviewing. Who here is coming to qcon I should just put that out there in Chicago.

A

Okay, Karina and me because that's where we work for a vendor that has a vested interest there, um all right. Well, uh we do have I'll just say very briefly: it's it's in it's linked in that meeting uh in that number 408, which is in the in the notes Here we will. We do have a project meeting space reserve for the pre-day.

A

um We do have a booth spot in the project Pavilion every afternoon, so we can talk about what our tag does and and projects too, like especially the longer tale of projects that aren't as well known. You can talk to us about using that booth and the folks here aren't coming, but um if you have folks on your team that want to give a presentation from that Booth about you know the long one of the projects related to have delivery, we're open to that um yeah.

A

Nothing else, huge I, guess on that right now, I want to share any yeah and I'll stop for five seconds in case anyone wants to talk about kubecon, more.

A

Cool um yeah, so let's jump in then to the project. Presentations, um Alex, we'd already talked so you're, let let's let you go first and take you know about 15 minutes, um we'll ask questions and stuff, and then, after that, we'll get next on. The agent is the API extensions proposal so we'll get to that uh next. If that's okay,.

G

H

G

So let me first share this screen.

G

Can you see my screen.

G

C

G

H

So, um just a very short introduction of myself: my name is Alexander block, nickname, codoblock and slacking everywhere, I'm, an open source developer and freelancer and I work a lot on my main project, close video or clue cattle which I'd like to present now.

H

So, um let's start with what is glue cattle clue, cattle is multiple things. Actually, it's a way to define deployment projects, no matter if simple, complex, small, large, it's all possible, and the idea is to make everything as easy as possible and boiler played free as possible. It's also a CLI that allows to deploy these projects.

H

It also has a git Ops controller to do it in a pool fashion and as of recently, the last release also contains a web UI to visualize everything and to give you some kind of visual control about your github's deployments.

H

So, uh let's start with deployment projects, a diploma project um is easy to set up and very cheap initially. So the idea is that, whatever you want to deploy it's as boilerplate free as possible, so you really have to only write down what you really need. For example, in this example, I have a very simple deployment project. Oh, what was that sorry defines the project kind of the entry point right now and the only thing that it does is it deploy.

H

It defines one deployment item which refers to this directory, and this is just a simple customized project having some yamls with some special uh with some differences to normal customize, because you can also do templating here, which I'll go to later.

H

The idea is that if you have that very simple project, you can just go into the directory, do a close CDL deploy tell it that the argument environment should be set to test, for example. Here you see it being used. Can you see my mouse pointer by the way good?

H

um So you would simply deploy this very simple project with this command line invocation. So this is a simple project. The idea is that you really start with such a simple setup and if you realize you need more features, for example, Target definitions in code which are kind of your list of possible environments targeting clusters targeting namespaces, whatever you like, you would send, for example, introduce inclusive yaml, which somehow it's switching back and forth and I. Don't know why.

G

H

Stopped here um so in this case, for example, we introduced the close dl.yaml, which has two Targets defined. One is named simple, one is named another and what you see here is the same that I previously passed as an argument, so I'm kind of pre-configuring how this deployment project is going to deploy it. uh It's the same as now, I think before that we use test here.

H

So in this case you would set environment to simple, which would then cause the templating here in the namespace that you've seen before, to deploy everything to a different namespace, um yeah I. Think that's explained. If you have questions uh just ping me in between and I can explain more um templating, as already mentioned briefly, is used as the glue between your deployments and deployment items.

H

The idea is that you have a um you, have different types of variable sources or configuration sources. For example, you have files, you have other git repositories where you can load configuration from. You can load stuff from an HTTP endpoint. You can use Vault, there's AWS, Secrets manager and so on different kinds, there's also Subs support, so you can encrypt your configuration files. Your yaml files with Subs same way as, for example, used in in flux.

H

In this example, we introduced two configuration files and another and and simple.yaml the deployment yaml that we have seen before can now load the variables, while already using templating, to figure out which configuration to actually load, for example, depending on the target name, we can now load a different configuration um depending on the target asset. So, for example, one is replica set to one one has replica set to two, and these configuration files are variable sources or variable files.

H

Whatever you call them, they contain arbitrary yaml, so you it can has have whatever you want and you can use it as you want in the templating. So there is no schema or something like that, you're completely free in what you want to do. What you can do um templating can really be used everywhere. That includes customization yaml. It includes M values, there's also hand support um so really just everywhere. You can even use templating in variable sources.

H

So let's say in this list of variable sources to load. You can have multiple of them, for example, if you would load two files, you could have one that loads some configuration and in another one already using that configuration. So you can layer configuration on top a common configuration, have overriding configuration and so on.

H

um Oh one thing I didn't describe before when targets are introduced, you use minus t to actually deploy that Target instead of giving the individual configuration arguments before. So you see that here as well. Okay, next slide um the deployment projects allowed to Define order and dependencies in a very natural way, as I feel. That means that we have seen before before that it had just one entry in the deployments list. Now it has multiple and the idea is that everything is deployed in parallel until it encounters a barrier.

H

So, for example, very often you cannot deploy your application into a namespace that hasn't been already deployed. So what you would do is you you tell it to deploy everything until this point in time. Wait for everything to be deployed and then continue with you see next things and so on.

H

This works in all levels involved in the deployment project. One thing I didn't show here is that deployments can also include sub deployments. That means, for example, it could have an include here that has another deployment with the same structure, also allowing to load variables allowing to include multi-planet and so on, and the barriers work with that together. So if you, for example, include a very complex base deployment sub deployment here, you could also wait for everything to be finished before you start deploying your applications.

H

um There are hooks hooks are comparable to what Helm does so after stuff is applied to the cluster. It will also apply the hooks and wait for Readiness of the hooks and do automatic deletion of the hooks afterwards, depending on the configuration how you used it in this example. It's a simple one after this deployment item, which means everything in this folder here is the Hang chart- is deployed, run this job and wait for it to to be ready.

H

Any questions so far.

H

Nope, okay, um next thing is closed, allows to do githubs. The idea inclusial is that githubs is just another interface to a deployment, so that means the project structure and the complete deployment is defined through the closer deployment project independent from your way of deploying it. That means you do not get dependent on on cluster components. For example, there is no, so you can use the controller, but you are not dependent on it.

H

You can still continue working in a push-based fashion, uh while everything else may be working in a pool based fashion, um so the idea is uh githubs is implemented through the inclusive controller, which provides inclusive deployment, crd custom resource definition. It implements the same interface as seen in the CLI, so the CLI has multiple Arguments. For example, you can say it's a dry run. It should not wait for for barriers of weight, Readiness instructions and so on, and all these arguments that you can give to the CLI are also configurable through the crd.

H

So it's really just an interface to what you already have used before.

H

The idea is that you can start with the push-based approach and figure out how to make your deployment work, which maybe requires like a hundred retries until it actually works as you expected, and then you can switch to githubs, you can even mix it. So, for example, you could have a prod deployment that runs po-based githubs and you could have a Dev environment where you push from your local CLI.

H

A

Can also mix I just had a question: I mean that that's pretty cool I have this situation all the time or like everyone get Ops, but like I, just want to push up my little change, but doesn't it reconcile like won't it just override like if I try to directly push something from a repo where I haven't committed? It won't the existing stuff, just wipe out my changes right away.

H

um It depends so what I describe now is when you have two different environments. For example, you have prod where you do strict uh github's pool, and you have a deaf environment just for playing around until everything works, and then you push it and then it goes to product. You can also mix it on the same environment. That's what you just described.

H

um The default setting for the closer deployment crd is to not override everything, but just to just detection. So it will tell you what has changed and then you are free to actually do the deployment.

H

If, however, something changes on the source side, for example, something gets pushed to get, this will definitely trigger deployment, and that might overwrite your changes, but only if it detects that it is done by a tool that is known to be used by humans, for example, inclusive, Cube, CDL, or if you did a local, closer your deployment or K9s and so on.

H

If it sees that a change was done by another operator, it will not touch it. It will just show you the the information that it has lost ownership of a field. So the idea is that yes, I know, the idea of githubs is git. Is the real or the only source of truth, but everyone realize after some time it's not really true, because there are so many factors involving um involved that might actually still change your uh object and so on.

H

So they kind of have to live together, losing my voice a little bit right now,.

H

So um the idea is really: you should always be able to switch. You should always be able to make it work together and very important, even if you don't want to mix it. For example, you have plot and you really follow very strict gitups pull based webflows. You can still do a diff against protein, so you can do your changes and run a div and a diff really means do a dry run, deploy and see what would have changed if you would have actually applied it.

H

So, at least you know that your change as an example. There are situations where you change something in a Helm value, a values file and it looks like a small change, but it has a Cascade of different changes resulting through the templating in the helm, chart and just reviewing that in gits might lead you to miss a lot of real changes.

H

That could happen so real dry run is um giving your more Security in that regard, so um need to read more see my voice, so here's a very simple example and the deployments that I've shown before. Let's assume we pushed it to a git Repository and we are using this simple with targets. Example.

H

So I'm not the one with the order, but you won before it's very simple, closely deployment manifest just saying the interval, then it should try to reconcile time out it's just a safety measure where is The Source located, which Target to deploy I'm, not forced to use targets here, I could also just pass ax instead, so I could work with our targets, but I don't have to then I have to provide the context but deploy it. The default context of C1, which points to the cluster that you're currently in you can enable pruning.

H

Oh yeah, also very important, which I didn't show in the project. um If you set it up properly, you can also do pruning. So if something gets deleted from your Source, it will tell you, there's an often object and you can prune it because then the DT objects, which also works from the CLI through githubs and later you will see through the eye as well.

H

So in this example, can you see everything because I have that pop up? Here we see.

G

Okay, now it's moved.

H

So yeah the um the status will show you. If something has drifted. This deployment is okay. If validation has succeeded and so on here, let's assume someone has changed the replicas, that's what I did in this example and deleted the service of the nginx example. You will see that um comparing the source to the actual State, we have one missing object.

H

One new object and one changed object and validation is failing, because the engine X is not up and running properly, um then you would force a redeployment which would then fix it or you could just change the source which would cause a reconciliation with the deployment right now. This is not implemented through the CLI, but that's going to be there in the next version.

H

You can still do it through the UI, for example.

H

Okay, next things, if you don't have any questions to that point, the next thing is.

C

A

Just then I'll hold I have a we'll save the question for the end.

H

Okay, so um next thing is the web UI, which got released in the last major release um this one. Yes, this is a simple screenshot, um so the web UI does not require an installation. That means you can just run it locally. It will connect to your current contacts and just read all the necessary information and visualize it for you. You can, of course, also install it, which gives you more advanced stuff, for example, IDC integration, um so that you have proper um at least authentication authorization is going to be implemented later.

H

Multicluster support is regimentary implemented. So, if you run it locally, you can give it multiple contexts and it will visualize multiple clusters. um It.

F

H

Show you everything that has been deployed through.

C

H

And also everything that has been deployed through your local CLI in locations, because both will write the necessary information into the cluster UI is still very early in that stage, with many improvements and features to come and I'm, also hoping that some people decide to contribute, of course, because I need some help. Sir um yeah next thing is demo time, because screenshots do not describe well what UI can do do I still have time actually didn't look at the clock. To be honest,.

A

Let's, let's give let's try to do another two or three minutes just to give enough time for us to talk the API stuff too, so yeah.

H

Great, so this is the basic UI assuming I have did. I did the closer get Ops deployment? Obviously simple: Target um I can do some simple stuff here, for example, cause a Reconciliation again Force the deployment I can cause a validation, can cause a pool. I can suspend it and so on. I can see past deployments. For example, here I see that this deployment happened uh one hour ago. It performs these changes, change replicas to two. A new object was introduced.

H

That's when I fixed seat drift by the way, so the service was recreated and the replicas was fixed um information about the Target and so on, and the ideal 3D that, um let's assume uh I, don't know, I think it was this one. So let's say I changed something here. Let's add enable a label test and change replica, so I said we can use templating. That means we can also do some Expressions here. Let's just do that for fun.

G

And deploy it oops.

G

Okay, I'm in the wrong folder, and what, from now I'm still in the wrong folder.

H

For information, I prepared everything like an hour before the meeting started so I'm doing a local deployment. Now, um as I said, we can mix push and pull. You can now see that it asked me if I'm fine, with the changes that are going to be applied. For example, I added the label, replicas have been changed, I say yes, I'm, fine with that and it's showing another div which actually tells me what has happened. So it will always tell you what will happen.

H

You say yes or no, and then it tells you what really happened very important. This div is based on a um on a real apply, so it's not just manifests being compared. It's real objects being compared all the new objects.

H

If I look at the UI I see there's a new card now which says Hey. Something has been deployed through the command line 45 seconds ago and I can see here again information about the targets. What has actually changed? I can go into detail mode, which will even give me information about which variables were loaded which uh manifests were involved and so on.

H

Then the GitHub space deployment will now tell me how wait a moment there's a drift that just happened by the way this card is the target. This card is the result.

H

um I said many improvements are going to come to make all this clearer, and now we see there's the drift. This label would be removed if we would redeploy and replicas would be set back to two.

H

um So let's say we want to fix that Rift and it's fixed now the controller did a redeployment and so on. um Another thing that I find pretty cool is there's also a Mode called manual for the deployments.

H

So if a deployment is set to manual, it will always just shows the drift and even a source code changes. It will not deploy to just show the drift and ask for approval. So let's say I actually commit all the changes that we just did push it.

H

Reconciliation currently happens every minute to speed it up, I'm, going to force a Reconciliation now and what you see now is there's a drift and even though the source changed it did not deploy it. It now tells us that we can trigger the manual deployments by approving this source, so we can check the drift. Okay, looks fine and tell it to deploy it and there. It is now it's unhealthy because replicas have changed and it needs some time until the cluster has actually reconciled everything.

H

So, let's fast a Reconciliation of CD deployment to take a few seconds until my local kind cluster is actually ready.

A

So Alex um are you getting? Is this kind of towards the the end here, because I have a question that I want to talk about while you're- and that is this is very cool? Stuff uh I mean this. This fits the Paradigm of like Get, Ops and and template rendering and things that we see in a lot.

A

um I guess kind of my question would be this reminds me a lot of like Argo CD I'm, more familiar with that than flux it's probably similar to where is it differentiate it like is it provides something that they don't do? Is there something? Is there a particular value here that you know stands out.

H

um I think the value is that you don't have to go full in from day one, um because you never get dependent on the operator. You never get dependent on anything that runs on the cluster. You can start simple and build more and more into it.

H

um That's one thing, also argosity I'm not very much familiar with other CD mobile flux as well, but I think it's kind of the same. You are working a lot with the custom resources so to have a more complex project set up. You would have to use custom resources that, for example, do a customized deployment that then does more custom resources. It does more customizable and so on.

H

Also when it comes to order and dependency I think it's better solved in or easy, not better, but easier in other cities and flux, for example, but still I believe that this approach, where, for example, let's go to.

G

The orders I think that this is.

H

At least for me, it feels much more natural to deploy it to to Define it this way, because it's a just a tree structure, just a folder structure and um I think it's much easier to work with this compared to, for example, dependencies that you have in flux, or these I think sync waves in other City, so yeah.

A

Can we call it they don't know? Is that would be I'm just eyeing time because I want to get to this one Alex? Are you pretty comfortable with what you've shared so far, I think we conquer right now to ask.

H

You mean if I'm done with presenting or yeah.

A

H

I'm done, okay,.

A

B

You um I had a follow-on question to what you're just talking about for simple deployments. What do you have a path to more complex deployments? Do you have a path to Argo, CD or flux, or are you planning on more complex on supporting more complex deployments through your.

H

Project, um it is already supported um so with what you've seen right now, you can already do a lot because there are also, for example, git includes you can include sub deployment, and you can include sub deployments from other git projects which allows you to build quite complex um stuff.

H

There's a pretty well working Helm integration, so you can reuse just everything that you find through Helm same as customize, um so I I've seen some very complex projects already and it's, at least in my opinion and I'm biased. Of course, it's nice to work with these complex projects and I cannot imagine how I would Implement them with fluxo agricity and how it would still be that easy to work with some but I said: I'm, biased and I. Don't have so much experience with argosidian flux,.

A

Would almost say you should talk to the git Ops working group. I, don't know, though, if they have a consistent thing, but it would be interesting to hear their take on this, and you know what what you all could learn from each other.

F

A

um Yeah, but in the absence of that, let's we've got this all written up: um yeah, I I, don't think we have any. You know you're not submitting it now anywhere. So we're just kind of sharing this info along people want to contribute. Give us you know, you're, sharing your deals with us, so thank you.

F

A

All right so I, don't Let's uh yusuke. Let's switch over to the discussion about. Thank you again. Alex. Let's switch over the discussion about an api's working group. um I am thinking. If you are okay with it yusuke I'll. Let you kind of introduce what you're thinking of doing and then we can discuss like things that have come up. Is that all right with you.

E

Yep uh I think that sounds good and again sorry, the noise and, let me know if I'm, just in audible but I'll, try my best so um okay, so I think um just as a quick intro there's a tag, app delivery issue, that kind of summarizes I think what we're trying to get accomplished. um But, oh okay.

E

Okay, now that's shocking to me because it's really noisy to me in this coffee shop uh but I'm glad I'm doing a good job, so um yeah thanks Josh for for posting the issue so um basically um just some background.

E

um I used to be the API design TL for Google um and in particular, I was working really closely with Google Cloud to kind of solve some of their problems with resource oriented, API design. I. Think that's! You know pretty important to Cloud native Technologies, because often these declarative providers like terraform and whatnot are um you know, require some sort of resource model to operate against and um there's this open source AIP group. So aips are API Improvement proposals. They are generally specific to Google, but there's also sort of an open source variant as well.

E

So you know um if we want I can try to. Let's see I can try to share my screen and let me know that just doesn't work.

E

um Okay, it's not letting it's giving me a bunch of explanation points, so it's probably not gonna. Let me I'll just post links in for now. um So if you look at those, uh it includes a lot of the AIP link I just sent. It includes a lot of information about the aips sort of a resource where the design looks like at Google and I.

E

Guess one side note here in particular, it does describe everything primarily in protobuf, um because Google speaks in front of that as well, but I think that um you know in general, these principles apply to sort of any sort of transport protocol, including HTTP and Json, which most I think clouds and services exposed.

E

um So uh basically there's this open source, AIP group um and our goal is to try to basically take a lot of the learnings from you. So sorry, I did the API design and try to maybe put them in more of an agnostic organization.

E

um Those who are involved right now are coming from Roblox IBM, Netflix, Microsoft and I've started to talk a little bit to um palumi and uh cross plain, as well as possible clients of apis that adhere to the specifications.

E

um So you know, if you, if you look at the aips, it kind of includes the general structure that we're hoping for, but we don't have any any tangible project that we feel like we're ready to submit yet, but I think one of the it's kind of a chicken and egg in our in our minds. We wanted to see if we can establish the right place to kind of um find the people who want to do this work and be involved, um and then once we do that we can.

E

You know you know how much I think better discussion around what that API I. Think specification is what we're going for at this point here.

E

um That's one sort of major difference is the aips were primarily kind of a style guide, their best effort and the feedback that I've heard from you know, both internally and externally. Is that um you know style guide isn't enough. It needs to be some sort of specification because people aren't going to basically go right. You know some provider against what they hope.

E

An API is going to be like in terms of its ability to support, create, read, updates, delete list operations as well, as you know, very the various sort of um behaviors that we're expecting around things like you know the pagination behavior for list apis, and um you know whether you fields are used edible or not, so so that's kind of the the gist of it.

E

You know I'm happy to go ahead and you know have deeper side conversations discuss more about the API design itself, but I think primarily the discussion that we've been having with with um Josh and the folks in this room has been sort of you know. How can we get some structure around this and can we get some support to just like have discussions and find the right people involved and work towards? You know some tangible unit of code.

E

um I do have some codes to show if people want to see it, but um you know I think it's a lot less important than I. Think the you know finding the right place personally.

A

F

A

Thank you for the background music.

A

um For for me, like one of the questions and you started to address, it is like, when I think of these working groups, it's to it's to kind of bring together a few of our projects, to kind of facilitate greater synergies, and you know make it easier for end users to like succeed with this stuff.

A

um So I I, like that, you called out like cross-plane and pulumi um I, would throw in their operator framework. They haven't worked for the same company as me like anybody, that's producing or k native, even they're they're, designing crds. You know I, always think of them, because they're duck types and I think that might apply.

A

um Yeah I think that it's a big step to know that those folks will come and participate. I think that that even that helps me understand a real value that could come from your group. If you guys could come up with some standards for crds kubernetes apis or something like that. That would I could see that being really beneficial to the industry.

E

Yeah and I mean I think just to expand on that. A little bit too I mean I, think there's a lot. We could say about maybe best practices for crds themselves and trying to align them. But um you know I think a common design pattern now is that crds and the resources that they are representing are actually sort of mirroring like a cloud API right. So this is kind of like what cross plane does or you might want to just go ahead and Mirror to some sort of internal SPF resource.

F

E

And I think that if you look at the Practical work to make that happen, it tends depending on the design of the API, it tends to be very expensive to integrate, um and so uh you know fundamentally, the same problem stuff, for example, terraform runs into are very similar to the types of problems that um kubernetes crd integration runs into.

E

You know if something doesn't have a create operation, you have to go Stitch that together from scratch, if, if update, isn't literally just sending the resource, it's actually calling five or six different things different apis and like orchestrating them appropriately. All of those are problems that I think they get pushed to the clients, um and so yeah I would love to see kind of how far we can. We can take this, but definitely I, think crds and and operators in general are are beneficiary.

A

That's a really good Insight I, really like that that yeah, a lot of these things are just projecting provider apis, and is that really what they should be doing? Are there? You know more.

B

Efficient ways.

A

F

Good experience for users.

E

Yeah so I mean I, think um you know these conversations are going on. We have this sort of like completely separate from the cncf AIP weekly meeting, where we kind of talk about these problems. I've also engaged in some discussions, like I, said with lumion and cross-plane, but I think you know everyone would be much happier if there was some sort of you know.

E

Frankly, like cncf sponsored place where we can have these conversations and feel like we kind of get the support, and you know we can reach out to you know I love you you know, and then we can get your feedback um sort of a place to make. This feel a little bit more official, if that, if that kind of makes sense, um so.

E

So yeah I'm kind of curious. Maybe you know if there's interest and we believe, there's there's value here: I, don't I've been trying to figure out the right structure for us to fit into. You know, I think the um this tag as well as the cncf overall, and if there's some guidance on what to do next, I'm happy to you know, rewrite docs or whatever. That needs to be done, I think. As long as you know, we get to some place where we can say you know.

E

If someone says hey I'm interested in this area, we can say Hey, you know, here's all the resources, here's you know some level of structure, here's the the meetings to join um and have some sort of I mean again frankly, cncf. You know affiliate stamp to make people to also help convince their organizations that this is actually worth contributing to I. Think it's! Another problem is right. Now it's sort of a ragtab group of of Engineers.

E

You know, um despite the fact that we all come from different organizations and we're officially representing them in the same vein. So the feedback always comes back is like sort of you know what what larger open source group or organization are you? Actually, you know working with. Are you working with cncf? Are you working with you know, open API um and by extension, Linux Foundation yeah? We don't have that backing per se.

A

Yeah yeah um yeah, that makes a lot of sense. I mean I I, like I'm thinking about like how we formed artifacts and basically, we had five or six interested parties from different projects.

A

So, like someone from Docker someone from oci someone from zoc, which is a project we have from from the oras project, um someone from ortalius even on the uh from the CDF side, um and we almost even as the Ouija was getting formed, we were even able to have the first meetings and we had you know like 10 or 12 folks contributing what they think the use cases that are going to be solved and and how to solve them.

A

um So yeah I think that you know it sounds like you're already on that path, with reaching out to like blooming and cross-plane I. Think that that would go a long way to kind of say: we've got these five or six related projects that would that are interesting. That are going to come to the meeting.

A

I think that's kind of key, um then there's the greater question of of where it fits uh uh and yeah yeah Karina. You might have thoughts on that, so I'll. Let you uh that that's kind of where I'm at now. It's like this. This thing is valuable: where does it fit.

B

Okay, you know um so we've been having side conversations about it, obviously um and I put in the issue what my my perspective is- um and we've also been talking about whether um it makes sense to start it as a working group within the tag and then, as it gains momentum, it could be moved somewhere else.

B

If that's, it makes more sense, but at least for a starting point and we'll have to I mean you know, get consensus within the tag, but um I think that it should be started as a working group within the tag.

B

um And then we talked about removing the tooling specific items and then, as tooling um as the spec is written through the working group, The tooling, as it meets the spec, they could for each tool look at um applying to sandbox, and then we go through that process.

B

um So that's kind of how I see it, but again it needs consensus. And um what do you think Josh.

A

Yeah I lean in that direction. Also, like you guys, have something that you are ready to start. uh It applies to a lot of our projects um better to start in a tag than somewhere out on your own. Like you know, you can stay connected so yeah, like kind of like air on the side like like, let's go ahead and do this get the work going and worse guns, the worst. We shifted to the open, API.

F

Foundation next.

A

Year or something if that's where it really goes,.

B

But it doesn't have to be a set in stone kind of thing: um it's not donating a project, yet um it's a working group. um It does it like. You said it brings others in that normally wouldn't and you're bringing them with you, and we know other people that you can reach out to and bring in for the working group, and hopefully they would.

B

Because they're working in kcp he's another, which was just admitted to sandbox, which is awesome, but you know there are more people you can bring in and um I would like to see the tag support this.

A

You know one thing: I wonder if it would help if we emphasized infrastructure apis in your I guess, I'm I don't know. Maybe now already you do it sounded more like you were just general service apis, but in in our context here, infrastructure repairs are most relevant yeah. What do you think of that? What if we, what if we focused on that in the proposal.

E

Yeah I think that's fine um by the way I like that structure as well sort of the working group and then we'll figure out what projects to kind of spend out of that. So um if the uh it sounds like what's keeping that from happening is, is this uh Charter, Dock and kind of cleaning it up to better align with what we discussed um is so is that true, okay, cool, okay,.

A

E

So I, what we're saying, okay got it so I mean, if that's the case, yeah I'm totally happy to kind of fix that up a little bit more.

F

E

I think, um although I will no, you know, I think the the goals and expected outcomes will probably be similar, which is cultivate and maintain a series of cncf OSS projects.

E

um So um is there, aside from making it I, think focused more on infrasture apis which I'm happy to do? Are there other things, I shouldn't change.

B

I was going to say, I'd, be careful with saying that the goal is to cultivate the projects.

B

um That's a great end goal and I would love to see that, but at least as far as the working group more focus on the spec itself and standardization.

B

um What were you going to say? Josh.

A

um I was actually going to kind of say it might be good to kind of know who the projects that are dotted line related I, guess it doesn't have to go in the in the actual proposal. I guess to your point, Karina that might be a distraction, but maybe a list or something in the issue or something and just them because just to know like Stefan, let's say if we ask him to to contribute like is he like? Okay, yeah I want to help. He probably is actually I can reach out to him um yeah.

A

Just just so. We can kind of see again I'm kind of thinking with the artifacts. We, you know what you've got. The Ouija you've got that uh Channel already in slack, so maybe we can start inviting folks to that like that would be an indicator too like if we can shift the conversation to stuff, that's relevant to them or.

B

Maybe it already is.

A

But I'm saying have those conversations.

B

A

Guess another Next Step that I'm saying is: maybe we can start bringing them into your channel too and showing some evidence that yeah they they are interested in this.

B

um And I would love to see the follow-up from this go soon, but um all right, thank you. Yusuku and Alex.

E

Okay, thanks I have a couple other questions, so maybe I can just reach out to you offline or something like that. um Okay, but but the high level idea sounds good and thanks for the time.

A

E

If you want I mean I'm still here.

A

For another minute, if you, if you want to discuss further, you know but yeah, it definitely hit us up on Slack.

E

Okay, sounds good I'll, probably ask about this as well, but I think my well. One of my questions was you know, um uh I guess who so I can add people to this to the slack room. You know what I've noticed is these asynchronous conversations don't aren't nearly as productive as sort of like having an hourly working group where we can actually discuss some of the problems in depth um around the design of those?

E

So you know if the goal is to basically Loop someone in and say Hey you know, are you interested in contributing and they say yes and I think we can accomplish that. But you know um is that kind of all you're looking for it's just a list of people who say yes, I would you know contribute time to this? If it was a thing.

A

That's something and we have done that a lot of groups do that, like just a list of interested parties just to show interest, you know with artifacts. We actually to your point. We actually did have so like. We gathered people and said, let's at least get together for an hour, and you can use this Zoom chat.

A

If you want or something and we can even record it and post that I don't mind even without the official you know, Ouija um yeah, so I guess the answer is the basics are just somehow that we know not just me. The other leaders in the tag know that yeah these 10 or 15 people, and these four or five projects are paying attention or interested. That's the minimum viable and then yeah.

A

If you want to have a meeting and we actually bring them in a room- and you know talk about the charter- and you know the goals I'd be happy to help you with that I'll be getting better.

E

Yeah yeah I mean I was thinking the the well I think yeah I, don't know 215 people authoring a charter seems like a very difficult thing to accomplish, but um uh certainly I think you know we can get a list of interested parties and get their get their support on it um and then I don't know yeah. We can kind of figure out the next step. My comment about the meeting was more about once we get to the nitty-gritty of the design of the specification um and and to do that we need.

E

Actually, you know, probably several in-person conversations. I know that already there's areas that the existing participants don't exactly agree on. So um tactically, that's kind of the work that we're doing right now is trying to refine the specification to a point where at least all the existing producers and consumers are happy with it um and then yeah. So if if we re-home that particular meeting I think that would be beneficial, um so, okay, but anyway, as.

A

E

Yeah we do have a weekly meeting actually for those and we've had ad hocs with polluting and cross like it's it's beyond just sending an email. Hey. Are you interested? It's actually, you know. We've sat down and discussed some specific, precise details around you know what the API model thing would look like.

E

You know what are some strategies that might reduce the costs of integration for producers, because you know this affects not just the scope, I think of cncf projects, but also you know clouds overall as well, but you know those clouds obviously tend to align pretty well with what the cncf is doing, or at least want to, based on the ensure that their work interoperates well with them. So um that's that's. The second reason why the cncf I think endorsement is pretty important, um but um okay well, I think it's at least that's the next step.

E

I can certainly you know, update the dock a little bit focus on the infrastructure apis a little bit more focused on the specification where I can remove the the tooling part of the conversation from the proposal for now and then I'll. Try try to add a list of all the I guess existing people who voiced in and sort of you know who else might support the project.

A

Yeah sounds like a good good strategy and, let's try to maybe Karina will come to your meetings and stuff. That would be great if we could get one of us from the tag involved. You know just to keep the channel open. Let's try. Let's try. Let's ask her if she wants to join.

E

Okay, yeah we'll do awesome. Well thanks a lot for the time. I really appreciate it.

A

Yeah, thank you. Thank everybody. I know we're a couple minutes over. So thank you, Alex for your presentation. Earlier I'll get this video up soon. um Yeah and I'll see you all online.

F

A

Thanks a lot, bye have a good day.