Cloud Native Computing Foundation App Delivery Special Interest Group, 14 Apr 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CNCF SIG App Delivery 2021-04-14

Description

CNCF SIG App Delivery 2021-04-14

A

Hello, everyone, hello, your background, I think totally wins. Tracy.

B

It's my favorite sci-fi.

B

A

It expense is great. I was a bit struggling after the first uh season, the end of the first season that but yeah, I think it's it's definitely one of the most interesting ones. I.

B

Liked all of the uh the politics and stuff around it between all the kind of reminds me of um there was a really good, uh I think, was called the left, the left hand, side of darkness or the left side of darkness.

B

It's a really awesome sci-fi, and it has a lot of about the different, the different like planets and the the politics of it.

B

B

And that's a piece of artwork that that I have behind me of the taiko station. I would love to have that in my house. It's beautiful.

A

So I'm still posting in here the link today for.

A

A

As you already have some people here, I'll see it later on again for the co-chair and tech lead voting. um Obviously, submission deadline ended on friday.

A

uh I just have to wait for amy to get back from her well-deserved vacation, which will be end of this week. So that's why I did not start with the voting process. I want her support there and, yes, I could have asked her during the during her time off, but I'm definitely not doing this to anybody. So I decided if people take time off. I know that they're reading slacks I'm not even posting to them. So I hope every everybody understands this and then we have to start with the official voting process.

A

So that's the reason why I didn't kick off immediately this week, so I think it's important to honor the other people's time of work, but we'll definitely get this done before kubecon. So we'll be there it's a lot of new people today. That's that's actually amazing.

A

Okay, so let's keep some people cross playing garrett.

A

Chairs over here on cross planes that jared also posted some feedback already on the due diligence document. Great work by the way yeah.

C

Thank you. I saw that uh that feedback come in overnight here and I'm definitely grateful for you spending some time to look through advanced thanks for making time for that.

A

Yeah we try to like work through those uh as much as possible. um Also in my comments, I think the only critical piece for moving this in the time frame that you want to do is really, I think, agreeing on the end user interviews. These usually take some some time and some coordination, but the earlier you can get started on these. uh uh I think the better.

C

Yeah, absolutely I can. I can facilitate making that happen. uh Would that would that be with like you and harry uh as well.

A

uh Toc member, um but they will obviously decide who to who to pick it might be corny cornelia, in this case uh who's going to join but harry having harry on there. I think they can definitely help to get these schedules.

A

Awesome, cool I'll, definitely facilitate making that happen because from from experience, this is usually what takes the longest, because it needs a lot of people to be coordinated across time zones right and.

A

Good have still some people joining. Usually we wait until five minutes after the hour to get started.

A

A

A

So while we as while we are getting started here, if you are new- and some of you are actually new to this group and want to quickly introduce yourself feel free to do so, while we're waiting, so we have one half more minutes to go so feel free to just briefly introduce yourself to the group.

D

Yeah I can go. um I think we've spoken a little uh on dm yeah, uh but my name's alex so I'm currently working at jp morgan transferring somewhere else in a few weeks. But the app delivery group um is extremely interesting to me, specifically around some of the challenges and opportunities that we see in infrastructure and application space. I've been working a little bit with jennifer and thomas to uh to put some ideas together and look forward, looking forward to contributing a bit further.

B

And I'm tracy reagan, I am on the board of the ced foundation. We have an initiative to do more cross-pollination between the cdf and the cncf, I'm also the community director for the ortillius microservice management platform. I'm the ceo of uh deploy hub.

E

uh Yeah I can, I can introduce myself, uh I'm hong kong, I'm from alibaba right, I'm not new. Actually I I participate in since the beginning, but then I I'm I'm a colleague of uh harry so yeah, so I also like participate in a lot of open source projects, because that's what our team do. I I also like uh contributing cosplay and cooper, I'm also cooperative, as well so yeah, also working on a chronic efficiency space. Yeah.

A

And I know that I still owe you a reply was just like super busy until cubecon so expect a reply for me. The next one to two days as well.

E

Yeah no problem yeah, I think harry also like asked me to I- can help provide more input as well right, so we are all together. Looking on the slides.

F

I'm also new, so just introducing myself, my name is anais. I'm currently developer evangelist at code fresh, but I'm also similar about to alex I'm also transitioning somewhere else right now, um yeah um also cncf ambassadors are just looking into different calls and.

A

You were actually super famous with kubernetes 100 100 days or something.

G

A

Everybody should know about yeah.

F

That's yeah, I mean, if there's something that I can like collaborate on and help with and also share in my youtube videos. That would be amazing, so just yeah, it's basically for those who who haven't come across it, which I guess are more people, it's a challenge that I set myself out to do when I got started in the space which is like learning cubanitas across 100 days and just showing people how they could follow a similar path. I.

H

Guess- and my name is robert and I'm uh I'm I'm trying to be more active in these kind of groups, so this is one of the the things that I thought kind of fitted. My working- uh I guess my my work obvious, I'm a cloud architect from crayon in norway and I'm currently working at in the githubs working group as well so trying to do a little bit of everything, but just trying to be more active.

I

H

B

I

Is dan um and I work on cross plan, I'm a maintainer of cross planes, I'm here uh for jared's uh presentation of our incubation proposal today. I'm also a tech lead of kubernetes.

A

Miracle, I think we have a lot of new people by the way. There's also, I posted the link over the doc for the the meeting notes which we try to keep there um looking at the time. I want to give everybody a fair chance to uh speak today, and we want to start off with the cross plane incubation project, which is uh pretty exciting, a pretty exciting project. I think it has also a great logo by the way, whoever came up with that logo, I'm totally big fan of it, so I'll just pass it over.

A

I think to jared you want to present it.

C

Yeah, absolutely I'll I'll. Take it here. Thank you so much for for uh letting us come and speak today and get a get an audience with the sig here uh definitely appreciate that and uh we hear that all the time about the logo as well too, that people people love the little popsicle and imagery around that so I'll pass that, along to chris, the designer he likes he likes hearing that all right. So let me get this presentation going up on the screen here in the agenda doc.

C

There is also um the links to this presentation uh and also, I think, the um the upstream proposal as well too, and the due diligence documents, that's all linked from the agenda document, so you can uh play along at home with all three of those there. Let's dive into this, um so my name is jared watts. I am a co-creator and uh current maintainer of the cross plane project.

C

um We are going to kind of dive into our proposal to move to the incubation phase here um it could be about 10 minutes or less so we'll try to get through this quickly and give everyone else time today to talk too.

C

Are you seeing the screen too, by the way you're seeing the slides cool? So uh let's talk about crossplane, uh maybe a bit of a refresher for some folks in the audience here, but basically it's an open source, kubernetes add-on that focuses in two main feature areas. uh The first one is around allowing infrastructure owners or platform teams to create their own custom abstractions and to sort of enable uh their developer teams to be able to self-service and create the infrastructure that they need on demand.

C

um Crossland has a kind of a broad surface area across a lot of vendors clouds, environments et cetera, so we also have created a um a consistent api across all those environments as well too so we'll go into more details on both of those, but that's sort of the two main high level uh feature areas or focus areas of the crossman project in its charter.

C

We first open sourced it back in december of 2018 so just over two years ago, and uh we are also the creators of the rook project as well too, which is a fully graduated cncf project, so super happy to see the progress and adoption on that project. Also- and you know, taking cross-plane through the same sort of uh maturity, um growth as well too, we first got into the sandbox uh last year june of last year, so it's been about nine months since sandbox entry and um we had our first 1.0.

C

uh You know major stable, milestone, release uh just a few months ago. So, let's dive into some more details here. So let's first talk about you know: what do we mean when we're talking about platform apis and custom abstractions, and all that? So basically, the idea here, that's the core concept of crossplane is that we enable you to assemble or bring together low-level granular resources. uh They could be from multiple clouds, multiple vendors, multiple environments, and then you can expose those low-level resources as a higher-level abstraction.

C

You know a self-service api that your developers can use to get the infrastructure that they need. A good example of this is around composing, these low-level resources of a gke, node pool network, etc. Maybe some helm charts as well too, um like for platform services like uh you know, fluentd or jager, or things like that and then taking those local low-level resources and then offering them as a simple higher level abstraction so that your developers can just interact with a simple cluster object, as opposed to all of this.

C

uh You know, complexity and details of the infrastructure underneath and then, with that you know below that api line below that abstraction. You can put you know all sorts of your policy, organizational policy, specific configuration um and the developer just gets a simple uh api to deal with, or you know, a limited.

B

C

Of configuration that you're letting them uh you're letting them touch, this is all done with the kubernetes api, so everything here is going to be. You know a kubernetes object uh that you can interact with. Basically anything that speaks the kubernetes api, so a lot of the existing tools in the system are very compatible with uh with crosslink because of that approach. And finally, here it's uh it's all declarative, so you don't need to write any code to build your own custom platform. You basically describe it and declare it.

C

um You know in a declarative way, so let's visualize that because that's maybe kind of hard to kind of rock just by talking about it. Let's go left to right on this diagram here so on the left of the diagram we've got our application developers and what they're going to interact with is a simple cluster object that might have a couple configuration settings such as the number of nodes they want in their cluster or maybe the size of the nodes like small, medium or large.

C

They get a very simple abstraction or interface to deal with there underneath the covers behind the api line. Here, what we're going to see is a set of compositions that bring together all those lower level resources. So we have an example here in nws for a cluster you might have eks and then vpc and all of its friends and then for gcp you're, going to have gke network subnet, etc and all those so the example we're showing here is that a simple cluster object can be. You know something in aws.

C

It can mean something else in gcp, but um that's you know one example of multi-cloud sort of thing. It could also be all within one cloud as well too. You could have a composition for like a fast and a slow instance or an expensive in a cheap instance, or gold and silver or whatever um you know you can have multiple compositions that satisfy this simple abstraction at a higher level.

C

um The so the low-level granular resources are probably exactly what you think they'd be they're. Basically, you know cloud services, on-premises infrastructure, all sorts of stuff can be represented as a crd in kubernetes, so extend the control plane with the knowledge of new types and things outside the control plane. um So, for instance, uh amazon's relational database service rds. There is a crd in cross plane that represents the rds service, the user. Can, you know, configure all the different settings on rds they want to, and you know, interact with the low-level resource there.

C

There's a controller in the kubernetes control plane. That's watching for changes to that rds crd and then it's calling out to amazon's apis to you know makes basically make the actual states in amazon match the desired state. That's on the crd, and so these are the low-level resources that we, you know can compose together to form higher level abstractions on top, but they're, probably exactly as you would have guessed, their crds and controllers.

C

So let's talk a little bit more about the the crossland resource model as well too. So you know, as we've seen, the crossplane extends the kubernetes control plane with a whole bunch of different providers, environments, vendors types, etc, and so we quickly realized too that it would be very useful to have a consistent api or a standardized way to interact with all these uh different types of resources um that you know crossplane supports.

C

So um you know, basically, you can think of the crossband resource model or xrm, as we call it uh as an extension of the kubernetes resource model um there. It adds some opinions to the the kubernetes resource model and basically with the intent of providing a consistent management experience and api across all these different vendors. um You know what you end up with.

C

Is that all the objects uh you know for amazon, rds or google cloud sql or whatever it may be, or even abstractions and compositions, on top of those they're all going to have a similar shape and behavior to them? So, for instance, things like cross refresh references like one reference or sorry one resource depending on a field and another reference, uh sorry another resource, that's all going to be. You know very consistent across all the different vendors and clouds. um You know, status conditions are going to be fairly similar.

C

The way you do credentials and get connection information to connect resources, like databases, caches clusters, etc. um You know there's a lot of different ways that we've made some commonality and sort of consistent api across all these different resources and abstractions, etc, and kind of make some more sense of it and have a reasonable way to deal with all of them in a consistent manner.

C

We'll talk about what we've done since sandbox, so that was about nine months ago, um and so, as I mentioned, we had our 1.0 release. That was probably the biggest thing that happened in the last nine months is that you know we have with the 1.0 release. We've declared the core apis to be stable and the project is ready for production usage in response to that, we've seen uh end user adoption increase and people start taking it into production as well too.

C

Now that the project has reached a certain amount of stability, and you know backwards, compatibility and things like that. um In addition to that, something that I even find myself to be more exciting is that the non-trivial contributions we're getting from the greater community have picked up as well too, so that people are deploying it to production they're. You know having a use case that they think is important to them, and then they basically will implement that upstream uh for the benefit of the greater community.

C

So seeing those non-trivial contributions come in from from you know, outside of the core maintainer set is, has been really really exciting. um You know we're also working with um a lot of partners in the ecosystem and doing a lot of collaborations with uh various um cncf projects as well too. I've got a whole bunch of different links here for each one of those that you can.

C

You know, click through and learn more about, um but I want to give a quick shout out there to dan on the call that he, you know, he's been doing. A live stream show that, basically, you know every couple weeks does an integration with another cncf project, about how you know open policy, agent, falco, etc. So that's been really cool to see ways that crossplane is compatible with or it can work alongside other projects in the ecosystem. So a lot of cool links there. Another other thing here is our pack. We have a package manager.

C

We did a v2 of that. Where basically say you have a you know, a bunch of platform apis, you've designed and they have dependencies on- maybe provider, aws provider, gcp provider alibaba, and so you can basically declare hey this platform. Api needs these providers and we'll handle those dependencies for you, we'll fulfill them and make sure they're installed and available for you, and then you can do upgrades and rollbacks and stuff as well too.

C

So that's been really nice for the maturity to be able to you know, version and iterate on the the providers and things that are installed in your crossband clusters and then, finally, we did a really successful community today, event, alongside of our 1.0 release- and we had a really awesome speaker line up there uh with some of the creators of kubernetes as well too, so that was really fun to see the community get together and have some really awesome um time that we all spent together there.

C

um So, let's talk about a couple stats, real quick! So as I mentioned, sandbox entry was nine months ago and uh a lot of stats have increased by 50. Since then, um some of them are even further like container downloads, we're seeing a 10x increase in and then also our slack members have uh increased two and 2.5 x. The um that's where we could uh congregate mostly, is in slack and we collaborate with community pretty heavily there in addition to github.

C

So it's nice to see more people coming in there asking questions answering questions as well, too, and kind of participating together in that community there uh and then uh getting close to wrapping this up. uh Let's talk about some of the partners and adopters um those are kind of outlined in more detail in the due diligence and the upstream proposal. um But someone kind of want to highlight here real quick is the is deutsche bahn they've. You know adopted crossbane into production and I find it.

C

I find them interesting because you know they're the largest railway operator in europe, so seeing an enterprise of that size, kind of say, cross maintenance production has been really pretty pretty impressive. Accenture has been a huge help to them. uh Kind of you know guiding them and helping them get get that uh all deployed and find success with crossplane and they've also been a huge. um You know, source of feedback, and um you know collaboration to contributions to the project as well too. So that's been great cloud, checkers kind of interesting.

C

I think too, because they're, uh what they're basically doing is they're replacing their terraform usage with crossplane they're buying in and they're adopting uh this you know a control, plane approach as opposed to a single one-off. You know, infrastructures, code tool, execution that you know they're buying into having a control plane manage all their resources instead.

C

So I think that's interesting and the last one is mothership. It's interesting too, because not only are they, you know, building platforms, custom platforms for their developers and you know being able to provision infrastructure, but they're, also extending uh crossband and writing new controllers to do day. Two operations, so kind of ongoing operational tasks uh are being implemented by them as well too, which is kind of it is pretty interesting. I'm happy to see that um so final slide here. The uh there's links to all the resources for the project.

C

uh The upstream pr is 620 in the toc repo and there's a link to the due diligence documents as well too starting to get some feedback on there and really appreciate that um and then also some other. You know general projects resources there that you can peruse as well too.

C

So I think that's all the resources there, all my spiel. uh Thank you super uh super much for listening, and um you know I don't know if we have a lot of time for questions today with the agenda, but um I'm always available to you know. If you ask questions on the due diligence documents or the you know, proposal 620 upstream uh we're happy to engage and get everyone's questions answered. Make sure that everyone's everyone's getting the information they need. So thank you very much appreciate it.

A

Yeah thank you for presenting. Obviously we open it up for questions, so we do have some time for questions looking at at least five minutes, so I'll definitely want to open it up for people that can ask some questions.

A

Okay, if nobody starts- and I start with a question- and I was actually having when looking at crossplane. First of all, I think it's a great project. I think it's really helping to model these resources and it's a very interesting use case of kubernetes for like really using that kubernetes api server and and operator bot um modeling there when the controller model.

A

My first question is: how do you handle security? um I mean obviously you're giving developer the freedom to create crds which might, um behind the scenes, create a massive amount of of resources and um something we've seen also uh when we're looking at at resource modeling that usually the rights that you need to create. The individual resources in the cloud are kind of different from the rights that you need to create your own clusters. What's the the security model, uh how do model security and access control, more access, control actually than security.

C

Yeah yeah, I agree a great question and so that touches on something. That's that's really a foundational concept in crossplane. Is this whole idea about separation of concerns, and so there's kind of I didn't really get into it too much, but there's kind of multiple uh personas that are going to be interacting with crossplane on different sides of the api line. Let's say so: your infrastructure owners, your platform team, you know they will have full access uh to. You know, define these platform apis to assemble resources together to express policy all that sort of stuff.

C

So you know that's they have this one persona that has you know a certain high level of access and then with the abstractions that you build. On top of that, uh you know those are intended to be consumed by lower level. You know lower privilege, application teams and they're very focused they're scoped. So you can do you know. Kubernetes are back to all of those resources so, for instance, this cluster object.

C

You know you can grant access to that and only that object to your developers, and so they can interact with the specific configuration on that object. But they have no access to the underlying resources themselves, to edit them or tweak them, or to skirt your policy, or anything like that. So the separation of concerns is, uh you know, pretty major uh architectural uh aspect of the cross playing uh design.

C

um That kind of you know helps separate out who gets access to what um something that's also fairly interesting too, that you know access is granted on the abstraction and not to the underlying resources.

C

Specifically, is that um you know you can programmatically determine like what are the im roles that you would need in a cloud provider in order to be able to do, or uh you know, operate on all the types that are going to be underneath the lower level resources, you can kind of make sure that you know you have a good understanding of those you can lock that down, etc. But the separation of concerns thing is, is a really really important concept for the security model um and dan.

C

Do you have anything that you would want to add to that to kind of fletch out that thought a little bit more.

I

No, I think that was a a pretty good covering of the offering, I will say, that's uh definitely a big proponent of crossplane compared to something like infrastructure as code tools um where your level of abstraction, uh your permissioning, does not match that. um I also would point to uh some of the resources uh on the crossplane blog, um where we have some kind of like compare and contrast with uh additional models or traditional models of infrastructure, abstraction.

A

Yeah, thank you because that's one of the key points for something like this to work, uh because you need like very different rules for like lower level resources than for the high level concepts. uh um I think that that would be well worth if you could maybe also edit. I mean it's not necessarily part of the due diligence document, but I think it's an important resource to have somewhere, especially how you handle um this type of access control, because, like one of the work streams in sick, active delivery always was okay.

A

How we can we provide kind of like higher level primitives for people to work with um and where you say. Well, it's not just about helm charts. There are some, especially when it comes to access control. It starts to get a bit more complex and just pure composability.

A

So I think that that by itself would be an interesting topic to to discuss um a another question. The second question that I would have is um obviously some of the cloud providers- and you also mentioned it in the due diligence document, they're kind of like building, I wouldn't say, like their own controller versions, um um to some extent that they do uh so so how far along are you that they would and again this shouldn't be like just one solution out there, but how's your collaboration with them.

A

Or do you have some joint plans that maybe xrm is a model that is going to be adopted by multiple of them or by like the major cloud providers for their resources as well?.

C

Yeah yeah, that's that's a really good question as well too, and I took a note by the way too, to update the due diligence document with the security model and access permissions and all that sort of stuff, so I'll incorporate that into the doc um yeah. So for so, for you know the cloud provider um apis and partnerships with them. So, basically, um you know the concept of bringing external resources such as a cloud provider. Maintenance service into the management purview of the kubernetes control plane is, is not a concept.

C

That's exclusive to cross plane now, other the cloud providers are starting to do some of their own uh projects as well too. That's you know basically enable you to create crds for the um you know, resources that belong in their cloud right, and so uh we actually have a pretty strong set of partnerships going with all the major cloud providers there. um Aws is probably the one that's furthest along, I think so. Nwx has uh the ack is the name of their project.

C

um It's you know, uh amazon's controllers for kubernetes, and so um you know, we view those as kind of uh low-level resources that you can use to directly control the ws api aws api, so we've partnered with aws and hooked into their code generation pipelines so that you know, services in aws and generating code for the ack controllers or in the same process can be used to generate code for the cross plane providers as well too.

C

So I think, there's you know, there's space for both of those projects where you know ack or something specific to amazon. You know if you're want to. You know, lives exclusively in the database ecosystem and you know only consume those apis not and not necessarily have any sort of abstraction capabilities. On top of it, then you can choose to live, uh live in that ecosystem.

C

But if you want to do anything cross-cloud or you want to have some sort of consistency across cloud providers or if you want to start building abstractions with the machinery and cross plane, then I think that's, uh you know a really strong story to start getting. uh You know consuming or using the crosswind project and then, at the end of the day, to the you know, collaboration with these uh cloud providers we're doing with azure as well too starting with gcp. um You know have a decent relationship with alibaba as well too.

C

um At the end of the day. What I would like to see is the uh cloud reticent cells, you know, become maintainers and take ownership of the cross, plane provider, and so I think, a nice step in that is sharing code generation pipelines and then continuing to increase the amount of collaboration between the projects. So they become owners of the classroom providers as well too.

A

Yeah, I think that that's exactly my point and maybe the cncf can then also help once you move into incubation to make it more um attractive, obviously, for them to do so, because I think that's the real value I mean when you actually go in there and say I want, I don't know a load balancer and I can get a load balance on gcp as well as I could get it on aws or azure, because and I'm just kind of curious how closely you're working with that, because all of them have their own resource model like azure has arm.

A

You have cloud formation, obviously on the aws side, and google has also their own version of this model. So that's uh that that was my own thought. I only thought okay, but which is also not part of the due diligence, but whether you have some clear ideas on how to bring cloud providers closer together on something. That's like really core, like the core service models that they have, which kind of differ between those cloud providers a bit more.

A

Maybe it would be a nice idea where you could it's it where they also would like to learn more about, like which resources? Do you see, lend themselves well to um say more like a an agreement across cloud providers and which just don't because they're so specific on on the individual cloud providers, um but again it goes beyond the scope of cross plane. I just think it's it's a great opportunity there, where, um ideally, we would see more.

C

Yeah, I I totally agree with that yeah. I think that that's a really good point, that's something! That's really nice! With the you know, flexibility of the abstraction model in cross plane as well too, is that you know the it's more of a framework or a tool set to be able to define lots of different uh attractions and compositions etc. So we're not necessarily stuck to any lowest common denominator single.

C

You know abstraction of truth uh that everybody has to use, there's lots of opportunities to create more flexible abstractions that you know fit use cases, uh somebody's use cases better than others, but also you know, do our best at creating more unifying abstractions as well too. So there's definitely opportunity to keep growing uh that those endeavors there. I think.

A

Yeah, I just see the opportunity that at some point, xrm or a combined version like working with the other cloud providers on xram might become its own, maybe sub project. Even I would not like do it right now say that it necessarily needs to happen, but once at some point we want to have maybe the definition independent from from the implementation.

A

um But that's maybe a discussion for for another day.

C

Yeah- and I see that being a trajectory that we can go in as well too, like the next step- is kind of more formalizing the xrm and putting a more formal definition to it, and then you know letting it mature that way and giving us opportunities to. Perhaps you know make it an independent thing.

C

I can see that.

A

But you're already working with alibaba and microsoft was also working on oam. So I might see this at some point: maybe not converge, but look any closer yeah thanks for for sharing I'll open it up once more for a question before we jump into a working group proposal.

J

I had one question.

K

Just in terms of how you bring existing resources under management, that would be a.

K

I know you have these uh dynamic and statically provisioned resources, but is it is it that you just have to start with uh nothing or do you have a pattern for bringing existing resources under management? As you.

C

Yeah great question great question, and yes, there is a pattern for that kind of an adoption model where you can have existing resources out there in your cloud providers um and then the you can.

C

Basically, you know create a cross-plain uh representation of that, a crd that matches that resource and uh there's an annotation, that's called external name, and then you set that to be the name of the resource in the cloud provider, and so when it uh or that map, when that annotation is matching an existing resource, crossplane will kind of adopt that resource and say, okay cool. This already exists. Let me get all the configuration.

C

Let me populate the crd. Let me make sure that you know now. This is under management of cross, plane and kind of adopted into the control plane, so that is set up to to do that to take existing resources and bring them into the management of their cross plane. uh One interesting aspect of that. An extension of that is that um I think there will be two use cases or two variations of that. One could be.

C

Maybe you want crossband to manage it and you want to be able to update it and change values over time. Another one could be you want it to be observe only you want it to be a read-only resource where you want to adopt an existing resource that maybe is managed by another tool, but you can have it in the control plane across plane in a read-only fashion, so that other resources can reference it and pull values off of it.

C

uh Crossband won't update it or manage it, uh it'll be a read-only sort of thing, but it's then still available in the crossband control plane for references from other resources and that's not supported yet the observed only concept isn't supported yet, but that's on the roadmap in the in the near term.

J

All right thanks, so this doesn't work. I just had a quick questions, but in some ways it seems like this is uh a superset of what's available in cluster api. Can can you quickly comment on that.

C

uh Yeah, hey dan you've uh you've been working with cluster api, uh specifically on a number of dimensions. Do you wanna? Do you wanna take that one yeah.

I

Absolutely so, um I'm sure a lot of folks here are already familiar with how cluster api works, um but it has a similar model where it has uh core components. uh Then providers uh for different, you know kubernetes offerings um whether they be uh you know, a hosted offering uh like gke or eks um or it's. You know, spinning up ec2 instances and installing kubernetes with cube admin, or something like that.

I

um So it's you can envision it kind of as a very focused uh version of crossplane in terms of it targets very specific infrastructure, and much of that is kind of hard-coded.

I

One of the things that we've worked on with the cluster api community and we're continuing to develop is using crossplane as a sort of back end for cluster api. So you can imagine that their providers are very scoped, so, for instance, the gcp provider for uh cluster api exposes a machine type and a cluster type. uh The cluster type, its controller when it reconciles uh creates things like a vpc or a network, is what it's called on.

I

Gcp some security groups, subnets, etc kind of all of the peripheral things around setting up a cluster, and then the machine would be actual compute instances on gcp, um so so that cluster type uh has those components that it creates hard coded in the controller uh on. On the contrary, for crossplane, you can design sort of your abstraction, so something like a cluster type dynamically by writing.

I

Yaml and mapping it to granular resources, so the gcp provider for crossplane, instead of exposing a kind of abstract cluster type that provisions a bunch of network, is going to expose actually all of those granular resources. So the subnets the project, et cetera, all those different things that would be required.

I

It would be granular resources and then you would define a cluster type that that basically maps to all of those which makes it a little bit easier right for a user to be able to configure exactly what it means when you're setting up a kubernetes cluster on gcp and that can be changed and repackaged in minutes or seconds, as opposed to you know, having to rewrite the controller um and rebuild it and that sort of thing.

I

um So, there's a lot of advantages to that and you actually because the crossplane composition model allows you to um you, know kind of define any schema that you want. We've actually shown a demo of being able to take that gcp cluster type in the cluster api controller create an xrd, a composite resource definition with crosswind that defines a new schema and have that map to those different granular resources.

I

um So, instead of having a single controller that watches that and creates all those different resources on gcp, you have a controller that watches that abstract cluster type spits out those granular resources which then cross plans. Provider gcp has individual controllers for each of those resources to go and provision them, um and- and you can imagine right, you could change that mapping to to your liking at runtime with crossplane.

I

So we like to envision crossplane as kind of a potential back end for cluster api um and in the short term, that likely looks like having kind of a uh cluster api provider, that's backed by crossplane in additional to the work that they've done on all those individual ones.

I

And it's my belief in the long term that that would be a kind of desired direction to go right because it does allow for much more configurability and customization uh by end users. Okay, great, thank you.

A

Thanks man, I have to jump in here because there's lots on the agenda. I mean it's good that we have this discussion. We might want to do maybe once in a session specifically on xrm and dive deeper into those questions.

A

If people are interested, I still want to give the adults, especially the working group proposal, a chance to to present as well, and somebody has to play the guy who watches the clock and I unfortunately have to be today. I want to give uh the team here a chance to propose what they're doing, because it is kind of related, not directly but kind of related. So I think it fits very well in this segment here.

A

C

This is the opportunity today, too appreciate it today.

L

Okay, um yes, so, um as arlo has announced before um we are talking about the real, um similar topic to that what we heard before um we are uh so so we are jennifer and alex, and me um we put our hands to our heads together and um try to discuss such things as bringing infrastructure and applications together um and therefore we came to the currently working uh working title application enablement working group. For that, um yes- and I think we'll try to just try to describe this in the next few slides, okay.

L

So um at first we um we identified a bit of a problem, so um application delivery and deployment is often about artifacts and their configuration. But um all of these um artifacts intercon artifacts need infrastructure at some point at some point.

L

So, um as we heard from from from crossplane before um we have databases, we have file storages, we have message queues and so on um and at some point in time we um we might deploy applications um and we might have a bit of a shared response of a of a broader responsibility between the application, infrastructure developers and the infrastructure guys and at some point we want to apply application, configuration or deploy applications and there might be infrastructure infrastructure components which are not available in the in the target infrastructure, so um think about a development stage of an oven deployment.

L

um You configured an efs provisioner by yourself and you try to push the to to provision the the same application to a staging or production environment, and there is not uh the efs provision is not deployed there and therefore the deployment will break and we think that there's almost no solution. So you could obviously do do something with crossplane regarding this, but there's almost no solution which handles both infrastructure and application deployment.

L

But we also think that end users might have found ways to deal with this, um so they might use terraform prelumi crossbrain for infrastructure deployment um and they might use argo spinach or captain for the application deployment and they might um do a link via ci tool or other other different things.

L

But um we also think there are no best practices at the moment, and this is a gap which we we as the same as the cncf working with a good fee. And, yes, alex, will tell you what with what we think about application, enablement.

D

Thanks thomas yeah, so just to give some annotations on those previous points, and just to add to that you know we, we look at things like the service mesh working groups and we look at things like smi spec and how that's a ubiquitous spec. That's an agreed upon standard across vendors and across cloud platforms, and we think about that kind of standard as an end goal, but more immediately.

D

We think about application enablement as being the anecdotes that we kind of pull together to understand what are the best practices and when I was trying to work with jennifer and thomas. We came up with sort of this. This sentiment here of and I'll read it out. You know application enablement is accomplished by describing the requirements for an application workload to operate within a hosted environment and provision components as they are required. This domain encompasses the pipelining, provisioning and distribution of necessary underlying infrastructure components.

D

A ubiquitous yet agnostic pattern to ensure applications are deliverable to any appropriate cloud native ecosystem, and you know: we've been speaking about this for the past sort of 30 minutes or so, but this resonates at so many points, and I think that now more than ever, we're really at a singularity where we need to have an opinion within uh the cncf uh sig of how this is how this is done in the real world, what the best practices are and potentially, where can there be?

D

um You know standardization, and you know if you don't use such a strong word. You can say at least attunement between methodologies, and so that's what we're looking to do, and you know just to double down on this thomas. If you go to the next slide, please you know I I draw this very, very um terse illustration, and you know we all understand kind of what I'm getting at.

D

I believe in that there are quite a few steps, and and really the application code doesn't get a look in and there's no clear way to really describe what I'm doing here. You know we talk about provisioning in ingress. We don't talk about the wait timeout that I have to introduce, so that I p address is provisioned as it's coming up through the stack. We don't talk about. um You know how does the ci, the ci uh provision its credentials to then execute a remote command to check if a health check passes?

D

These are all the kind of discrete behaviors that companies throughout the world are having to introduce to accomplish this multi-layer process of application, delivery and enablement, and so what we would like to be able to do in the sig again is to sort of look at these kind of um edifices of how do you deploy and how do you deploy on top of that uh is and figure out what are the most common patterns that we're seeing in utilization and uh thomas?

D

If you go to the next slide, please you'll see that, where we're trying to find our niche and where we're trying to sort of come together on here is looking at the infrastructure, the very strong infrastructure, provisioning space and the very opinionated very strong space in terms of ci cd enablement, um you know as we as we were introducing earlier on. The call with you know, folks that folks, like tracy and and people from the continuous delivery foundation, are thinking about and looking at the synergy there in terms of you know, end game.

D

How do we have some sort of specification schema standard opinion on what works really well together? What technologies potentially have projects that can grow out of them? And what could this working group foster in terms of collaboration? And so I believe that- and I hope that that kind of captures- a sentiment and with that I'll I'll pass on to jennifer jennifer.

G

Sorry, I was muted hi uh yeah, so just summarizing here is it and putting this uh matching with the big charter. uh We are looking from the whole uh end to end right from the application definition, uh configuration packaging and deployment the application delivery, uh their workflow as well, and the yeah I already said the workflow.

G

So uh could you go from the next slide and now propose that goes here is like is to get the uh find out about the current practices and the landscape, because there's a lot going on at the moment, uh various cutting-edge stuff- and we yeah want to make reason about those and and make sense of what's going on, give the end users the ideas, examples we want to run some pocs and see how they could integrate that with the solutions we have now on the landscape.

G

um We are also thinking of the personas. Here too, uh we don't want to overload uh like operations. For example, we want to make sure that production engineering can uh be. We can be able to participate, get involved on that too. So, um balancing the involvement of various personas across application delivery and provide best practices uh through uh what thomas and alex have said could be.

G

Standards could be um uh some of just some uh recommendation or white paper um yeah and, as you can see in the future, possibly create a specification for handling infrastructure components in called native environments.

G

Next slide, please, um I think, until we are not going to run something and then recommend a specific food chain, uh we want to look at an agnostic way and and present uh the parent. What is currently going on in the community what people are using and uh come up with an opinion best practices, and I will just get back back to thomas. Thank you.

L

Okay, um yes, so we also thought about how we could um how we could achieve our goals and how we could um start using on this on these topics, and um we thought in the first step.

L

We don't want to write another white paper because um jennifer and me we wrote the white paper in the last few months and in the first step we want to beat something and um want to try want to try to give to give the end users something back which which provides value to value for them similar to the potato head um and that they can try things out, that they can try to find out what could what could fit best for them and so on.

L

So we try to build something usable and um we want to achieve this by trying to integrate tools and find solutions. So um I think there are many many infrastructures code tools as alex described before, and there are also a lot of cicd tools um which can be integrated, and I think everyone is interested in integrating these things and after we found out which solutions uh there could uh we we have on the market and how? How best practices could look like for such for such efforts?

L

um It might be that there are some some things which uh we find out, that that aren't currently covered by anyone, um and then we could um build a white paper, a similar document um which, under which is underpinned by our practical work.

L

um Yes in the future, as alex said before, and we could possibly possibly also design something it like a like a uh like a cloud infrastructure interface. So as my like um that there might be a database claim for databases, um and it might also be that we create an abstraction uh which um creates a database and proxies that for the for the application, so that we need the that we also need the endpoint of the database.

L

Now I have the endpoint of the database and possibly possibly also access control, but this is um kind of a future topic. um Yes, and this was our proposal, um I hope everyone liked it and um yes um we'd like to ask how we should proceed.

A

Yeah I mean we, we briefly talked before and that's what was encouraging uh this. This work to be done and, as you mentioned, jennifer specifically, this is in line with a lot. What we want to do to have like this application definition topic, whether it's on the application or underlying infrastructure topic, which is in the last time I was talking about this on twitter. I got roasted by some journalists that the kubernetes doesn't have an app definition, and I should never say that um I think it is important.

A

um We also have harvesting data where the work is over, then around xram. I think the idea that I totally like how different tools are doing it today and starting there is uh is, is a great approach. I would definitely reach out to the existing projects to do something in a space, not just because it's nice to do things that there are things there already and uh then maybe also reach out then at some point to the end user community.

A

Obviously, I think the working group makes sense, I'm still personally struggling a bit with the name, because I think it's not immediately what it does, and I know the names are always hard to find. I think it hits a very important point that we also saw with zxram and others that we need a better way to define our application constructs dependencies and how we model something there.

A

um So I I'd say having like the charter and ideally reaching out to maintainers of those individual projects that you have already listed and what daisy would fit in there. I think it's just super important to get also their feedback um on this topic, um because the official, uh the official setup of the working group is produced. Wait for your proposed shares. You need to find a charter, and then we go to the toc.

A

I think it needs a bit more work before we go to the toc. What I would personally leave out for the time being, is this smi kind of thing or wants to be that king maker, and if it has this okay, let's define a standard on top of other think so jared all the way left. But I think they might not be super happy if they say like this is the uber xrm model and then and now that that's not what you want to do.

A

I mean just making it uh uh easier to to digest and maybe having a couple of key users that you use. Cases that you want to address is is great, but also the examples obviously make make a lot of sense. So uh it's it's interesting. I think everybody understands the problem, but then putting it into words still feels a bit hard, which kind of means we don't have a even a word for what we're trying to do here.

A

Also, we all agree on the problem, and that might actually also be the trick here, maybe to again put in that. This is the problem that we need to solve, that people run into. So that's how I would propose to to go forward and, like start with the charter document, reach out to the projects and ideally have somebody from those individual projects joining you, because the worst thing and I'm coming from a standardization background, is a standard that nobody wants to implement. So what what do individual projects want to adopt? What adopt?

A

What do they already have available? Is there like a common agreement? um How are they complementary? I think that by itself is, is a very great starting point to see what's what's there, but I think it's a it's great that we're moving in that direction. That was from the very beginning, I remember when uh brian harry and myself started to to work on this uh so yeah. We need an application definition and make this easier and really move up the application stack here.

A

So I think uh great initiative to get started, and maybe we regroup what your findings are with those those other projects that doesn't what I like about the seek to do more like in in general. First of all, have more end user related impact. Well, I don't like the word end user. It feels fine like people who are actually using it on a day-to-day basis, and sometimes it's hard to say, because I know I'm obviously working with thomas. He is an end user, also he's also working for the vendor, uh like like many others.

A

uh I think that's an important part and also we bring projects closer together to collaborate on problems that exist in the industry. If you can do both together in a working group, I think then that's ideal.

A

It was like me very long saying it's a great idea, talk to a couple more people and let's recoup the next time and see what you found out- that's the tlr. Maybe it's good.

B

K

Helpful, it might be.

B

K

To do a data model of what um you're proposing kind of like you, can look at the xrm or what what they've done and say like it? um Overlaps with you know what cluster api is doing. That would be helpful and like in an area you're working on, I would say, like part of the model, might be actually multiple applications right, because in in a space, let's say where you have 200 microservices.

K

You know there might be a a hierarchy you might want to represent, or some dependency that you might want to represent, or you might want to encounter like environments being a first-class citizen and rolling out applications. You talked about monitoring those all kind of different parts of your data model. So if you have a data model of what you saw, your crd, looking at or crds looking like, that, would that would be helpful in terms of or even not thinking about a crds just in terms of how?

K

What is it that you guys want to model? And that might be enough to describe in words you know like, because those are written kind of what you see as your.

K

B

I was just curious: if we can uh uh is there a place for that slide deck? Is that the.

L

Yeah we can drop.

D

It in we can yes that.

B

Would be great uh just in case uh anybody else from the cdf would like to see what you're up to on the best practices yeah.

A

I think putting the slide deck in there and also starting the discussion with the charter document, where people can chime in there and I'm leaving at that time to you to reach out to to the projects and things people were also asking here where there's a specific slack channel robert was asking yes, there is a specifics like channel. There is the app deliveries like channel in the cncf like not in the company. This is like just sometimes confusing for people.

A

I think most it's like that. You can't go there or you can also use the mailing list. I think the most people go to the to the slack channel and when we share something we usually try to share it on those channels. Here, yeah. I think we ran out of time today, which is good and bad. At the same time. The good thing is, we have a lot of things to discuss a lot of momentum made about time. We have to postpone uh some of those conversations to the next time.

A

Like the operator working group update, uh that's maybe a very short one that the operator working group can do, but I'm afraid we won't be able to talk about conveyor today. So we won't have those 15 minutes. So I'd like to postpone it to next time and use the last two to three minutes. Just for a very, very quick update on the operator working group, progress.

L

Okay- um yes, it's also me um or us. Let's say this way. um So um in the last I think about two weeks before uh ago, we've um finished our collab collaborative review of the operator white paper. um So currently we are, uh we are starting the next review phase or, let's say the public command phase.

L

um Therefore, I've I've uh we've also create uh opened up a pull request, um which I will share afterwards in the in the um slack channel and on the mailing list, and um it would be nice if some, if someone of you would take a look on the operator white paper, um do comments. um If something is not really clear in the document um feel free to comment it or make proposals, suggestions or whatever I think we will.

L

The public command phase will take two hours a week two week weeks as we discuss and discuss with others, um and afterwards we will try to publish the the white paper um yes and that that's all of our progress at the moment. So we we have no more content to present.

L

So we're on in the in the last steps.

A

Yeah that was a long time in the making. I think it's great, and then let's um get this also shared, maybe on the tc mailing list once everybody feels comfortable to do so, definitely something to have ready for kubecon coming up and yeah. We are up on top of the hour already, so I know amy usually scales this meeting for 45 minutes and she knows that we're already running over, so I'm causing some of you not having a break and um so for conveyor.

A

I would like to move to the next one and then give them like the start to start right off this conveyor that they have a fair chance to present in two weeks from now, and otherwise I want to thank every all of you for your presentations.

A

I think it was a very good and very well prepared meeting today a lot of different topics also interesting to see topics converge on the working group side, as well as on the cross plain side. So that's definitely great and again, if you have any proposals for topics to present for the next meeting feel free to post them in the agenda. We try to arrange them as uh as as much as we can uh and accommodate them.

A

If you want to do a presentation, usually good at your head you've seen today like if you stick something that was ten minutes plus five minutes discussions, that's fine! If uh topics go deeper, I'd rather schedule a follow-up than cutting it off and not coming to those points thanks everyone for today and hope to see you again in two weeks from now.

K

L

Okay, um so I think we are on the operator working group meeting now yeah. Is it the same? Zoom call I just left same zoom chord, but I'm not sure if someone will join there. Okay,.

L

um Jennifer is also asking if there is a operator working group now yeah.

L

But but I also think we don't have much to discuss today.

D

Okay, uh I guess I can you know I can, if there's anything I can help with in the review cycle. I'm happy to do so.

L

Yes, as well as we talked before, we should make the whole wording a bit more fluent if possible. So if there's something you'll find there, okay, it would be nice if you could do this and.

D

I'll go through I'll, go through the document again and look for more, like not just grammar but also maybe like phrasing, and you know taxonomy hello again,.

G

L

um So I think the operator working group will be very short short today, um so I thought uh alice alex and we talked about um what we should do next, so um I'll kick off the the public comment phase now um or tomorrow or whenever um and afterwards we should try to make the whole document a bit more fluent. So I think this is the thing we talked about as um narrative voice and so on.

L

um Yes to make it better, readable and so on, and I hope um yes and um I hope in two weeks we have a final version of the operator white paper um which we uh and that talk. I also talked to emily from music security, and she said um we can uh reach out to the help desk of the cncf and they help us with publishing the white people so also make it making it pretty and shiny and whatever.

G

Is this public comment? Is that the toc going to look at it and will we I mean if it's available tomorrow or something like we and we are making the narrative voice like we do at the same time, then, like the is like in parallel that we're going to be submitting the changes, because it feels like.

E

G

Be like a change everywhere, maybe not maybe lots of chapters we want to keep the same, but to have the same flow. Could we could change the whole thing so.

L

um We can also we could also um start the public command phase on monday. So if you want we can make, we can try to make it a bit more fluent until monday and start the public command phase. Then.

G

L

So I think it's I think from a time perspective, it's absolutely no problem at the moment, because I think we we we said that we will be finished on the on this. On the start of may, and I think we'll uh this we can accomplish this.

G

Okay: okay I'll try to get on friday some hours to do that.

L

But only if it's okay for you, we can also say we start the command phase on wednesday. So for me it's absolutely no problem, but we should start it at some point.

G

Yeah yeah yeah uh before the next week now are we having sea gap delivery meetings the same day we used to have every other like things like exchange.

L

No, I think the next gap delivery meeting is next week.

G

Wow, okay, I heard you're saying.

D

G

D

We have to do the uh reaching out to the community by next week and the other projects.

L

um We could try this, but um I don't think we have to.

D

Okay, it might be good to start I'm happy to help like yeah, I'm not sure we can guarantee that everyone will respond, but maybe we can at least say that we put it in progress and also maybe have a look at a draft charter. Yeah.

L

Yes, um I've I've created the document so that the google doc in the in the beginning, in a way that it might be easily converted to a charter.

L

So I copied that this from the github's working group charter, okay, and so I think we only have to add some things which we which we described in the presentation now. But then we can use this as a chart also, um and I would try to send out the charter as soon as possible, to get a bit more more people in.

D

There I feel like we would make this a lot more successful if valerie's uh agreed with the name, so yeah that's difficult. I I did have. um I did have one other idea and I don't think we mentioned it on the initial ideation on the document, um and that was the word interoperability.

D

So maybe we need to go and think about more names, but I'm not sure he doesn't seem keen on it, though maybe he just needs to get used to it.

L

um Yes, I think some point with interoperability or integration um working group would be coolest.

G

We can get uh opinions from the projects as well as a good.

D

G

D

G

And stuff and then.

D

How would you reckon, how would you want to contact them like what's the best? What's the normal way,.

L

I think there is no normal way.

D

Yeah, it's just.

L

Random for the for the operator working group, I often try to ping some people on slick or some projects.

D

Could we use? Could we use the sig app delivery channel because they'll all be in that channel and just say, hey, please, participate and do like a google survey or something or maybe.

L

D

We wouldn't get enough people to participate.

L

Am I thinking the first step, we could use the app delivery channel and reach out and try uh try to find some people from from there.

D

How about I will um dm us in our group I'll put a a sentence or two together and we can look at it and then, if you think it looks good I'll paste it and then maybe I can just invite people to contact us if they'd like to participate or something like that. Yes, so absolutely no problem! Okay, I'll! Take that as an action.

L

um At some point in time, if you, if we have um some people um which might join us, um we can also schedule a zoom meeting for that um and.

D

Yeah, like the the the guy that we met today, laszlo who's working on a similar problem space would probably be interested in this kind of sig. So this kind of working group.

L

Yeah, so um he would also be a good candidate for that, and also the cross playing guys would be perfect candidates for that.

L

I'm not sure if we have something from someone from tierra form in the special interest group, but I think I saw I saw someone from hashicorp there some time ago.

D

I know that tracy would be really interested. You know because she represents not only um uh not only deploy hub but also the cdf, and I I work a little bit on the cdf as well, so there's a lot of cross cross pollination.

D

Yes, I was also.

L

Also participated in some best practices meetings there, and I also think that this would be a topic which would be interesting for them for the cdf.

L

So it could also be that we will kind of join the effort with this working group. I think that would.

D

Be a really good idea: yeah.

L

So, um yes, I think we could reach out to tracy and try to find try to find out what what she thinks about this, because then we would have a lot of more a lot, a lot more people joining them.

G

L

G

The in the other, in the toc and other things and stuff, there are people from end users. We can try to reach out to them as well and just think somebody.

L

um Jennifer, I think you you what you were active in the end user community around you.

G

Yeah I was when I was not a vendor. I'm trying.

B

G

Reach to my colleagues to see if they are still part of like they renewed their membership and stuff. um I just been one of them just waiting and I know also some people that work uh cncf I'll just check if they they can like, recommend someone to talk to or something out just directing them. I think I I have had good experiences pinging people because by themselves, so they they are receptive. Sometimes they they just don't answer, but one or two answers.

G

So we could try to get some some use cases um does it have to be just by end users like if I talk to another company, that's not yet, but they they would like to share the use case. Would that be you okay, or do you think.

L

It's absolutely no problem, so um I I will also uh try to find some use cases um to be. uh The company I'm working for is also a vendor, but I'm acting as an end user. As I always said before,.

G

Yeah yeah, I will have a discussion tomorrow as well with my couple of people, my team, uh because um one person in my team is the chair for the state service catalog and they are trying to like they are looking into this community's service bindings back um and then we are going to discuss if there is any thing that's like I like, such as use cases or any shared interest that could be contribution or something I I want to understand more, but I'll have a chat.

G

Tomorrow too, we can so I guess, as an action photo. So we will reach to these people try to find use cases. Some feedback um should we, like alex said, start a charter draft, and we can also give our you because we thought a few use cases right. You had a problem thomas you you mentioned on our uh working group proposal doc, so we could start with that, put something and then um take it from there and then start contributing synchronously. So we don't need to arrange a zoom meeting straight away.

L

No, not really so. um um Yes, I can take the task to start writing the charter.

L

And to um yes um I'll start, writing the chart and share it to you both um afterwards and when we agree that this is an uh a good starting point. We can share it on the on the mailing list on the slack channel, but I would do this in a uh in the next few, uh so I would like to share this in the next few days if it's yeah possible good. I think that makes sense.

L

Okay, then, I think we'll get back some minutes or ever.

D

Thanks very much see you later thank.

G

You very much bye.

D

G

The presentation bye.