Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2018 (Copenhagen), 7 May 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: WG App Def - Intro – Antoine Legrand, Red Hat & Gareth Rushgrove, Docker (Any Skill Level)

Description

Want to view more sessions and keep the conversations going? Join us for KubeCon + CloudNativeCon North America in Seattle, December 11–13, 2018 (http://bit.ly/KCCNCNA18) or in Shanghai, November 14–15 (http://bit.ly/kccncchina18).

WG App Def - Intro – Antoine Legrand, Red Hat & Gareth Rushgrove, Docker (Any Skill Level)

With Kubernetes emerging as the clear leader in orchestration, more tools are being developed to manage the lifecycle of Kubernetes applications. How can we help authors to coordinate their efforts to focus on smaller components that are interoperable with the rest of the ecosystem? How can we help provide a clean user experience around using declarative primitives? How do the community’s use cases affect their application lifecycle management needs?

Those are some the questions being discussed in the new “app-definition” working group.
The talk will explore these questions and explain what progress the working group is making on each, including demos of the prototypes, explaining the current proposals, go through what has been tried, rejected, accepted, and what we’ll work on next.

About Antoine Legrand
Antoine Legrand is a software engineer at CoreOS co-leading the working-group 'app-definition' and works closely with major actors in the ecosystem to improve the user experience in managing application in Kubernetes. Antoine is involved with open source and cloud native software projects, and for the past two years has helped the community and led related projects including Kubespray, kpm, and app-registry.

About Gareth Rushgrove
Gareth Rushgrove is a product manager at Docker. He works remotely from Cambridge, UK, helping to build interesting tools for people to better manage infrastructure and applications. Previously he worked for the UK Government Digital Service focused on infrastructure, operations and information security. When not working he can be found curating the Devops Weekly newsletter or hacking on software in new-fangled programming languages.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.

A

Yeah, hey hello: this is a the app definition in workgroup. Are we calling it's a deep dive session, just an intro? Oh just an intro. So this is the app definition working group intro session glad you all came to this last session on this long week. I'm Brian Laos- and this is our.

B

A

So what we did is actually more entwined, then I prepared some slides to show you what we're working on and how we're actually trying to contribute back to the sick, apps community. Sure.

B

So yes, the working group started like six months ago. Around three mission. One is to improve the class affirmative in committees. Other one is like to understand the ecosystem, so beside em like why there is so many tools what the community is trying to solve, and the last mission is how we can make this tool work better together. So what's missing to to interpolate them clear.

A

Is from what we understand, there's 60 70 plus tools, doing something in the app ecosystem and we're just trying to figure out what are we all doing.

B

So quickly to align on so watching on cigars, we define like some user profile, so the application developer writes a code for your application. The application distributor, so example in M will be like we write to the chart will write to manifest and the application operator is like would deploy them so sometimes many times, people have to say different ads, but make easier to agree on on term so well.

B

What we, what we are trying to serve is like the Abdi Twitter like how to better improve writing application and operators already have plenty of tools to deploy. Six, yes, okay, so for the ecosystem, yeah.

A

Earlier this year actually been working this for a long time, we went to prepare survey. Actually, the survey results will out soon.

B

Yeah, the raw data is the.

A

Raw data is actually out on the mailing list right now, but we need to summarize that and present it in a better fashion, find the key trends and also we are looking at the landscape. Just seeing really a good example would be we have. We have applications that are for generating optics and we have objects, our applications that are for applying optics and some that are mixed and really what we're trying to do.

A

As you see this, it's a very long list of tools that are written using quite a few technologies, just trying to find some some common patterns between them and then, ultimately, maybe we can do some things for the community. Like saying we have sample apps right now. Maybe we need a better sample app that all the applications can use to to demo their features, rather than just saying, I can do this.

A

I can do basically deployment service, so we're just trying to level the playing field just so the actual users of the applications know what they're going to get I'm.

B

Not sorry where.

A

There's a reason why Antoine is touching this I don't understand how to use these. This application.

B

Yeah I think this is where I failed. Okay, going back to the slide. Okay, so yeah for underthings a good thing. We did try to figure out different tools and did the survey also. So the survey is a rigid we're trying to extract, but some some data, so first like mini but simple, more than 400 something and yeah some data, so more than 90% are using keep scale and are very don't look to something else.

B

More than a third are using em and also third are interesting in him, and the last are using more than 80 different tools beside yeah.

A

I think the most one of the interesting things we picked out of one of the slides in the actual survey was 55% of the people using kubernetes are using jobs and in current jobs, actually, who here is using jobs and cron jobs. So this is what we thought and somewhere out there in the in the ecosystem.

A

There is a small group who is all using jobs and cron jobs, but we really want to know just for the ecosystem, who's actually using these tools, and if, if people are using these tools, actually how can we make it easier or if that's, if I mean these objects are actually pretty great, more people should be using them. So we should be talking about them to talk and document better. So.

B

For example, if you're interesting, it's like who's using job and why it's just like so question was not only do you use job is there is like a twenty five or six following question? What's your use case, it's rough, so we can dig into the survey to understand what really is a use case for each each question was like quite complex array.

B

Okay, so that's that's more with graphic. What we say.

A

There is actually there's a lot of questions in here about go ahead.

A

C

Is application equivalent to a bunch of manifests that you apply to some cluster and then you I, don't know, delete them or update them in place? So is it the cool in terms a group of resources and application.

A

So you are not the first person to ask that, and and you're probably I've heard this question a lot, and actually there is a project. It's probably later in the slides we actually there isn't. There is thought around. Creating an application object to actually define what an application is, so it could be a single manifest. That could mean could be a single object in a manifest, but it could be more complex than that. So there is work coming out of the work group right now to actually codify that and bring it together. Yes,.

B

So far, we just like the committee refused to define what an application, but this is that girl I demand things to represent an application, so we are working on what what could be yeah.

C

But if you don't want to define what is an application, then maybe we could define what what can we do with an application.

B

True, but so if we go back like through the simple definition, what's an application is a bundle of resources. So the work is a reside slide for this, but the works like oh, we can find this application, this resources in your cluster to regroup them in one application. So the.

A

Theory the problem is, is that an application is a very nebulous term. It doesn't mean it means if I were to ask everybody in here what an application was. I might get a different answer, so a good example would be a wordpress site. I have a wordpress application. What is in wordpress application?

A

It's uh it's the wordpress PHP app, but it doesn't work without my sequel so and that's actually deployed as a staple set in the cluster and then maybe we haven't maybe have an engine an engine x, ingress in front of it, so we can actually get to it.

A

That is our application, but other people might say well just the PHP WordPress app is our application and what we want and what we don't want to do is say no, no, no, we're gonna, give you it what we're, but what teams to have the flexibility to pick and choose what their application is and once they've identified their application to be able to operate on their app or application as a whole?.

C

What do they get out of this? So when you give them this flexibility, FLE what-what flexibility, do they actually need? So what will they do with this application thing?.

C

A

Of the first things that comes out of it is that there is no concept so with labels and annotations. What we're trying to do is well there's a couple things here. So if, with labels and invitations, if we had a, if we had a single way for defining an application, we could actually operate on all the components of an application at the same time, but there's also since, since we have an application object that is coming, that will probably come as an extension. Maybe papi I see I think it's coming as a CRD.

A

Yes, what we'll be able to do is do cluster operations based on that application. So maybe we could. We could actually someone, if you're, trying to monitor your application as a whole and determine if the whole thing is up. You don't have to identify all the individual pieces. You could actually as a tool say: I'm Monica, Bernoulli's applications, and if you have this application, CRT defined then I can just talk to that and I.

A

Don't have to worry about all the rest of the components and you don't have to worry about applying labels in a certain way, because it's all it's a standard, and now we can just take care of it.

C

To summarize, what I understood is managing parts of the application together components: individual resources, like ingress service, a config map, a secret deployment, so the standard way of grouping them and operating on them as a whole. Monitoring them was deploying them anything.

B

We have slide later so we coming back to the to the to the.

C

Okay, thank you. Yeah, we're sorry.

B

Yeah, so so ecosystem is frequented. This is like many tours used. Ok, so one of the mission is like to improve primitive. So it's a bit like in the working group now there's like project name customize, it is like you, you JD is like you writing a cube. Gl create config mark, for example, so how we can capture this in a declarative way. So that's part of the work. So this is an example. We can just say config map create.

B

So this is things but also apply works for about when you do keep CL apply for fun. Now, it's like very doing a lot of logic. So we need to improve this and put it in server-side, so that some description, but the idea of a working group is like to start description and then to ship it to also sig. So now sickly. Isn't it yes.

A

To customize it's actually this tool, customizes customized with the case. Ok us do. Mi ze was created by the people who run cube control and they have these ideas. So after seeing this tool for a few years and not being able to change it because it is a standard, they have thought of different ways of managing manifests and then actually taking a manifest and applying transformations to it. So the working group is giving them space that they can actually experiment outside of what they normally would be doing and they're doing it in public.

A

This work right here. It's been done by the team that runs, keep control at Google, so.

B

Yeah, for example like if you see the community most using template, only so don't play it, go down, play Jinja, 2 or whatever template is or even checks on it, but there is like other way to transform resource all the way to define to other transformation. So it could be like patch, for example, so you can define like explain some overlay, so you want to modify it the payment or something you just write the patch. So this is like template less you just don't have to think about everything. A user can modify.

B

You just apply patch, so it's like some generation or knowledge about, or many people just change the level on annotation. So maybe, instead of forcing people to template every annotation every levels you can just say, yeah.

A

B

Put your object levels here and we're going to apply on everything. So that's part of the discussion we are, we can decide templates.

B

A

B

Customize, you can fight, for it will move out of the cube CTL repo, but you can try it already. So this is like exploring different ways and using templates I.

C

Want to ask the audience who uses other things than cube control apply to create or modify resources.

B

Brian is using different.

C

Because we only use cube, CTL apply, we don't cube city or create, don't don't do anything else only apply and well sometimes delete.

C

So that's why I think it's already declarative.

C

B

This like in this example in customer, is basically you're using qcl apply, so it will just generate a on side. You just customize expand, it will apply the patch client side and then you pipe to keep CL apply. So that's a way to is a way to easier and overlay things together.

B

Going through yeah, this is like a full previous customized things, so we can declare which resource we want to deploy. Something like in every template, I see is the name prefix, for example, say I want to deploy prod what price they have. What price was it just templates the prefix, for example, to to the resource? So you can like get like your deployment with prod what price as a truck. So this is like without template. We can just here you can just say name prefix, and this is going to be applied everywhere without template, etc.

B

A

B

Part of the ID, now the cube sickly is exploring and.

A

The important thing to remember is this: is a working group we're working to the there's multiple groups with different interests that are working to come up with new ideas? A lot of these ideas may be implemented in different pieces of code somewhere else. So since we are a working group, we're not we're not sig apps we're not actually maintaining official repositories, but what we're trying to do is figure out where, where are the best ideas and that's what a lot of this experimentation it might never show up anywhere.

A

You might never see this again, but we are still are experimenting with it. So.

B

Yeah that that's one of the go like you, we, we don't own any code in the end, so but it's a space, so people can try, IDs of so Chauvet, try them and then we ship to also cig like okay. Now it's sickly implement these c-cups we're going. We talk about application.

A

In particular, its its cig apps, so KPI machinery sometimes and then 6ci and those are the ones that we're really just trying to feed information back into.

B

So another part of it it's improved anti-piracy. So for this like we have like concrete action, one of them was some easy with no code is just like the Lebel recommendation, so it's not easy to agree on the set of levels, but this is a first step to align every tools say some some people, just that name also like cavities name or up so can we just agree on one thing: name your application or your version, and then you can build confidently when you create tools or UI or dashboard to think.

B

Okay, probably the user is going to get disabled, so we can have some special behavior special UI to set to extract these levels. Yeah.

A

This is really important for tools writers, so as a as a tool writer when I create an object, it would be nice for me to have some kind of default or I could actually say that this is a case on it maintain thing: it's not owned. It just case on it touch it case on it created it. So if, if someone comes through and they do a get on this, there was some other means either through the API or with a command-line tool.

A

They could actually see that this object actually was created with kiss on it rather than helm or something else. So.

B

This is like produce set of labelled, so it's going to be in the official documentation, the library and annotation recommendation. So, for example, here you have deploy manager. So we know it's been creating for a case on a tour. M2 is description about what's name or component so as there is a PR out there. If you interesting, you can comment on it before it's much.

A

This is a very I know. Some of you are seeing this for the first time. This is a very tumultuous subject. There was lots of debate on.

B

A

This is like lots of debate on this series.

B

Like five levels, but it took like it's, for this- is for four months so and yeah. One of the latest description is like. Do we need perfect like do? We want to prefix all of it? Yeah.

A

Because the standard says that any any label needs to be named spaced if it's being created and being used, but at kubernetes the io / name, is very long versus name and there's great arguments on both sides. But now we're working through this to see what we can do to make. This actually be a standard.

A

But I think everyone in here can agree that, even if you don't understand what every single one means it would be nice to be able to query those if you're looking for something and that's how and that's what the group is trying to give to the community and yeah.

B

And so, to give some insane for all the world community to a doula base is going to be implemented really quickly in communities dashboard and keeps it here or so to have some special feature like show me my component or get some special UI in 2012. So.

D

B

Was one work in the community just lost again, we keep doing it yeah, okay, one other thing was greater, so in in him is a shot hmm how to describe an application, for example or package, and this can be extend to any kind of package manager or configuration manager, even no package manager. Let's say you have just a deployment of similar services, American figma of the CML in a directory. You don't know what they are doing. Just / to descriptive.

B

The Chamorro equivalent of chato Gmail in this recording and the user will go to the drip, just open this file and understand okay. This is again so example. What press this version? It's right! You don't have to read to decrement everything, so we're trying to adjust part of agree on what a description of an application- it's quite simple, just like with the maintainer salvation what Cinco icon, where you can find the source etc.

B

So this part of the description so maybe at some point, get a descriptive spec that you can use in your go type to just agree. This is how you describe an application. Yes, okay, so we go back to the point application resource, so the last piece is also what's an application or we can group things together. Even for you, I say: okay, I, deploy with him or anything now, I have a bunch of secret config map service that are not related together. Oh I can create a UI to show them together.

B

Currently every up doing the own things just just own crg. Even in Korres we have our own TRD in open gyms. They are their own and every UI of their own. Basically, so can we agree on a way to just find this resource, and the proposal is quite simple: its base selector so just create our fans, which libido I want to find. So this also relates to the level recommendation. Now it can maybe create some default about that which component it's which kind of resource you want to find.

B

What are the dependency, and this is what you find your application. This.

D

B

Like opt-in, you can your pitchin will work if you don't create this, but now you can have like some representation about yeah.

A

We can actually create gooeys or better command-line tools. It says, let me find this application, and this is an interesting example, because we have this wordpress application that has a service and a staple set, but notice it depends on another application in my sequel. So now we can build these hierarchies and we can actually determine if I take this.

A

My sequel out of the app or out of out of service I can ask the cluster what am I gonna break, because now we have that object graph, and we currently mean you can build that, and probably people have built that I think I built something similar to that just the other day. But I don't need that if we actually have this application CRD that allows us to query and.

B

Yeah and beside to just so face information, it's awesome so there's an application controller to garbage collect. So now you can just delete this resource because it know also graph. You can delete your config Mike just recreate or together it's a bit like tighter, if you're using it. So this is one things also it's like to show up application level else so to, for example, to WordPress to work. You need to API to be to be app and medical to be app, so we're. How do you know that the boss are up now?

B

You can have its probe in the application resource? Oh it's not yet implemented, but that's the end goal. There are some special behavior in it, and this is going to be also integrated in cubes here and community that world very quickly. Yes,.

A

B

It's it's not as a CID, and it's probably going to migrate to the core like to the workload API or ship which keeps it ship with communities too. Yeah.

A

Yes, actually ask kubernetes, there's a lot of cigs and there is quite a few working groups, but as kubernetes gets more complex, you see. There's 43 people hundred people at this conference. There's a lot of interest- and this is the purpose of these workgroups- is that someone needs to sit down at least take a first stab at what if we all work together on this one specific thing- and this is what we've come together- to create yeah.

B

So we continue to discuss about this in a definition, but it also we ship it again because we all agree in the working group. This is where we want to go now. It's mostly gaps that, following it. Typically, if you have ideas about new things or we want to try out to improve life or if you have ideas about what's an application, we can just come to the working group, explain it we try and then we can ship it to so related, see.

B

Yeah, I think this is so.

A

Let's we yeah what.

B

A

Any questions or anything.

D

Thinking of dependencies, are you considering also the open service broker API so that dependency is coming via a service binding? So just keep it in mind. I think that's! My screen must not run in your cluster, but can you just come in where service binding so.

A

So, with OSB and service catalog I'm. Actually this is interesting. I.

B

Think areas like to represent dependency in terms of resources. So if you sing about cabbage collection, if what you will need to go through and also even if you don't get watch collection like where you see your application on UI, why do you click to see? What's rated? So you know SB who just gets like bind to something external? So it's not like really dependency. You need to get much back, so it's not like you want. You won't put it at the dependency yeah, maybe yeah. That's maybe things.

A

To say you should go to issues and raise that yeah.

B

That's a good like if you want to unbind when you did it this application and do trigger things there. That's a correct just.

A

A sidebar, OSB and service, and and whenever do to service catalog it's a this- is a very particular thing. If you're a provider, this is very, very important to you. But if you're, not a provider, OSB and service catalog are a little more a little fuzzy and what they actually do. So that's, but as a representative of a service provider, you should definitely be saying hey what, if we wanted to do this as well, because this is not how this started. Actually this.

A

This was proposed as an extension to kubernetes right out of the box, and there was lots of commotion. It was exciting couple days on the mailing list.

B

So yeah it should have started in core but say start as a adi get adoption and go back. So if you like this just say it so maybe we can.

C

This looks similar to close formation stacks on AWS. Let conceptually you see the same or am I missing something I.

A

Know this is a little bit different because with a CF staff on AWS, you are using it to bring up software. This is more of a description of what you've deployed. So I mean you couldn't there are some parallels, but you would actually need something to deploy your app. You would actually have to create these resources. So in case of that WordPress service, you would have to actually have some code, whether it be just manifests on disk or some tool, to actually create that.

A

Then you can create the application so to manage to tie them all together. Yeah.

B

But by itself it does nothing so it doesn't create the application it represent. It not yeah with some behavior so is. There is a controller that can now link on a reference to give a correct and check the ease of your application, but it will not you deploy it the same way. You have a deployment service. No, you are in service an application. If you want to link things together, so you still have to deploy the two first service and deployments yourself Oh with case on it or ham, but.

A

Actually, as this becomes becomes the CR D and then potentially one day moves in the core if you're, this is the incentive, and this is why that keep control. We see the shortcomings and keep control if you're, using a tool that I can actually do this. It could just automatically do this along with this deploy. It already knows what your components are, so it could actually write it for you. Ultimately, we would hope that people aren't writing these if they don't want to, but we still want people to be have the ability to do that.

A

B

So yeah you would do keep CL get application to represent, but, for example, if you're using hem, mp3 or whatever, if they agree on this, maybe you can generate this automatically or case on it can generate this automatically or so yeah. It.

C

Doesn't yeah don't.

B

Create your app just.

B

To speak about this, like in incorrect technique, we have something similar that actually creates application, but this is two different way to things and I personally prefer just to be loosely coupled to your resource lock. So you can choose to not choose on this.

B

That is another question.

A

B

So yeah the working group is, are every Wednesday wait.

A

Other when they were other Wednesday, so I guess it'll be next Wednesday yeah.

B

Next, one yeah maybe think.

A

So but we have a mailing list a so. If you go to the community page, you can actually see all the SIG's and all the work groups and we have a link somewhere. I just searched for yeah.

B

Ab definition and yeah.

A

So you can just go there and you can see our meeting times and we meet on I know it's noon: East, Coast, United, States time, I, don't know what time it is in the EU, but.

B

It should be yeah depend on where your lives for your it's 9:59. A perfect I'm just find this. That is okay, so.

C

B

Even if you just use case, you want to explain, like it's painful in our company, to do this in this. It's bring ideas to solve this issue or how you solve it internally, even if it's not open source. We solve this way, so we can just find maybe interesting ideas and I.

A

Believe, that's all we have right, yeah, okay! Well, thank you all unless you have any questions but either way. Thank you all. Thanks.