Cloud Native Computing Foundation App Delivery Special Interest Group, 10 Mar 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: CNCF SIG App Delivery 2021-03-10

Description

CNCF SIG App Delivery 2021-03-10

A

B

B

A

Good morning morning, well it's early in the morning in your time zone.

C

Yes, I'm in california, so.

A

Okay, so where are.

C

A

uh I'm from austria, so it's even my.

C

Time soon, yes and my daughter was in budapest, so I used to track that type of time zone for a while. The covet stuff brought her home. So.

A

Okay, so a few.

A

A

D

Hey um hi, everyone, I think. Well, I think I can start, and so I'm omer, I'm part of the culture of the this working group, thomas, is another co-chair here and jennifer is the third coacher. She had some maintenance at home, so she doesn't have an internet right now.

D

um What we'll do today is as usual short introduction and then talk about our draft question and task assignment and before we start john and only to pronounce it- and I froze right, I'm just.

A

A

So here so he freezes at some point. um Yes, hello um at first, I think so highly you are new here. Aren't you.

C

Yeah, this is the first time I'm attending this.

A

Yes, and where do you want to introduce yourself shortly so that we know and give us a an idea? What do you expect from our working group want me to go? First.

C

Yes, okay, all right um yeah, so I work at a company called infoblox. um We do have a open source project, infobox open! You can go! Look at that body of work, um we're uh we used to be a public company, but now we're a private pe firm owned and we have had traditionally an on-prem solution in the ddi space, but for about last four years I've been working on our sas solution which, as you would expect you know it's a cloud native application. We built it from ground up.

C

um We have something like over 200 microservices and a lot of tooling, for it is on the open source project, because when we started the project um you know on twitter and but if you were writing java, you know or node or some of those um or ruby there were some tool kits available, but for go at the time there wasn't anything. So it's a lot of tool. Kit elements are open sourced by us, although we don't evangelize it or anything like that.

C

But you know, since we wrote it, we put it out there, and so we've also been on this journey of kind of building the cicd tooling. Today, today we use spinnaker for our pipeline. We've been using spinnaker for about over two years in production and uh in the new target state that I'm working on. We want to migrate to a kubernetes native environment, we've gone through the whole.

C

You know docker swarm on mesos kubernetes journey as well in that four years, but we've always been cloud container based architecture so, but but different orchestration piece underneath, and so we want to migrate our pipelines and we've done a lot of pocs around a lot of our girl projects. Today we use our go workflow for our ti pipelines.

C

Looked at our go cd, flux, flagger, rollout, we've written our own operators, we're heavily influenced by the work and by the backstage community in terms of the end-to-end pipeline.

C

You know that kind of a declarative way to define the apps and the overall life cycle, everything from creating it to monitoring it to so that's kind of a target state for us to get to, and so I we just see this, um what I couldn't what I call an application operator, um it's kind of a critical piece of that, um and so we do have projects that um are using.

C

You know the various everything from the group builder to the s operator, sdk and and and others. So um so, let's kind of give you an overview of our company, myself and kind of the overall target state that we're trying to get to, and so I I've been attending the application working group, and so when this got started.

B

C

That was, at least um you know, and I it's been on my list, but my mornings are filled with uh scrums and meetings, so this today I didn't have anything blocking me from attending, so I thought I would. I would attend so.

A

Yeah so nice to have you here and thank you for the introduction. um Yes, we are very, very short, short uh small. Sorry, not that's more working group. um We are dealing with operators, one of our main goals, guarantees to get to to write the white paper about operators so that um um users end users and all, but also developers of operators, know what you can do with operating design. um What you, what you can take care care of, and so on.

A

um Yes, this working group, I think we are one year older.

C

Okay, the most the most recent operator we wrote- that's open source, actually is a cluster operator that we wrote because one of the goals we have is to be able to declaratively um create clusters on demand, and so we are. We use today cops in-house, um but um there's a lot of flux in the company I mean around using eks or using eks control, and I mean there's just and so our goal was and there's cross plane.

C

So one of the goals we had was to create an operator that kind of overlays and creates a in representation of a cluster that that you can operate on and then all the choices we make around kind of what the tooling is underneath um is kind of hidden from us, and also it becomes more multi-cloud so um so that that piece is um and so mostly in poc. Right now but, like I said.

A

So you might have a lot of knowledge about operators.

C

Well, like I said, we've written lots of operators so that I look at it as more like a black blank canvas and then.

C

You can kind of do something, um and I guess the tooling, for it, like I said over time, has gotten better from the point of view of the the community. But uh yes,.

A

Okay, um so um yes, our current state is that we are, as I said before, we are currently writing a white paper. um We are currently in the state where we have our first. I think our first draft deadline next week, um but we don't have enough enough content at the moment. Let's hit this way. So um if you want to um contribute something to the white paper you're more than welcome um yes, um so yes, let's start with with the rest of the agenda, I think john is also here. I didn't.

A

I didn't expect him to come today, because it's very early in new york times or I think- um and yes, I just took a look over here work. Currently it looks really nice. um Have you? Do you have any questions currently or um do you need some input from our site.

B

No, um so we're aiming to have this draft wrapped up this week. We were trying to have it done before today, so I could tell you guys excitedly. um We both got busy on friday last friday, so we're I'm, I'm nearly done with my section uh cameron's hoping to spend thursday on it. So um we should have that uh in a state previous review.

B

So I guess the question would be: um do you want to review it in the the google doc or do you want me to uh submit a pr in a markdown format or what sort of your guys's next steps.

A

um For us, it would be cool if we, uh if you could um put it in our in in full request, um so we and all of all of the other people could review it.

A

um Yes, we'll try to we'll try to um accept the full request as soon as possible.

B

I've um I've been trying to get more folks interested to join this working group. I'm not sure if that's been successful yet or not, but I'm trying to help.

A

It's perfect perfectly fine. Thank you.

D

We'll also try to make the tasks a little bit more clear. I think uh just something that came up in the sig.

D

We view the task and make them make sure they're more easy to pick up for new.

D

A

How could I say this in english? um I? I could not understand you, because there were some some pauses in here.

D

D

We will we need to go over the tasks and make sure they are good for new joiners.

D

Some of them are good to pick up very easily. Some are less and we'll do it this week to make sure that if someone is want to join easy to do that.

A

So the next tasks, we'll we will write, will be a bit more clearer and a bit more uh better defined. Currently, um just for your information, we we copied everything from a google talk, google doc tool to mark down into the repository and so on. So um we we shifted our whole workflow in between of writing the document um to get this a bit more structured and therefore the tasks were not very clear defined.

A

But thank you john forever for your contribution and thank you uh and yes, I I hope the security tasks were defined well enough. Oh.

B

A

B

um A little surprised at almost comment, I I thought it was uh at least for me. It was easy to understand, but maybe for others.

A

Okay, so um yes, we have. Today we have the nice task of assigning the rest of the tasks. I think I mean these are these are a lot um so at first. um So how do you want to write something for the white people or don't you have gotten enough time for such things?.

C

Well, I was going to start by actually reading it, so I just pulled up the get site. um You know I was just looking at the body of work.

C

I'm surprised it's not using um mark make doc, um but but it's fine the way it's laid out, um mk doc. I think um you can you can if you're gonna do something like this might be a way a way to organize it. It's just uh typically that's the pattern. People are writing um documents and get but I'll I'll read through the we have a table of contents I'll go through and and read and sounds like we're generally accepting pull requests for updates right.

C

These are sections that are empty uh right now and you want people if there.

A

C

Lot of sections, which I empty, okay, all right, but I'll start just by looking at the document. First time, I'm looking at it so.

A

um But please, if you, um if you plan to do something or to to write something, please contact me or.

A

Then we could block this uh this chapter, uh so we have uh we've got tasks in our our repository and then we will send this to you and and hopefully no one. No other one will do this.

C

Okay, yeah I'll look through the task too, then that's good.

D

Mk docs is for presenting the document in html right.

C

uh No it's in markdown it just. uh It creates a uh doc yeah.

C

It's still marketed.

D

C

D

To implement, but do you think it will be easy to implement that in our current repo.

C

Yeah, once you have marked down, you could essentially just uh we create a project for documentation and it shows up as a doc folder. The pattern is more, for you know having it coexist with other things and because this right now the whole repository is one big document.

C

So let me let me find um and there's also um material version of it for fancier looking ui. So it's let me find a link to it and I'll put it on chat.

A

Okay, um then, I think we'll try to um go over the the current tasks so um well, let's see we're.

D

A

D

As this month or I'm sorry, we're aiming.

A

For a draft to come out this month right on next be exact.

A

So it's a lot, it's a lot of work for one week but yeah, but I think it's like in like on university. um This also worked the work the whole time to be honest. um Okay, then, let's go over the tasks. I'll share my screen.

A

Okay, um so for everyone who doesn't know this: um yes, we have, we have a lot of document parts and these are different marketing documents and for almost all of these um document parts one task, one issuing in the top exists.

A

We have some tasks which are assigned already and we have some tasks which are not designed, and we should do this now. um I think uh we will include jennifer afterwards, but um that's no problem.

A

Okay, so, um yes, um we have.

A

We have defined a lot of capabilities. An operator can help, so um you might. Some of you might know that there are capabilities that are maturity, models of of operators and um there are a lot of capabilities defined and we as a sincere project.

A

We didn't want uh to um to say that an operator who has capability x is more mature than another one, um because he could lack of other capabilities right to only to describe the capabilities and operate the control, and this should give an end user or also a developer of an operator, an idea what you can do with an operator.

A

Therefore, we defined a lot of capabilities um such as the most obvious one to install an application to upgrade an application seamlessly, and yes, backup, recovery, auto remediation and so on, and for all of these capabilities, so for the ones which are not written at the moment. There are also existing issues such like the article of the configuration team, um which means that the uh that we want to describe that the operator is responsible for managing the configuration of the deployment application, so he might be able to queue application parameters if certain.

A

um States exist, so I'm sorry um and the customer should be able to tell the operator which uh which thing could be altitude and so on so to keep. This short um is anyone there who wants to deal with this section.

A

Okay, I'll take it okay, okay, then we have auto remediation sections, uh which we will describe that an operator could could be used to um yes, get get an application, I'll get notification to run the state from uh from uh more or less future state, and then we have backup and recovery. This is all this faster.

A

Then we have such kind of uh other related work section which um this should should describe, which work or already exists. So there is an announcement from the phone call is where operators are defined and so on.

A

um We want to give readers the possibility to get get further information about operators, because.

A

A

Okay, then we have somewhere. In conclusion, this is not the real problem. Now. Does anybody anyone want to deal with the technical application of a technical implementation of an operator? So how can I write? How do I start writing an operator? How can we test and operate the um we could this?

A

We could describe all of this um using on our uh using uh using a new application if we want- um and we could give some some example- implementation of different capabilities and but also uh I wanted to describe shortly what what could uh hard in the problem to operate. This could be.

D

I think we need to change this task because we don't really want to run to to have to describe how to write an operator. We have a section about uh implementations of operators, and- and so maybe it's not just a introduction to them. If you want to write an operator, you can use those tools or see those guides.

A

And well, just just about to take the methodology, how you could start writing an operator so not not regarding the exact tool set. But what should you consider before you want to write an operation.

D

What you should consider is a better way to to say that I think.

D

But it also big task and maybe we need to split it.

A

I think we won't have a new application now.

C

Not sure the white paper is a place to talk about implementation. I think I think, an overview of um of kind of the landscape in terms of um what's available and maybe some links to them. I think and like I think we just somebody discussed. Oh man, I think, discuss just the high level of how to approach understand the use case. I mean in some cases, depending on what you want to do.

C

You might want to write a very low level controller um versus um because you want more control versus uh something like operator sdk, where a lot of the tasks have been abstracted out and you just write a control loop um so but.

A

Okay, um yes, um I've I've changed the task a bit, so um we should describe on a very high later, so not not too much in detail. What should we consider? Where could we start when we, when you want to write an operator, so um this might also include um find out which transactions for you um that then there is something to test operators?

A

um I think this could also be a thing which, which could be interesting for present people um and yes, the implementation of different capabilities. I think we could drop this, but also, um I think kinds of of the requirements for operators could also be interesting to get ideas how such things could look like. So um we only should make um it's very it's very late today. So english, it's very bad, um so um we should give user user idea how a delivery pipeline could look hard and not how they should do this.

A

So um I think this would. This would be um enough overview for a white paper, but do you think uh what do you do, but is anyone anything missing in the technical technical thing or should we should we.

A

uh Should we rename the section or include it to the best practices section.

D

It feels like this section is like kind of summaries for the whole white paper uh kind of the short version of what, because, if you read the whole paper, you probably know what to consider, because you understand what an operator is.

D

So maybe we need to postpone this to have to after we have the best practice and more.

D

Content in other sections.

A

um Just a second.

A

We also have a best practices section where I think nothing of those um so there's some kinds of. um Sometimes the placement of operators is discussed so um that they are separated from the workloads and so on. Then we have some matrix. Then we have available politics, which always also uh sometimes points of discussion.

A

um Yes, best practices for operating in the implementation. This could also only be such things. um Yes, then, I think I think we will include this. uh This additional thing.

A

C

A

C

Umbrella operators, we look at them at least in our system as being federated. I think that's the term we use for those operators that are managing others.

C

How do you talk? How do you call them?

C

We look at them as being federated and there's already several groups working on federation.

C

You know at least at infobox. We have this concept of kind of clusters that are more like for control functions and, and so they kind of overlay or they they federate other uh clusters that are running workloads. So there's that pattern we established so so when so the I guess, the term umbrella does kind of indicate something over something else, but but just terminology, I guess, but.

C

So it's all terminology here.

A

Yes, we talked a lot with, I think, we've this past half a year half a year to find out what an operator.

C

But even the term federation means something fairly specific in cncf there's a group working on, I think, there's at least three groups. I know or 30 projects open source projects that are doing federation.

A

Okay, um then, I will close this technical implementation thing or we will remove the section in favor of the best practices section uh if it's okay for one more.

D

I think so we'll probably need to revise this decision after the first draft and maybe we'll decide that there needs a place for that. But for now it seems like more content that we might have in other places.

A

Okay, then, I will remove this uh remove this chapter and um the content will be moved to the to the best practices.

A

Okay, then we have use cases for an operator. This is more end user related um injector. This will be done by jennifer, so she wrote that she will do.

A

A

Then we have have such funny thing is the lifecycle management of an operator. So um it's about updating an operator itself and.

D

Can you zoom in for the do your browser? Oh sorry,.

A

Can you zoom in the browser? Yes, yes,.

A

Okay, great, oh, it's a bit bigger! um Yes,.

B

Updating an operator.

A

A

And um and everything other which could be, which could be the scope. um I think this is mainly targeted to operate a life cycle, miniature things and so on, but I'm not sure if there are probably switched out which do similar things.

D

I'm thinking maybe it's a little bit technical talking about updating an operator auditing.

A

D

And I'm not sure what updating an operator itself is just updating an application. How can we describe it in a way that.

D

We want to say you need to think about versioning, your crd, for example. You know how to do that.

C

I think this piece is here because there is a operator. Sdk has a life cycle management piece. Maybe that's why this was put here just to maybe capture that element that comes with that environment. I think opera ck, I think, is a openshift project, and so they have a whole bunch of machinery around managing life cycles and so.

A

Yes, so this section was a bit targeted to this, um because there is such thing as an operator lifecycle manager, and I think um users of operators should know what this does or how such life cycles could be managed.

C

But that second bullet is the main problem, not for operators, but in general, once you have crs and kubernetes, you know, there's no good way that kubernetes allows you to manage them. They're not managed elements in the system, and so it's almost like, I wouldn't say, like very implementation, specific how you go about managing crs, and so we have a lot of yeah.

C

So that's that's area we deal with uh both for our our normal applications as well as any operators we have is just almost they're, not being they can't be managed as part of the application. They just need to be treated as their own kind of resources. So that's kind of the problem today.

A

Okay, so um does anyone want to deal with this section or whatever? This will be in the end.

D

I can try to load it button. Yes,.

C

Well, it's like that. Last section we talked to on implementation. It seems like it. um I don't know that you could. You could um as part of a white paper or really work on it unless, um unless you almost picked the implementation and then you could maybe drill down awesome.

A

A

A

um So try to get. uh We should try to get some some some uh some sections on this, so one two or three and um we'll see afterwards in the draft. If we need this or difficult difficult practice or if we, uh if this will be discussed very often roughly during the reviews and if there are additional things to consume.

A

Okay, then, we have operator frameworks phenomenal platforms. um Do we want to deal with this because that is explicitly um not in our house within the chapter? As far as I've seen.

D

I also think that we, the section where we're talking about operators in general and not related to kubernetes, is enough, and we don't have examples for non-kubernetes platforms, operators.

A

Not that we know about them, okay, um then that's the same thing as the as the technical implementation shaped. I will drop this and remove it from the from the.

A

Building, okay, but we have operator frameworks working with this and I think we have to. uh There are two operator frameworks described at the moment in a moment. So this is content cube builder.

A

I think we should create an issue for describing the operator sdk.

A

What do we agree?.

D

Yeah, we will probably find someone who is part of the team.

D

A

A

A

So this is, uh I should only give a rough overview, not the detailed uh description, how this, how this can be can be used.

A

A

A

New tasks to the uh so I I will post them on the spectrum today,.

D

Yeah, I think we had some people from operator sdk along the way, so I hope they were in the channel, but if not, we can find them. Okay,.

A

And a second thing I want I would like to have in the white paper is, um I think, what is it called kudo? um So, let's, let's try to find out if someone wants to.

A

Description about.

A

A

Okay, it's also submitted okay, so these are the these two sections.

A

um Are you where we have any other operator frameworks? We should write a description.

D

A

Okay, um yes, I will take the overview task for.

A

A

Okay um and then we'll write the executive summary in the end, so um I think we will give this task to jennifer if it's okay, for you.

A

A

Auto configuration tuning, I think I will try to try to do this and this will be it um okay. So um so so, um if you want you can uh you can add something or some some chapters or sections to the white people? But please, please don't hesitate to ask us if you, if you have questions or if you want to to have some sections.

C

Yeah, I'm going to read it more, I'm trying to understand the big picture that this thing is painting because, like I said in our environment um like I would like to see, if there's like a some sort of conceptual data model of how you think you see the environment, you know how does the operator operate on it?

C

Like um you know like at um you know, we look at some things like, like. I talked about like our argo, rollout or flagger as already being application operators. Okay, so the question is: um do I start with that and enhance them? In other words, or do I build my own when I build my own?

C

Is a single operator have a concept of multiple application crs that it manages, or is it one to one, and I mean those are all kind of like things that I would think that I'm gonna read through and but I mean those are in my mind, kind of fundamental things that I start with always and say like okay, then I can go down and so a lot of what we were looking at were like the uh leaf nodes.

C

Okay, I'm just trying to read alex I'll I'll read the document you guys have been working on it for a while, and I try to get the big picture and then we're appropriate yeah. I will sign up and.

D

D

We have some section while we're talking about the concepts of operator in general, so that will be good to validate them and see if they match what you accept. What you expect and also open vrs for changes.

A

A

I have the I'm, I'm thinking you have a lot of uh you have you have a lot of use cases for operators in mind and if you, if you want to, if you want to write one in in some in some lines, um don't hesitate to do it do it. We will accept the full request. I think um such things as you said before, um should I write an own operator. Should I extend another one? This could be a very, very important use case for individual for for available authorities.

A

So I uh I think, one week ago I had a similar problem, where I also wanted to do something and also try to find out if I should write the operator myself or if I should- um let's, let's say this way, do a word from another one, so this could also be a very valuable use case for, for readers.

A

Okay, so um yes um do you have any other questions or or comments or other.

A

Things, yes, then, I think we're right in time. um Thank you for all payments. Now the next meeting will be in about two weeks, so I hope then we'll then we we have to discuss the first craft and yes thank you for saying. Thank you for joining me.

A

A

However, this is cold all right. Okay, thank you.

D