Cloud Native Computing Foundation KubeCon + CloudNativeCon North America 2019 (San Diego), 22 Nov 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Tutorial: Deep Dive into the Operator Framework for... Melvin Hillsman, Michael Hrivnak, & Matt Dorn

Description

Join us for Kubernetes Forums Seoul, Sydney, Bengaluru and Delhi - learn more at kubecon.io

Don't miss KubeCon + CloudNativeCon 2020 events in Amsterdam March 30 - April 2, Shanghai July 28-30 and Boston November 17-20! Learn more at kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects

Tutorial: Deep Dive into the Operator Framework for Kubernetes - Melvin Hillsman, Michael Hrivnak, & Matt Dorn, Red Hat (Limited Available Seating; First-Come, First-Served Basis)

Please bring your laptop fully charged as we will have limited charging stations available in the room. This is an entry-level tutorial session for both application developers and system administrators interested in building and managing Operators for Kubernetes environments. It is designed for those who have a basic knowledge of Kubernetes and want to learn how to apply domain or application-specific knowledge to automate common operational tasks. Attendees will understand the past difficulties with building Operators with existing client-libraries and discover how the Operator Framework can ease development workflow. Step-by-step guidance will be given on the process of creating real-world Operators with Go, Ansible and Helm charts while mastering methodologies, design patterns, and strategies that can assist in avoiding common pitfalls. Attendees will use the Operator Lifecycle Manager (OLM) to define, install and upgrade Operators as well.

https://sched.co/Uaet

A

Hi, everyone welcome, hopefully you're in the right place. This is the operator framework workshop we are here to this is one hour, so we're gonna be going at a very hour and a half. That's right! That's right! Thank you an hour and a half. So we're going to be going at a very, very quick pace.

A

We're gonna learn all about the. What operators are what the operator framework is, how you can use it and for those you that use kubernetes every day, guess what you might learn a few things bug better going on behind the scenes that you never knew about before. So, let's go ahead and do some introductions. My name is Matt Dorn I'm with Red Hat I was previously with core OS and part of the core OS acquisition. Before that I used to work a lot in the OpenStack world and I'm happy to be here, Lily hi.

B

I used to work at operators, decay, I'm, moving on to monitoring I'm just here to help out ask any questions. Answer anything really: yeah I'm an engineer at Red Hat. Basically thank.

A

C

Hey I'm Michael rhythmic I've been at Red Hat for almost eight years in the criminais space for several of those years, work it on the operators DK and before that a thing called the ansible service broker, which is part of the whole Service Catalog ecosystem and kubernetes, and it just generally been in the space of automating workloads on kubernetes for a while, most recently, I've been involved in bare-metal support so running kubernetes on bare metal. It's pretty cool turns out.

C

You need a lot of controllers to make that happen too, and we're using this same technology to do all of that very happy to be here and I'm. Gonna talk with you about building ansible based operators in a little bit in the second half of this talk great.

A

And we're gonna be going very, very quick, so I do ask that you know if you have any questions. We were gonna have a Q&A immediately after, but please come talk to us. We have Red Hat people here in the front. These people also deal with Red Hat certifications of your operators. They can talk to you about that and we're here to help you if you need anything. So please don't leave with your questions unanswered, we're here to help you okay.

A

So let's, let's get an idea of what's going on inside the room here, how many people here use COO, Nettie's and OpenShift regularly? Well, we are at cube con, so I expect there's gonna be a lot of hands up in the air. Okay, yeah most you, okay, how about this? How many people here develop and golang weight regularly? Okay, great okay, that's like oh! Is that a quarter maybe and the people um yeah, but a quarter. How about this one use ansible regularly Oh way! Do we have more okay? Do we have more ansible?

A

People in here are more goaling people you think based off about the same okay. Okay, all right and anyone here attended to build or actually did build. Maybe you didn't get frustrated and get and gave up build a custom, kubernetes controller from scratch.

A

Custom community, okay, a few people in here; okay! So for those who have been hearing all about this controller stuff and you're, not really sure the internals work. This is a good workshop for you to be, and is anyone here, tried operator, SDK, ansible operator or the hell map operator, yet one two, three, four: five: six: okay, cool few people. So, hopefully you know today you are going to use it and get a get. You get a feel for how it works. Okay, so our goal here is to help you succeed with our tools.

A

So in this entry level session, we're gonna be talking about the operator framework, okay, what it is why it exists all that good stuff support channels, because you're gonna need it once you start diving deeply into operators, Google Groups, we have setup for you. We also have kubernetes slack. This is the official kubernetes slack. If you go to that channel communities, operators we're always in there hanging out and we can help you okay and the openshift commons operator framework. This is a group that gets together every third Tuesday of the month at 9:00 a.m.

A

Pacific to talk about all things: operators, maybe you're working on an operator, and you want to share it with the community and see what they think about it. That's where you do that. It's a live meeting. Okay, so, let's start out with what an operator is, but before we do that, let's make sure I haven't. We were talking right here. Let's make sure the lab environments work. Okay, let's make sure the lab environments work. So what we're going to do is work. I'm gonna show you what URL to go to here.

A

Let's make sure that stuff works go to learn, open ship, comm, slash, training, okay, so for those you who just showed up the reality here is that you are using a dedicated instance and a dedicated OpenShift instance running on ec2 on Amazon ec2, it's embedded into a program into a web app called cata Kota, which, if you haven't heard a cat, ACOTA you're gonna experience it now. It's awesome. It was just acquired by O'reilly.

A

When was it, thank you. No, it's it's training here. It was just acquired by O'reilly a couple days ago. So congratulations, the category for that I'm. Gonna, give this a second to see if it comes up again, the link is.

A

A

Learnin ship comm, slash, training, learn OpenShift comm, slash training. So what we're gonna do here is a little smoke test. Alright, go ahead and click on start training here. We're gonna need feedback from you all to know if this stuff works. Okay, click on the first scenario here start kubernetes, api fundamentals, click on start scenario,.

A

Okay, it might take a while then you're, gonna click start here. It says connecting because it's provisioning the instance it's spinning up on demand how many people currently on their screen. It says connecting its you're waiting for it to connect how many people calf you people over here. How many people have it's up and running how many people, these people- that didn't say anything what's going on?

A

Okay, I didn't realize that was that okay, blank screen? Okay! Thank you! Okay! So give it some time. Okay, because we provision this thing for 500. It's supposed to be able to handle capacity about 500 people in here right now, we'll see what happens there.

C

They're showing up slowly, okay,.

A

A

Let's go ahead and let that run while I go through the introduction. Okay, for those of you who have yours up and running, feel free. You don't want to hear me talk play around with this. This is pretty cool. It's just a little smoke test, we're creating a pod right, I'm sure we all know what a pod is in here with a couple containers inside so have fun and play around with it. It is an actual instance feel free to blow it up. If you want okay, all right, let's go to the intro here.

A

What's an operator now, look, maybe you were probably here like oh yeah, why are you going through what an operator is look reality? Is there it's totally possible that you all have been walking around this conference everyone's talking about operators? You don't even know what they are right, which I don't blame you? It's like it's kind of hard people throw in all these terms. What are we talking about here? It's kind of hard to there, two out there, the actual definition.

A

If you were to go to look it up would say an operator represents human, operational knowledge and software to reliably manage an application right. But what does that actually mean? What are we talking about there? Let's go back a few years so, two years ago, back in November to 2016 Brandon Phillips, the CTO of core OS, you don't know core OS a little startup right in San Francisco that specialized in something called container Linux an operating system. That's sole goal in life was to run. Containers right says talks about this on their blog.

A

This idea of this new piece of software called operators right, and it says an operator builds upon the basic kubernetes resource controller and application specific knowledge concept, those three things so an operator is nothing more than a nickname that fits this criteria. It's got to have these three things to be an operator again, operators, just the nickname people throw around. Is it an operator? Well, does it have these three things, what three things resource, controller or knowledge- okay, which what's going on here, like it, has to have these three things? But what are these?

A

What's a resource and kubernetes? Well, we should know it a resource is in cuba, Nettie's right a pod. We know a pod can have one container multiple containers right. We have a config map for application data, their key values route. For those you don't know what a route is. That's kind of like an open ship term, but it's the same thing as ingress right. You need a way to get people access from the outside world. You know access to your application, that's what a route provides. Okay and two. We have controllers controllers.

A

What are controllers and kubernetes I know you all know what controllers are right. The replicas set the replica set right when you specify the number of pods that you want running and it ensures that that number of pods are always running. That's what's really cool about replica sets. We're gonna, learn about why studying the replica set is a great way to learn about operators. If you study the replica set, it really helps. You understand what an operator does right.

A

What else is daemon set? Who knows what a daemon set? Isn't here right, Damon said they used a run on every node in the environment right and typically, they have host level access. They're, pretty cool I like daemon sets and then deployments deployment before deployments came around. We had to do client-side rollouts we had of. If we wanted to roll out a new version of our application right, we had to actually go and manually client-side send an API call to create a replica set number two and version number two, and do that deployments are great.

A

They manage our replica sets for us. They do this server-side rollouts. For us, you can determine the availability or specify the availability there cool. So what's domain or application specific knowledge. What are we talking about here? We're talking about knowledge, specific to your application, knowledge specific to your application, that only you know, because you know your application inside and out. You know all the the knobs you have to turn.

A

You know about being woken up in the middle of the night and you have to maybe SSH into the server and then modify something in this application. You know all its quirks all its things. We're talking about knowing how to install it, how to scale it properly, what I mean like scaling properly as if it's a stateful application, not just stateless lis, like a replica set.

A

You know if it's having one pot here, one pot here, one pot here right, scaling properly self heal self, heal, so aware what something's going wrong, cleaning up, updating, backup, restore, etc. Okay, so we're going to dive more into this still high level. I know you all know what this is. You probably look it up all the time when you want to create something in kubernetes, I'm sure you just go and you Google, you know okay deployment and then you just who here uses the explain command. So here knows the explain command.

A

If you don't know the explaining command, you should be using that every day right, they explain command. Actually coop CTL, explain deployment, COO, CTL, explain replica, set that thing talks to the API right and get you back information on how you can actually fill your declaration right out. So in this manifest. We have here the way to think about this. When you look at this in the future think about it as four different parts, you're looking at four different parts here, the first part is the gbk or a group version kind.

A

We also call that the type meta- that's the group version and kind here. We also have the metadata or object meta. You know about this right. You can put a namespace in there. It's your! If you're, not scoped, to a specific namespace when using cube CTL, it will our view our scope to a specific namespace when using cube CTL, you don't have to put a namespace in their annotations.

A

Go in there, the owner references end up in there, which we'll talk about later labels we can duplicate, though we can put a lot of stuff in there, okay, so spec, what about spec? So in our spec here this is our declaration. This is what we want right. So what we're actually doing is we're saying I want to have make this replica set that has five replicas right and then we can specify one container or multiple containers. This is your declaration. This is what you are requesting right. This is your declaration.

A

What you want now the fourth section here is status. It seems, like you know, a lot of people that get new to kubernetes are kind of like well I assumed when we talk about operators as kind of comparing the actual state of the cluster to the desired State, this being the desired state that this is the actual state. A lot of people think that, but no the status is only written by your controller and it's a way to provide information to your users.

A

It can also, if you're, creating, maybe a more advanced operator, be used as a way to store previous old state. So if you have to you want to check and only do something- and this is in ready status- your operator can actually look at that and decide whether or not to proceed if that's in a certain status right. But my point about talking about this is that this is something that a user does not fill out, that a human does not fill out your controller.

A

Does it to provide information to you, the human or perhaps other operators running in their cluster operators. Take advantage of custom resource definitions, okay, so those who have used these before you know that these things are incredible right because they allow you to extend the API via the API okay.

A

So you don't have to actually go and grab the kubernetes code and then like, for example, if you want it to add a new resource like let's say you want it to add a new resource called coop con, so you could do like coop CTL get coop con or something right or have an end point. If you're visualizing, the API endpoint right, you're visualizing, like B 1, beta 1, slash like namespaces, slash, whatever slash coop con like you've the actual resource you don't have to actually bring down the code to make that change.

A

You send an API call to modify and create that new resource on the API. That's pretty powerful command right, because I mean it literally impacts the entire cluster and everybody using the entire cluster. So here we have an example: custom resource just so you can see what one looks like this one is two called kinder fauna. What do you think this does? If you had a guest, you had to take a wild guess. What does this probably do? If you were apply this right? It creates a group on an instance of something right.

A

This is just an example of showing you this. You could see, though, there's some configuration parameters in here right, so it's kind of like, rather than using, like a configuration file or a config map and having to mount that config map as like a volume mount right, it's like a deployment or whatever like we're, just specifying what we want as a declaration and the spec section here right and there were cool thing about creating si RDS or a custom resource is, as an author I can actually design.

A

This I can design what users can put into here right, which is really really neat. So let's go ahead and just show you an example of let's extending the kubernetes api by creating our very own object or resource via CR DS.

A

So for those of you, who've done this before just go play on your play on the lab environment and look around, and you know probably boring you you've done this, but if you're, never, if you're not really familiar with CR DS and you've been in the conference for the past few days and you're like what is this thing, CR D thing anyways. This is basically what it is really straight forward, creating the CR D. So what we do here is we create the CR D.

A

First, the custom resource definition and literally the kind is custom resource definition. This is the first step to creating a CR D. You can see that this right here is the customized piece, my sequel's I'm, just making that up I'm going to create an operator that, let's say, creates my sequel, easier said than done as we're gonna see. But point is as I'm creating my custom name here and my custom group here in here.

A

What's the group, the group is nothing more than a way to disambiguate different kinds that may have, or let's imagine, that they're in companies all come out with my sequel operators. Right and they probably all create the same kind, my sequel right. Well, how do you tell the difference? One's gonna be company, a one's gonna be got company B right so that B company, a Company B and then here we're saying here the version. Well, this one starts with v1. I should probably fix this slide, because what do we want to start with?

A

Usually if you want to start with V 1 alpha 1 right, that's what you want to start with. Usually so I should probably fix that. So don't look at that right now: ok, scope, namespace, scope, namespace! This is saying after I create this custom resource deficit definition which, by the way, let me ask you a question: what do you? What do you think that this object here is gonna be scoped to first of all, do you think you need privileges to do this?

A

Would you think you need pretty wide privileges to create something like this you're modifying the kubernetes api? Think about that? Yes, you need some pretty cluster wide privileges to do this right, you're literally modifying the API, but then this block here says scope namespace. Why are we talking about scope? Namespace right, namespaces are how we do multi-tenancy and kubernetes, because after we create this, and we want to create the kind, my sequel object right, we're going to want that scoped to a namespace.

A

So if we have example, my sequel- that's gonna, dis a namespace one, but if I want to create another example, my sequel, that's the name literally. We know that we can't have multiple objects with the same name right, so we gonna put that into another namespace. So that means it's gonna be scoped to the namespace. What do you think the opposite option of this is, if you had a guess cluster, why the exactly right very good yep.

A

So this is just a couple ways to refer to it and some aliases that we can refer to, which is very awesome. I can say, coop CTL get MS and it understands what I'm talking about. How does it do that? The client does discovery right right when you type the word, it's gonna scan the API for the alias or nickname, and it puts that in the dot. What is it what's the directory? Do all those with a dot cache or dot yeah the coop needs a coop CTL.

A

Is it a dot cache directory that it puts it in I? Think it's a dot cache directory or something right, yeah.

C

So we're aware that the environments are having a connection error right now: I've reported it to Ben and we're gonna try to get that fixed as we're going so hang tight as they say in this environment with that, hopefully we'll get that back up. No.

A

Thank you appreciate it. Thanks.

A

And then you create the CRD after we verify the creation of the CRD. We then go in and we say, oh see, get CR D and how do we know it's there? Is it there? Yes, it's there. We say, oh so you're wondering probably wondering what's OC. Okay, so OC is the OpenShift client all right, maybe I'm a little biased. We originally an open ship, it's just like coop CTL. What is it? What is what do these things do behind the scenes?

A

You're just an API calls right, they're just talking to the communities API, so OC is just like coop TTL OC just has some extra open, shifty stuff added to it, but it pretty much acts like coop CTL same thing. You can use both interchangeably. Okay work really great. So we're saying coop CTL get CRD and it comes back and goes. That's your CRD. Let's now actually verify our new my sequel resource and object.

A

So what we're gonna do now is we have to see our D created check this out now, I can actually go up here and say: OC or cube CTL get my sequel. I can actually say get my sequel and what happens? Is it comes back and says no resources family. Everybody thinks at this point when they see this they're like oh, it didn't work, but this is a good thing.

A

This means that behind the scenes, coop, CTL or OC talked out to the API and actually found an end point when I say end point: it found a Ning spaces, /v 1, alpha 1, slash whatever right. It's a slash, my sequel, it's bound it, but guess what there were no objects in there. Yet we haven't created one yet so let's go ahead and create an object. So when we create this object, it's gonna look like this DP'd example.com class v1. It has my sequel in it. It says: name: WordPress user password.

A

You may be thinking like. Where did you get this stuff from? How did you? How did you know to put this stuff in there? Well, let's just say that you can actually customize your custom resource definition with the same information you get back when you use OC or coop CTL explained so for those you who use the explain command. You can inside your CRD, you can actually program or customize what comes back so I could actually have it's.

A

A coop CTL explain my Segal and it will tell me okay, you can put you here, you can put a password here, you can put you know food here or whatever it will give you and you can program it to give you instruct instructions, but the reality is that this right here you know this is not really doing much yet we're just adding this to the API. That's it. You can probably guess what we're doing here, we're trying to create a wordpress database right.

A

This is an example, so we go ahead and create that object after we create that object. We say: coop CTO get my sequel, it comes back, it says WordPress. If I want to actually look at the object. My object is actually in here. It looks good. It's persisted on the API, thus persisted where I see it. Where is it actual? Where is it? Where is it sitting where's the source? That's exactly right! It's sitting and EDD right, the greatest key value store.

A

Okay, so that doesn't really do much for me when it comes to an operator, it gets me done with number one which is number one. What's number one required for an operator, I need a customer resource definition right, yeah, that's right! So it gets me that that's good but a custom resource. It needs a controller to act upon its presence. So without, if controller, you just have this thing on the API and that's it and that's not gonna, get you much right. So what do controllers do?

A

Controllers literally sit in the cluster, they sit in the cluster and they just watch the API now how they watch the API. We usually talk about. We run a version of this coarsely over a couple days and you get into the hardcore specific sorta, like it former shared informers, to watch all that good stuff. That is you who use client go, know and love. I know you all know what I'm talking about but those years ago now just know it watches the API okay. It just watches the API for the current state of the cluster.

A

Then, after it watches the current state of the cluster, it analyzes okay, it compares the current state of the cluster like how many pods are running and how many config maps and whatever I want to look for, is the current state to the desired state. Where does the desired state live? You tell me yeah, that's exactly right. It's in the custom resource, in which section the spec section. That's your declaration right! That's what you want! That's what you want in the declaration, so it compares what is do I have any my sequel.

A

Things run how many pods, whatever do I mean deployments deployments so I have any you know Peavey's whatever to what I want in my spec and if I don't have what's desired. What does the controller? Do it acts right? It performs all the actions necessary to make a current state me the desired state, that's what it does and that loop runs whenever it receives an event and guess what for those you use kubernetes and you've never used it internally, which is totally fine. Maybe that's not your scope of your your.

A

You know what you do day to day, it's totally possible to use curators every day and not to get into the internals, just not part of the scope of what you do day to day. But this thing right here we talk about events, we're not talking about cube, CTL, get events, that's not what we're talking about we're, not talking about coops, to kill, good events, we're talking about events that you see you get when you watch the API.

A

You get a add event when it's something when that CR is created, you get a update event when the update is modified. I'm. Sorry, when the CR is modified, how do you modify a CR? You can run coops ETL, replace right on the command mock command line. How else how about some of you probably use coops ETL edit right, you do coup steal edit, and you go in there and you like using your editor of choice and editing it that way.

A

Maybe you do it that way, that's an update and then, if you delete it, so this thing only happens. This trigger only happens for those three different events, add update and delete okay. So it's not like your regular coop ETL, get events totally different type of event. This is an internal event here. Replica sets in action. Remember I told you that understanding what how awful the sets work can really help you in understanding operators.

A

Here's our replica set, which notice how it has a selector. It's saying, I'm, looking for a P equals my first app. We have a pod here, app equals my first app. That's how it knows to pay attention to this pod. It has the same label. We have replicas set to one. What happens here? We change replicas to three and it's smart enough to know to spin up two more okay, pretty straightforward, that's our primary resource! These are our secondary resources, there's just a nickname that we use to refer to something like this.

A

So imagine the replica set as sort of like the custom resource. Think of it like that kind of like the custom resource right- and these are the kind of the children or secondary resources that are being controlled by that thing. By that replica set now owner references are things that you may not know. You may not know what an Oda reference is right. You mean to have no idea what a reference is, but you use them all the time if you're spinning up deployments, you may name never just see. Maybe this is really notice.

A

Owner references help us with garbage collection right. Can you think about it? You're creating these things called deployments or you're, creating a custom resource, and it's creating things like you, don't want it when you delete the custom resource or you delete the deployment or whatever or you delete the replica set in this case, you don't want your pods hanging out right after the things been deleted, you want those to go away too. That's what owner references do okay.

A

So if we have a replica set, we spin up a replica set and we have pods here we want three replicas. They all have an owner reference now, what the owner reference does for those who don't know if the owner goes away. So in this case the owner here is this replica set. If the owner goes away and gets destroyed and never comes back well, I just say: if it goes away, guess what happens to the orphan here, it goes away right kind of sad right. So pod has a parent parent goes away gone!

A

Guess what child is gone, that's how they work, who controls that logic, the garbage collector that runs and is embedded into the controller manager in kubernetes. So this is embed it you get this for free in kubernetes, it's not an operator thing! Okay, it's not an operator thing very important when you're creating an operator understanding how owner references work, because you don't want to have people, you don't want to leave a mess right. You don't want to leave a mess when someone deletes your custom resource okay. So how does the replica set control loop works?

A

It gets triggered. This is Kylee. We showed you before. It finds pods matching the label selector. We compare the matched versus desired pod count right so think about this. If there's too few pods, when we're comparing the match versus desired right, we're gonna create some additional pods. We'll have some logic in there. If there's too many, what are we gonna do delete them right, we're gonna, delete them, get rid of them. We think and guess what this thing gets triggered anytime. There was an event: okay and then you can actually see these in action.

A

Here, it's a little getting a little more detailed. You have a cube API, you have a replica set controller. The replica set controller watches for replica sets, so it sees you bed space up, replica sets and you have pods. It also watches for pods, but not just any old pods. It watches for pods that are owned by the replica set and check this out, create a replica set, there's an event that gets triggered.

A

We then do a list against the API, for the actual state turns out that the actual state is zero, it's less than the number of desired replicas here. So what do we do? We then create one pod, guess what we get an event based off the creation of that one pod. We then list again get the actual state. Now we have one all right, you all seeing how it's playing on here right- and this is the triggering here- everything you see.

A

These events and I think it's important for you to understand that these events, your controller, doesn't actually know that they're at events. It just knows it there. You get an add event sure, but your controller isn't care if it's an ad of end of the care with an update event, it's something you have a bunch of if statements in there in your logic, right, it's not like that. We're gonna look at the logic coming up here and some you're very fast, because I don't have a lot of time. Oh now, we're happy!

A

Why are we happy because now three is equal desired is equal to what actual and that's good and then guess what, when that happens, your controller stops.

A

Every time you see an event, that is the controller. What we call it was a nickname for it. Reconciling that this is the control on the right is the controller on the right? Is the controller right so just take.

B

A

C

Said he's asking: if you want three pods, you might code it to loop three times or something along those lines or.

A

The 301 burst you.

C

Have three in one burst.

A

C

Oh three, all in one go is that it yeah, and sometimes you do want to do that, but sometimes it's sufficient and even desirable to just take one step in the direction of the desired state. Instead of going all the way to the desired state, it really depends on what exactly you're, deploying and and why, in the case of pods, probably it would be fine to go ahead and deploy all of them, like figure out the entire difference between desired and actual and just do it all in one shot.

C

But in other cases it's advantageous to not you're gonna come around for another reconcile loop immediately anyway, so there's really no harm and in some ways it's a better pattern to do it. Incrementally it's easier to test and that's reducing yeah.

A

And the performance is obviously doing it's very fast right when you do it this way as well. Okay, so, and look what happens when we delete a pod, we get a bleed event too right. So there goes our delete event and then we're gonna. Have you that actual state again, which says there's two so I obviously have to go and create a new pod. We get a pot event and then we're happy okay.

A

So going back to the my sequel example check it out. So remember we didn't create a my sequel controller, yet we were to show you how to replicate but notice how notice how the the the CR is sitting on the API here by itself. Nothing is happening in our cluster, but we don't have a controller. So what we have to do is that we actually have to go and very quiet. We have to go and deploy a controller in our cluster right and notice. How we have these controllers now.

A

This is actually I got to change this slide because, even though we want to have multiple controllers running guess, what only one is allowed to actually contact the API and act on that object. You don't want these controllers stepping on each other's toes here right, so we have in place leader election, so one controller will be able to actually watch the API here. So don't be confused by this I need to fix the slide and that's gonna dust or my seagull pod, but obviously I'm missing a lot of stuff here.

A

If I want to have multiple, my seagull pods, you probably want to have these deployed as a staple set. We probably want to have multiple TVs deployed. Didn't have tiny, but all those graphics on there, but here's reality. It's actually more complicated than that I mean you're, free to use any of the primitives or any of the artifacts you get with kubernetes to actually go and deploy this my sequel instance.

A

However, you want to do it right, you're free to do that, but it's really a lot more complicated than just creating something or reading it and updating or deleting okay think about anyone here. My sequel admin think about all the things that you have to do when you manage my sequel right, like all the tasks. These are the things that you need to think about when you build your operator. These are the things that you're gonna codify and put into your operator. Now how you put those things in there, you can use the go.

A

You can use, go to do it and if you're comfortable go and that's where you feel comfortable, you can do that, but think about this, you can use ansible to create an operator to with the operator, SDK and you're. Probably thinking to yourself, huh I can use ansible to do create a my sequel operator, Matthieu by you. When are you pulling down? There's my sequel, ansible modules right, maybe that you, those either use ansible right. So we'll talk more about that we get to date.

A

The Angelo piece as well, but point is, is that this is the human operational knowledge we're talking about how you codify that and how you put this in to your operator and again, we'll think about more of an app think about an app that you know and love. This is kind of general, my sequel, the ins and outs of that app. That's what you're gonna put into your operator, how you do that we're gonna talk about coming up. So to recap: we custom resource definitions, plus custom controller, plus your knowledge equals operators. Why they matter.

A

Here's the whole idea we're trying to build an as a service experience right. What we didn't buy that you go to AWS you go GCP, you go to a sure. You know you want to go, you know, go ahead and provision a database or something is very easy. You just you know, fill out some fields put some radio buttons. If you use the UI, you know you're, probably using terraform or something you know to actually programmatically interact with that infrastructure. But you get what I'm saying it's like a very easy experience.

A

You build an ecosystem of software on kubernetes that can be as easy and safe too reliable to use as a cloud service touch remotely managed, one-click updates right, giving you that really nice, you know low touch experience an example of an operate on the dashboard looks something like this. When I say dashboards, I think in what dashboard? What do you mean? I just mean from the UI.

A

If you were gonna, deploy an operator like this is what we're going for like making it look like just clicking a button and going enable it here's the sed operator, that's where your apps running on kubernetes Prometheus and for monitoring and volt for a secret storage. It's really easy to play an operator, accumulators environment. How do you create your own operator light before the operator SDK?

A

It's really for those you create an operator, you know, can be a little complicated right, it's not just as simple as creating some resources and then defining a controller with some pseudo look code. That looks like this right. It's actually a lot more complicated.

A

They require many building blocks of boilerplate code. You have to research and download tools to interact with the API. First of all, you start at the client libraries for those you who know there's not a lot, there's not a whole lot of Doc's on these client libraries right, maybe some of you who or you know- maybe you may not be you doesn't matter. You just read the code anyway, but you know: go Python Java dotnet. What is this stuff for it's for programmatically, working in the language that you're most comfortable in to interact with API.

A

We interact with Guber Nettie's right, specifically the communities API yep, okay, well, I didn't realize that knowledge of Informer's shared Informer's or object, cache and event handling. Why do we care about object, cache and event handling? Think about this? What do controllers do? They watch the API? You have a whole bunch of controllers running in the cluster right tons of controllers running in the cluster they're, always wedding in the cluster there watching the API. What does that mean for the API that poor API is getting beaten right? So how do we make it more efficient?

A

The controller has you know. Caching, we have a work queue where they work off of communicating desired state. You have to build that. If you do this from scratch, it's really annoying. This is an example, we're just using client code directly, just showing you what a sample project would look like, and we need an easier way to create operators and easy way to manage operators. That's what the operator framework comes in. It consists of operator. Sdk allows you to create an operator with go and or helm.

A

You can convert all your existing helm charts to an operator. If you want to okay, we're not competing with helm, helm is awesome, but if, for some reason you want to create an operator and put it in the operator hub, you can do that with or put it in your operator hub, which is nothing more than like an app store. You can create your own if you want kind of your own registry of operators, helm, helps you the helm, helps you do that with the or the helm side of the operator SDK.

A

This is the full lifecycle going through five stages of maturity and every we're up to autopilot, which is fully running on its own, auto scaling, responding to custom metrics for those you who are devs you're, developing operators, Prometheus metrics, definitely learn how to instrument your app that's really important using the previous libraries, because autopilot responds based off thresholds of those custom, metrics. Okay,.

A

That's exactly right, you can choose either one you can choose helm, you can choose Anne's floor, you can choose, go.

A

Yeah, that's exactly right! That's what that's what it's saying here! Maybe someone if you all work on hell more, if anyone here works in the home community, we can we can talk about this, but that's what this is suggesting. Yeah and then OLM OLM handles the installation. The upgrade did the whole dependency management with operators. How do you install them? How do you upgrade them all that good stuff?

A

This is actually giving you that app store, like experience, think about this when it comes time as an operator author for you to roll out a new version of your operator to your audience kind of like an app store, oh I, have to you know one of you: do you want version 2? What is it gonna do? That's what LM does for you. We might have time to talk about that later. It's totally upstream fully available operator hub that IO. This is these run in any kubernetes clusters.

A

It's kind of like your App Store, like experience, caddy coda, housecat ACOTA doing. Is it working? Oh well, good? We.

C

Apparently had so many connections so quickly to Kai Dakota that we set off an alarm and it thought it was under attack and they were all from the same IP. So it said: oh, no, no, we're not let these people in so the threshold has been increased. We should be in good shape, anybody's still having trouble well slow, ok, hopefully, it'll still be workable and I think we're about to dig into that now. Yeah.

A

C

A

Thanks, oh yeah, by the way, there's some bonus materials I forgot to check the website. It should still be up workshop that quarter less trained me a lot of slides. You can download from there workshop that core OS trained me trained me. It's like there's lots of good stuff on there. You can download lots of resources, lots of access to the support Docs and the chat and all that good stuff. Please take a picture real, quick, I'm.

A

Sorry for speeding through all this stuff, I told Michael, I promised Michael I said: I won't make this one long and it ended up long. The intro long I have a hard time condensing this is a big topic. It's a big topic! Ok, ok! So, by the way get your! If you want, if you all working on an operator, you want to get it certified and featured in openshift, so everyone sees it and even if you, you know, have a an enterprise offering or something that that you know so after people are evaluating.

A

The upstream version come talk to the people in the front. Here: that's what they do. They help people get their operators certified going through a series of very strict testing on your image and your operator metadata. Okay, so we're gonna go ahead and create an operator with go, and it's going to be very quick and I'm gonna ask I'm not trying to be rude, but it looks like let's hold the questions. I hate rushing through this type of stuff I. You know it's just, but this is where we're condensed for time, but stick around afterwards.

A

We're gonna answer all your questions again. We're gonna be here for you to help you. So what we're going is we are going to not the Cates fundamentals, which is a really cool exercise that helps you explore the API, but the operator SDK with go pod, set the operator SDK with go pod set and what this is now. Look. Here's a reality. I noticed with these workshops.

A

A lot of people like just this is the quiet time where it's like go work on this, but I'm just gonna walk through it; okay, maybe it helps okay, so I'm, just gonna walk through it very quickly. You're free to do this. On your own time, oh yeah, when you leave, you can hit all these exercises, not add flash training, but just knock off the slash training right and just go to learn. Dot, open chip, calm and you'll have all the modules on there. All the modules will be on there.

A

So when you leave learned open chip, calm we'll have all this stuff. So it's not going away so I guess: I have to wait for my environment to come up now. I! Guess: okay, let's do this! What we're gonna do here is we are going to use the operator SDK command, okay and we're gonna say here operator, SDK, new pod set operator type equals go operator, SDK new and now the gentleman in the back set was asking: do you have to use all these things? How does it so?

A

No here we're saying I want to create an operator with go now you may be wondering, should I go with go or should I go advancable which one do I use here? It really depends. It depends on what you're comfortable with that's all, if you're an ansible shop and you all- you have tons of antal modules that you all have stored and tower, or you use you like to fetch them from galaxy or, however, you want to you know you use ansible. You can use ansible to create an operator.

A

If you all are go shop, you love, go you code and go you can use, go we're gonna create a go on here. But let's see, if this one provision in time can anyone I don't know. Maybe if we want to maybe your house, yours yeah, this I thought we did this in the beginning. I thought that okay, yeah, okay, oh here, we go okay, I'm, good, yeah, okay, alright, good yeah, okay, alright, so I'm gonna create a new project which, for those you who are like what's project, are you talking about it's a namespace?

A

That's it open shape disease use the word project. That's all it is so I'm gonna click this to create a new project and then I'm gonna create inside my little sandbox and my go path right. This Red Hat directory and then I'm gonna, navigate to that directory and.

A

Then I'm going to create the operator with go and I'm gonna say skip get in it, because guess what we're not pushing this any we're not pushing this to get so we don't going to worry about that and there it goes now what's happening right now is you're getting all the boilerplate code and you can probably those either use go. You know what that screen looks like right. Third, your the depth screen.

A

That's going out and pulling down all those packages, okay that are being referenced in that boilerplate code, so getting all of that all that go code from those import statements that are in the top of the code. This is talking out to the net to do this, so you do need a extra. You know you need external connectivity to run this, and so what's that you're probably thinking is this creating my operator? How does it even know what kind of operator I want to create? What's happening right now, I'm confused! This gentleman has a question.

A

Yes, thank you.

B

A

What's the error.

A

Was that okay, oh.

B

B

A

Oh yeah, by the way, real, quick yeah. Let me show you that somebody asked about this. Where do, though I know we just want to show you it workshop? What is it core? Os train I have a minute. Anyone for us trains I mean okay. If you want the slides click on support' and you can download either of these links. This was a from previous cube, Khan and ansible fest.

A

Yes, we did not update much content here from the previous cube con, but we are totally revamping this for the new version of OpenShift of the URL workshop core OS train dot me I'm, gonna put it on a big screen here.

A

A

Thanks, thank you.

A

Yeah so he's talking about the difference between COO builder, which is another way to easily create operators now with, if you're, all familiar with COO builder, so they both both operator SDK and both COO builder, both use something called controller, runtime right controller runtime is a series of packages, go packages that makes creating operators really really easy. It was developed by the kubernetes authors actually or developers of the API and there's some news that actually operator SDK has been working with COO builder directly, so are we gonna?

A

Do we know anything about the changes with the operator, SDK Michael and up and cue builder like as far as what's working together still pending, but we have weekly meetings with each other, so we are not we're, not that's they're. These aren't two competing platforms, so.

C

To speak right, ultimately, the go features are gonna fold under queue builder and operator. Ck will bring operator specific things on top of what key builder currently offers that'll, be things like packaging for operator, lifecycle manager, perhaps testing, and that sort of operator specific stuff that opera CK does today, but that'll all be go focused and probably the ansible in home pieces will stay under the the operator CK umbrella, as it is now yeah.

A

So there's you're not this is really a good. You know which one should I go with. They're, basically gonna be this have the same workflow operator. Sdk is great. If you want to be in the operator, lifecycle manager and get those kind of features right, building out all the artifacts, you need for OLM.

A

Yeah we don't know yet, but Oh Willa doesn't say this, let's add, is a good question if I want to get started, creating an operator today. What do I pick, though, right now tonight, I want to get started with the operator and work throughout the weekend, which one am I gonna pick. What should I do? I.

C

Would say I mean today like right now tonight operator. Ck still has the operator bits that are gonna, get you all the way to the finish line with operator hub, for example, or does generally. If you want to package and distribute your operator even internally, it's got that essentially, that package management feature built into it and ready to go so I would do that today. Oh thanks.

A

Yeah, okay, so.

A

Okay, I'm gonna go to this pretty quick, so the first step to creating this go based operator is to add an API is to add an API. What does that mean to add an API that means to create the custom resource definition, the custom we basically telling this this code? This is the custom resource definition that I want to use, make the code aware of that actual object. That's what we're doing here.

A

Okay, so we're saying operator, SDK, add API, API version, apt, example.com, /v, 1, alpha 1, so very creative name group here apt at example.com and then we're gonna be calling creating this thing called a pod set.

A

Now, if you're wondering what is a pod set a pod set in this objective, the kind of operator that we're actually creating does this right here, it's very simple: it's almost like a replica set and that, like you, specify the number of pods that you want right and then it gives you that number of pods- and it also has the pod names in the status section- very, very simple, maybe thinking well, how is this different than a replica set?

A

It's really not much different at all, just we're showing you how this is a good place to start to understand, creating a controller okay here if I change the size to one the declaration, the pod set should be smart enough to scale down, so you want to have in the spec section replicas in the status section pod names, okay, so just to give an idea what we're doing here, all right, so we're gonna go and create this add API command.

A

Again, you see what it's creating here. You can see it's creating the CR, the CR D for us automatically. It doesn't know the contents of our custom resource, it doesn't know what we want to have inside the spec. Yet we have to tell it that right and never we want in this case we want, was at replicas. That's what we one side of respect right. We haven't told it that yet we're gonna have to tell it that okay.

A

So now, if you actually look at the CR D, the CR D has been created for you, alright again I'm skipping over. So much usually you know this is a lot of stuff and we're skipping it over now. We're going into here is hopefully, if this works for you, you can drill down on the very top to this file. Location, pod, set operator package, api's.

A

And I need to refresh.

C

It seems, like the environments, are working again for many people, so if you're still staring at that red error message refresh one more time, hope it'll work.

A

So what I'm doing here is.

A

There we go okay, all right, so what I'm doing here is I am going to create my cuss. My spec I have to design my spec right. What do I wet in my spec I want the number of replicas, so what I'm gonna do here is I'm gonna drill down and do this types go file right right here, posit type let go this is where I design my spec. This is where I say these are the fields I want right, and this is what I want them set to. This is the status that I want.

A

Please open. Okay, there we go all right. So what we're gonna do here is we're. Gonna say this is very easy: okay, Dakota, all you have to do is click copy to editor, and it literally updates it just like that. But if you're wondering what this is spec section status section, this is a field that can be referenced and go. It's gonna be a 32-bit integer and then we're. This is a JSON tag. It's basically saying inside the JSON.

A

In that declaration, we're gonna have replicas here and in the status section or have this array of pod names. That's all we're doing, and after you run that command after you update that code, you run this generate Kate's command for those of you that are super controller savvy. These are just the code, generators that are running in the background here. You know you don't know what that is.

A

You never use controller, don't worry about it, but just understand that just a little command that we did to run those code, generators to create the type objects based off this information here and OH generate open API. You all. Have you know it's a really cool thing about this is? Is you all when you all do coop CTO edit?

A

Like an object, you know when you go in there and you try to change something and you hit return I'll satisfy that. You made your change and it like comes back at you. It's like you can't do that. That's validation, that's validation, taking place right. You can do that in your spec, so you can say users can't put four replicas. You know some like long string or something they can only put it 32-bit integer e, you can say they can only put an integer between 1 & 5.

A

You can put all that information into your validation and that validation goes into the CR D, and now we have our C Rd created and there it is pod sets tapped at example.com. Now comes the really cool part where we have the logic controller. The controller is the thing that does the reconciling remember that screen that we showed you with the back-and-forth the the kind of the looping. This is the controller that does that, reconciling and then now what we're gonna do is we're gonna, go and update our logic.

A

Let's see, let's go into our controller pod set.

A

Pod set controller. Okay, there we go, you get some default logic, but it's very, very basic. It just creates one pod, we're gonna create them. This is what our logic looks. Like okay look, tipless and I promised we would typically go through all this and tell you everything that's going on, but we don't have time. We got to talk about Hanceville operators, so please come I. Want you to scan this I, want you to get your questions ready and come see us afterwards and ask his questions about this.

A

Okay, but the most important logic, if you're kind of wondering what's really going on here, the real meat of it. If you will like the actual thing that we should focus on, is if the number available is greater than the this is the comparison right. The number available is the actual state. This is our declaration. If this is greater than this, we're gonna go ahead and scale down. If this is less than the Declaration we're gonna scale up. That's the most important piece in all this right, so I'm gonna go ahead and apply this.

A

And we should be updated there, you don't have to click, save or anything. It should save automatically, and now we're gonna run this controller using the up local command. You may be thinking, I, don't understand. Why aren't you deploying this with a deployment like into the cluster? It's because the up local command allows you to compile this code and talk to Anne Cooper IDs cluster, if it's externally accessible, so you're literally running a controller outside of the cluster, and it can actually think about this.

A

The reason why we have this command and we made this command is because can you imagine if every time you made a little code change you'd have to literally compile your image push it up to a container registry go, and you know put that the name into a deployment. Imagine how annoying that workflow would be. So. This is a really nice workflow to just instantly connect to your cluster and try your code against that cluster. Ok and it's running.

A

Ok, how do I know it's running well, I have a my screen here and let's go ahead and make sure it works. What I'm gonna do here to make sure it works, is I'm gonna, open up a new terminal and I'm going to go into this directory here and I'm gonna notice that this is set to size. Remember we set it to replicas. So how do we get this to actually work but step the replicas? What we're gonna do is we're just gonna open up our deploy.

A

Folder go to see our DS go to see our y amyl and click our magical copy to editor button. Isn't this great? Wouldn't it be awesome if Cody was actually like those you just like clean fun? No that wouldn't be fun all right. It would be very fun all right, so then we then we actually go and we say OC create F, deploy, CR DS, let's go ahead and create that and then we say OC get pods and the control or check this out. This is pretty cool.

A

If you look at the terminal notice how instantly when I created that custom resource my controller got an event, you can look at all the logs here and then it scaled up based off that event, etc, etc. You might be thinking I, don't ever say this is such a simple example. While you do it showing us, it's like a replica set was a big deal.

A

It's academic or just you study this, and this will prime you for understanding how op the controller or loop works play with this play with this play around with it, modify it break it get comfortable with how this controller is actually working. We can actually see that if we actually go in and we say, OC get pod set. This is pretty cool.

A

Just like this OC get pod set.

A

Oh yeah Mille. You see that how the pod names got put in there. Okay, pretty cool you look at the code. You'll see the relationship there again we go through more, but we have to move on to the anvil operators right. Let's go ahead and do that yeah Michael. Do you want to talk about that.

A

Yeah, let's yeah real, quick guards. The gentleman asked about garbage collection. Where is it? How does it work? Where can I see it? You can see it in the code, but did you want to see it in person? You just say: OC get pods. Let's look at a pod and see if it has a garbage collection owner reference on it, I'm gonna grab it grab the first one here, I'm gonna say: oh yeah mo let's grep for the word owner ref and let's just go down.

A

See it it's in the metadata section there, it is so these pods have owner references on them right and what does that mean? It means that if I actually go and say, OC delete, pod set example: pod set, guess what permeating okay.

C

All right how's this looking fantastic we're plugged in we're ready to rock. Let's look at the easy way to do this with ansible now, I'm gonna apologize up front. I've got a bit of a cough I'm gonna, try my best and not coffin to this microphone, but we'll see how it goes. So why would you do this with ansible?

C

So the answer both side of the operator CK makes it particularly easy orders of magnitude easier to deploy, manage applications with operators as opposed to doing one from scratch, with even with the go side of the operator SDK and let's look at how it fits naturally. So a criminai is an ansible, very much playing in the same kind of mindset.

C

There are both all about automation and orchestration, open-source very popular active communities on both sides, but two things that I like to focus on are they're both yamo base from their heavily gamal top to bottom, so you're already familiar that they're, both also all about declaring your desired state. So in the antipas side, if you're not familiar with ansible, you create an answer, will roll such that you can run it over and over and over again and no matter what the starting state of your infrastructure was. It will end up at your desired state.

C

Very much like a controller in kubernetes already does anyway. So these are a really natural and a really good fit for each other. So let's look at using ansible as a client to interact with kubernetes. So on the Left we have normal yamo, like you might see in a file, you could cube CTL create F. This config map and you'd have a config map on the right side.

C

This is actual ansible code and you can see I've just inlined I've, directly inline to the exact same yamo, and this will using the k8s module directly, create that resource or it'll ensure that it exists in the desired State. So it's a little bit like, like the apply verb that you're probably used to now.

C

The one difference here you may see is I've taken the liberty of just showing off a little bit in template izing that last line, so we've templatized color and when your ansible code runs this way inside your ansible operator, any any key value pairs you put in your spec on your CR D, get plumbed in here at render time for ansible to use. So if you have color : read in your spec of your CR that you just created in your cluster, then when ansible runs its gonna use.

C

That and its gonna render this template using those values, and it will ensure that this resource has the value red as the string red there as the color. So pretty neat. Now what? If you don't want to inline your yeah Mille I? Don't really want to inline my mo either there's a better way you can maintain.

C

Your llamo manifests as separate templates and templatized them just the same way, but keep them in a separate directory inside your role and with these three lines we're looking we're using the lookup feature of ansible, which is a staple core feature of ansible. It's used in many different ways: we're looking up the template that has a certain name and we're just rendering that and the k8s module understands how to receive that as its input, and now it's just going to basically apply whatever has been rendered out of out of those templates.

C

So this is a very natural way to interact with kubernetes from inside of ansible. So why build operators are the antelope cutting right to the chase of these things? Ultimately, it's it's about the ease and the ease of getting started. The ease of maintaining you can have an operator with no previous experience at all, even if no experience with ansible even itself up and running in under a day when your boss comes to you and says, hey, we bought openshift.

C

Congratulations and now you have to go make operators for these twelve things, and I need them by the end of the week. So we can continue to deploy them.

C

You can get that done actually in a big hurry and you can take even whatever yan will manifest you've already been using to deploy your workloads meet with other tooling, maybe not and feed them in here and create an operator very quickly, but you still have the full power of ansible, so you can do anything fully advanced day to management of your workloads is still totally within realm.

C

So I'll say about that. For now we looked at this already earlier, so you're familiar now. How does a controller work Matt already talked through this we're gonna compare. This is a normal controller and we're gonna compare this in a moment with an answerable based controller. So in this normal one, we've got this controller in the middle. It's watching events right any time a custom resource or any a resource of any kind changes that it is watching. It gets that event and it runs that reconcile function right now.

C

What happens when we drop ansible into this? Here's what we have inside that controller we have created for you, a generic binary, it it's written and go using the operators 2k and it runs and it'll watch whatever resources you tell it to watch. So we have this watch file that we'll look at in a moment. That really just says for this group version and kind. When you see an event for that group version kind run this ansible playbook or run this ansible role. You do either one. So it really ends up being very simple.

C

There's some other advanced features around this, but then that ansible can run and do anything it wants. Now. Matt was talking about owner references earlier now here this automatically puts owner references on every resource you create for you. So it knows what your primary resource is and when you create secondary resources, it will do two things. It'll put the owner reference on it, and it will also start watching that other resource type also and it'll use that that handle or if you're, familiar with the go side. It will include.

C

Of happens very naturally for you here, finalizar czar, super easy and so on so automating with ansible doing this for the first time. What are you gonna? Do you are going to create a role or a playbook that understands how to deploy an application? So if we were gonna deploy Maria DB, we would create a Maria DB role and fill it out so that it understands how to deploy Maria DB on kubernetes.

C

In this context, in the operator context, the custom resource input is going to be passed into ansible at runtime. So when we run answerable all that input gets passed in and is used by ansible to render templates for a logic branching, you can do effectively like, if kind of statements in an table, you can run a block of code. Only when an expression is true. So all those things you can use that input to in very powerful ways make decisions in your ansible code and so typically you're gonna use the k8s module.

C

There's a collection of other related kubernetes modules around that to do things like manage the status and other things and then, if your interact with any external services, you can use that as well. We've seen somebody creating an operator with ansible to manage a physical switch, that's on the top of their their rack of hosts that turn into nodes in their cluster, so they're managing their kubernetes infrastructure from kubernetes, including the switch and they're using Angelo's, very rich collection of network equipment roles to to do that. So, let's, uh let's keep moving.

C

This is what that watches file looks like, like I, said, really. No, nothing fancy here, just group version kind mapping to a playbook, and this gets scaffolded out for you by the operator 2k as you'll, see now, with your custom resource in the spec. So let's say we're making this Maria TV operator. We've got a Maria DB custom resource definition when I create one of those resources in the spec. I can put any key value pairs, I want and those get passed into ansible. Now what about the stat? You have two options.

C

The easy option is we take care of it for you, the answerable based operator will create a default status. It's fairly generic. It's not going to be specific to your workload, but it is a useful and informative status. It says things about what happened when ansible ran didn't encounter any errors, did it change anything or not those sorts of things, and then probably the last thing we look at before we dive into the exercise. This is what the image really looks like for your ansible based operator.

C

All the stuff on the bottom is provided for you that binary answerable itself is in the image all the libraries and dependencies you need. All you add on top, are that watches file and whatever roles are playbooks get referenced by that watches file. It's really very simple in terms of artifacts, it's very little for you to manage. So you can already see this is a very different picture than managing an entire go based software project yourself, I'm, a software engineer, I, love, writing and go I would be happy to write good all day long.

C

But not every organization wants to throw a full development team onto every operator that they need to make. This is a very different way to manage the artifacts that you need.

C

Organizations generally are very used to having small teams, ops teams teams are not even software engineering, oriented owning and maintaining ansible assets as a regular matter, of course, and that's just as easy to do here with your operator.

C

Okay, let's go back to our exercises and, let's see mine is alive and well now. I've got good news for you about this exercise. It's a nine step exercise. The first four are start is just reading and I already told you all that stuff, so we're just gonna fast right forward past that all the way to step five and about halfway down it sounds like they're already tearing down the conference over there: okay, that's not an earthquake I'm in East Coaster, so I'm not used to that fear.

C

Okay operator SDK new, just as you might expect right, so we're gonna, run operators 2k new we're, saying type is ansible, we're giving it the group version and kind right up front and it just scaffold it out an entire operator for us. Let's look at what we got with this tree command right. What do we have here? Okay, so we have a build directory and we have a docker file. We could build this operator right now. We have a working operator already.

C

The bad news is, it doesn't do anything, but it will totally watch our resource. It will run in the an total role that it just created every time that resource has an event there's nothing in that role yet, but we already have everything we need. We could run operators to get build right now and this thing would run deployment. All this is all the usual stuff you'd get with the SDK, where there's go or ansible or anything else. So we have the CR D. We have an example CR and so on all the stuff.

C

You need to deploy your operator into a cluster molecule. Anybody familiar with an Sable's molecule project, yeah big fans in the antipope community of molecule. It's a whole testing framework.

C

It's super useful, as it turns out for operator testing, not even just ansible operator testing, by the way, if you're, even if you're, making operators with something that's not ansible like an NGO or otherwise do you think molecule it turns out is a really useful way to to stand up transient cluster very quickly, throw your operator into it, test it and tear the whole thing down and give you very quick feedback cycles as you're developing or your operator. So that's really cool stuff. I wish we had more time to go into that right now.

C

Okay, rolls ansible rolls, so we got a memcache D roll scaffolded it out for us already in in all this stuff. There's really only two things you need to worry about inside here. One is tasks main yeah Mille, that's where we go. That's like the main function in your program. That's the starting point for our ansible code. So that's what we're gonna we're gonna start putting ansible code and then the templates directory that's the directory. We can put template files so any mo manifest you already have.

C

You can just stuff them in there and reference them from that main dot. Yeah mo like we saw a few minutes ago on the slides and then last but not least, we have this watches diamo file, as promised. That's the file where we we look and express for the group version kind run this role, because the operator ck already knows what group version kind, you're, gonna use it already created the role for you. It has already filled out the values in that file, so we don't even need to touch it.

C

It's already ready to go so, let's press on. So our next step is to customize the operator logic cuz right now. There is no logic now, rather than sit here and type out some ansible code, we're gonna, take the easy path. We're gonna take the lazy path. Thankfully one of my colleagues, Dylan Murray created a role for us. It's to gonna, deploy memcache D. So what we're gonna do here is use the antibiotic to to download and install that module for us an answerable galaxy that I'm the command. There's really nothing fancy going on here.

C

It's just retrieving that off of github for us and putting that role here in our roles directory. So if we type one-handed and look okay, so in the roles directory look, we've got a way to die. I missed a step. Didn't I I missed a CD into my memcache D operator directory. Don't make that mistake like I did and I'm glad I caught it just then. So, let's run this again.

C

In let's LS in roles that looks better okay, now we have that old role that original one that got scaffolded out for us. We're just gonna delete that we don't need that anymore and let's take a closer look at that. If we look into that role, it's very similar, Dillons customizes a little bit. But again, let's see it's got that tasks file. He doesn't actually have templates he's not using templates here he inlined all the ammo.

C

You can look at it yourself if you like, but we just have that main Don llamo file, and that's really the only thing you need to worry about to get an awful lot of stuff done.

C

We can go and look at the defaults. Let's see this is not sometimes this text editor lags ah here. We go memcache D operator roles, villain memory, defaults may enamel. So in the defaults we can set defaults for different things. This case we've got a default of size, 1, we'll just stick with that, and you can see here. This is a printout of of what Dylan did and you can see that we're using just Jinja style templating here and where he's referencing so meta name.

C

That is the name of the custom resource that you just created. So if I had a memcache de resource in a Yemma file and I, keep CTO create a shelf that thing this operator is gonna. Wake up its gonna see that event. It's gonna run this code and it's going to use the name of that that CR to be to create the name of this deployment Andy, you did the same thing in the namespace, so that's pretty cool and finally, the size we're using the size.

C

Also in this template again, as Matt was saying, this is not a brilliant operator. In fact, this is kind of a silly operator. Why would you do this?

C

We're just reimplemented sort of what a deployment already can do, but this is the pattern it's academic and you can see how you could add logic onto this to do more advanced things like upgrade logic, fixing things when they're broken any kind of custom logic that has to happen when you scale up and down some things like Etsy D, you can't just start up new that CD processes and expect that to work, there's a whole workflow of introducing new nodes into a net CD cluster, for example.

C

Many other databases have the same kind of thing, so you can imagine how you would add that extra kind of logic to your ansible role.

C

Okay, we do need to correct our watches file, because that role is in a different file path. Now we just need to change this. The path on disk to that role and by the way, what this is. That's the path on disk inside to the container image for your operator. When you build your operators, image you've got a docker file for your operator. You're gonna put your role somewhere, and it's got to be wherever this thinks it is. So you just gotta make that match.

C

Alright, let's press forward it's time to deploy this thing, all right, OC did we cover what OC is great. Oc is just like cube, CTO.

C

Okay, we created our CRD great, so we've extended the kubernetes api. What a miracle that is by the way, how many systems do you know that? Let you just extend their api with your own things, they're really empowered and it enables so much it's a great feature. Okay, we are building our operator we're in the extracting phase. Okay, we have built the operator.

C

Now we needed to just hand wave a little bit we're gonna tell you have to choose a name for your your operator image right, your container image, the sdk doesn't know it can't read your mind and know: is this gonna be on docker hub? Is it gonna be on quay? Is it gonna be in your own registry?

C

Is gonna be somewhere else, so you have to pick when you build when you run that build command where you're gonna host, like what's the full reference to your image, gonna be so we just need to put that here in our deploy manifest operator. That's all we're doing here and we'll change the the poll policy just to help with the demo all right new project.

C

We know what new project does we did that in in the last exercise that creates basically namespace for us with some other features around it service account role, role, binding all very critical things for your operator by the way notice that operator seek a scaffolded, those things out for you and they're already ready to go with with defaults that get you started.

C

We don't have to muck around and there yet with with the our back. Ok, it's time to actually deploy the darn thing and cross our fingers and hope that our operator starts successfully and it looks like it did. Ok, congratulations. We just made an operator.

C

Thank you. Thank you very much. Ok, rising forward, let's actually use the darn thing and and just see it in action. So we have a custom resource here. It's the kind is memcache D like we were looking at earlier. Just this is the ansible implementation of that we can go. Look at that file, this one.

C

Yes, it is ok! So by default we have a size of 3 here, and this is our memcache D resource, let's, uh let's just change it to two cuz. Why not.

C

And let's create that CR and see what happens? Ok now, hopefully, just now we created that CR. Our ansible operator is running so that go biner that we created for you. It was already watching memcache de resources. It saw the event. It noticed the group version kind of the resource that the event was associated with it.

C

Looked in that watches, file that mapping file and said, okay, I know what role to run and it ran that role and it injected the entire CR all the concepts of the CR as input when it ran ansible with that role, and when, in that role and as we saw in that role, it should have deployed or should have created a deployment, and it did. We have our example: memcache D, memcache D. We have to find a better name for that deployment running. There are desired two pods there are currently two pods they're available.

C

Life is beautiful. This thing works, I love it. When a plan comes together.

C

Okay, we can come here, we can change the size, I, don't say five.

C

Week, a notice, the apply.

C

We can give it a moment there. We are: we've got five, only two are available, but there's five pods, okay, now they're all available. So it's that easy. We've taken ansible we've taken very simple: ansible code: we've made it kubernetes native as they say, and a big part of being kubernetes native is plugging into the crew Bernays api. So there is now a custom API in this cluster, representing memcache D and plugged into that we're plumbing all the way from that API we're calling an ansible and we've exposed this Angela code through that. Yes, sir,.

C

Oh, we we could I'm I'm trying to type one-handed, so I'm not gonna, but yes, we could and I would encourage you to run this. This exercising you get home and give that a try, but it absolutely would if any outside actor comes and changes. This deployment that's going to generate an event and that will it'll notice that event I mean that's. That is literally actually what we just did: I guess on the CR, but just like it's watching this er.

C

It's also watching the deployment that would kick off an event and fix whatever somebody had head screwed around with.

C

Okay, look: we've got the memcache D resource, let's just look at the status real, quick, so, like I said it's a generic status, but it shows that the last time this ran nothing changed. Two things were okay, these are ansible tasks and by the way, ansible code typically is a series of tasks in traditional ansible. It's often, you know a role to like deploy a service on like a physical blending.

C

So you might start by. You know ensure these rpms or Debian packages are present, ensure that these ports are open on the firewall, ensure that some config file exists in the right place, create a system D unit file enable that system to unify all those kind of things. Each one of those is a task going down your gamma file, and here we're doing the exact same with an tobacco. We had two of those tasks, apparently in here, and they both ran. Neither them changed anything. Neither them failed. So, overall, pretty boring, run there of Anton.

C

Now what? If all right, I'm gonna put the mic down for 10 seconds, be right back.

C

Is it on there are oh great, okay, thank you! Thanks, okay, as you get deployments, we're back I think it already fixed it. So I scaled it done too. We can look at the pods and probably see this summer are much newer. Yeah see these are only three of them are only 15 seconds old. So there you go scale down and the thing went right back up and did it yeah.

C

So if your intern gets in there and starts messing the stuff they shouldn't mess with your Angela based operator has your back and just like that. We can also delete this thing, get the pods, so we deleted the CR and it's uh did I loops I deleted the ability the operator instead, oh well, I.

C

Think we've made the point. The next part of this exercise. We're not going to go through right now, but you can go through on your own is a way to run your operator locally when you're, developing and iterating on your operator, whether it's go or anta below it slows you down.

C

If you've got to build a new image and push the image of registry and then wait for your cluster to pull that image and deploy that in a in a pod, so the operators ticket is a run local, commit or up local command that you can use to run the operator like locally on your laptop, and you can get very, very quick feedback cycles that way, and so the rest of this exercise just goes through that now, when you get home this weekend, if you're just so excited about the this whole week, you just gotta make yourself an operator.

C

You can do that just go to learn. Openshift comm without the training part on the URL and all these exercises are there. It's not there's not enough capacity there for 400 of your best friends to do it with you. So that's not why we didn't use it today, but you can individually do that anytime. You want there's no signup, there's no information collected. It's just totally free and open anybody can do it.

C

Thank you. Yep lots of other operator exercises on there so go. Learn. We're gonna stick around I think they might kick us out of here. So we might just move right out there, but come ask us all your questions. We're here you thank you for coming.