Cloud Native Computing Foundation Kubernetes Day India 2019, 29 Mar 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Kubernetes for Java Developers - Arun Gupta, Amazon Web Services

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

Kubernetes for Java Developers - Arun Gupta, Amazon Web Services

In this session, Arun Gupta will discuss how to package a multicontainer Java application using Kubernetes. You’ll learn the basic concepts underpinning Kubernetes before diving into packaging a multicontainer application using Kubernetes, scaling your application, deploying your Java application to Kubernetes using Maven, and then monitoring it. Along the way, Arun will explore service discovery, load balancing, persistent volumes, and health check concepts with live examples.

About Arun Gupta
Arun Gupta is a Principal Technologist at Amazon Web Services. He is responsible for the Cloud Native Computing Foundation (CNCF) strategy within AWS, and participates at CNCF Board and technical meetings actively. He particularly enjoys AMA on Containers and Serverless. He has built and led developer communities for several years at Sun, Oracle, Red Hat and Couchbase. He has extensive speaking experience in 40+ countries on myriad topics. Gupta also founded the Devoxx4Kids chapter in the US and continues to promote technology education among children. A prolific blogger, author of several books, an avid runner, a globe trotter, a Docker Captain, a Java Champion, a JUG leader, he is easily accessible at @arungupta on twitter.

A

All right, hello, everybody, my name is Arun Gupta I work for Amazon I work in the open source team, so I work with a lot of service teens across Amazon will help define their open-source strategy and I. Understand today is the kickoff of IPL, so go RCP yeah.

A

So no slides here by the way I totally believe code is king and always should not be presenting slides as developer conferences, which is we just jump into the code and talk about code and live in brief code? Okay. So this is purely a get approval presentation here: I'm a Java developer I was part of the original JDK team back at Sun Microsystems when we were doing JDK 1.1. So there's a lot of Java legacy heritage, Java Java champion as well.

A

I talked about Java quite actively, so as a Java developer know how many of you our developer, cheering day. Almost 90% of you are Java developers, that's a good sign, so Java is no matter. What you say is the most prominent enterprise language here and not just here but around the world as a matter of fact, but as a Java developer, and we saw how kubernetes is leading the mindshare in terms of orchestration.

A

If you are jumping on your journey to become a kubernetes developer, what it should look like, what are the tools I need to be familiar with? How would I get into that journey? So this is sort of my journey. You know our day in Java developers live with a taste of kubernetes okay.

A

So, let's think about what those topics are: I only have half an hour, typically I think about two to three hours to go through this in detail, but hopefully this will raise your curiosity enough so that you understand what it takes for a Java developer to get started with it. The key part is bookmark this repo here, okay, so this is a repo on AWS samples is kubernetes for Java developers is also available, as a recorded course on LinkedIn. So if you go to linkedin.com search for cooperate is for Java developer.

A

This is available as of course, so you can certainly subscribe for that as well. But let's see what it takes.

A

Now, as a Java developer, we can look at our application. You know this application is very simple. The spring boot application likely you are building your application using maven greater one of those tools. Hopefully you have a script. You know you're not going without a script. That's an important aspect. So we'll talk about it and I as B was talking about earlier. The first step on your journey is really to containerize your application. What are my choices for containerization shouldn't?

A

You use a maven plugin Gradle plugin should I use darker file, blocking lots of choices, so I'll talk about those as a Java developer. You know Java is known to be heavyweight, so let's take a look at it. How can we make our images? Small, that's an important aspect, because thinking of micro-services world, your applications are scattered all across and on. They could be downloaded at runtime at any point of time. So you don't want a big image size. That's important! Keeping your image size! Small, is fundamentally important.

A

So we'll talk about how you can create a quick image, particularly with features introduced in JDK, 9 onwards and I'll, give an example of JDK 11. Then you wanna pull the test using kubernetes cluster on your local desktop. So once you have done talkin ization lets you go onto the local testing part of it. How do you deeper get test such applications using the usual tools that are available is for IntelliJ, NetBeans and clamps. You know how do I D bug my kubernetes spots using those tools, then alright, my application looks good next.

A

Step in my journey is really to put it into the cloud, so I work for Amazon. There are other cloud providers who offer a managed kubernetes service. So I'll talk a quick example of how you can get started with, and it's on eks, which is our managed service and how you can migrate your application that is running on your desktop to very easily into the browser's very seamless part of it. Once you have done that, then you're really looking at it. Ok application is up and running now, observability across my entire application.

A

I know, in that case, I wanna, see that which service can talk to which service you know I want to see. If anything goes down, I want to see do every deployments or AP deployments. How do I do that applications and last, but not the least I want to do.

A

Maybe deployment by plants so feel free to go through this repo at your own pace, but I want to cover only the key topics today, where this is important to understand, because, most of the time when we talk to customers, the first thing they say is how do I get started on the journey? That's the most important aspect. Okay, once you understand what the get on the journey, which is a talker ization part of it, we'll talk about that first.

A

Well, let's look at the application here. So if I look at my application, it's a simple spring boot application here and if I look at my grading controller, nothing fancy just simple rest endpoint, you guys know to build far better much more complex, Java applications. So that's not the focus of this talk here. I just want to show what other application could look like, and how would you doctor eyes is essentially now in order to it.

A

What you can do is, if you're using in this case, I'm using maven so there's a tool by the Google container tools. Team is called as J. There are other docker maven plugins one by a fabricate team and that have one by Spotify, but the tool that I like really lot is J and it works very well. So, let's see how it works.

A

So this is my usual pound or XML here and I specify the JIT maven plug-in in the popular XML profile here, okay, in the profile, essentially, what I'm saying is I can do. The advantage of this is I can do maven package, and that will give me a jar file.

A

A fat jar file or I could do maven package and I'll give it a profile, then will give me a darker image as well, and the biggest advantage of this is I can use my maven language, the language that I'm used to to create my darker image, in this case I'm specifying use open, JDK and JRE as the base image. What the image name is going to look like what kind of options I want to put it in there right in the talker files could be scary.

A

Sometimes you know I mean there are documents written around. What are the best practices to write a docker file? So if you are starting new with docker, yes, you can start with docker file right away, jump into what kind of pace Amit. How should I use it? What should be my Fromm? What should be my ad? What should be my copy might starts running around in other tower developer.

A

Give me my favorite language and I'm happy with that and that's where business would comes in very happy, so it can easily build your docker image using maven language, push it directly toward our repository or build it to your local docker instance as well.

A

Well, the other option is, you can actually use a dropper file and in that case, what I'm doing here is instead of using a single docker file, like you know, instead of saying, hey, I'm gonna, build here and then I'm gonna rock my docker image separately, I'm using one file, one docker file and it's a multi-stage docker file, and in this case, if you see line one is what it says: maven as a base image at the end of that line, it says as Perl what that means is that is really my build phase and then my second fail starts with my second from statement kicks in.

A

In this case. My second statements phase starts from line eight, so what I'm doing is, in my first phase, essentially I'm using maven as the base image I, don't even need to have maven downloaded I'm purely using docker. As my build tool, as well so I'm using docker as a build tool- and that allows me to run my maven- you know have my maven repository- build my image because I remember when you build with maven it downloads the entire internet, on your machine. That's not what you want any rebates, though. So that's your phase.

A

One phase two is when you say all right: you know what, in from the phase one pick up my jar file on my war file and now use JRE as a base image. One of the big anti-patterns that we have seen in customer cases is where customer are using actually JDK as a base image. Do you really need unique as a base image think about it if you're using JSP?

A

If you need to compile at runtime things like those where it matters to have JDK as a base image, otherwise switch from JDK to JRE as a base image? That's important design pattern. You want to think about because it really pretty mmm science, there are JDK and JRE. Difference is huge, particularly if you are using JDK 8. Now, if you're using JDK 11, then you can do other fun things.

A

Jdk 11 has these: have this tools collars, Jane, Depp's and Jane link now remember: Java 9 introduces capability by which it has modularized that are JDK, so it kind of split the journey can do multiple modules and then what you can do is you can create well, just jdk level does not ship with a theory, because the idea is, you got to create your own custom JRE with it. So what you're doing here is, in this case I'm saying, take the jane depp's tool, which is right here.

A

Take the text tool, give it to my target jar file and give me the list of modules that it depends upon, in addition, also think the modules that springboard is dependent upon. So now you are taking all of that together and using the j-link tool to create a custom journey, and those are my standard jdk tools. Those are available, whether you are using amazon, correct or whether you using Oracle JDK or any other distribution of open JDK.

A

So both ways it would work, and now that you have created your custom jewelry, you can actually get rolling with this. So if I go back to my application.

A

So in my application now or in my app I, have a talker file to a cherry, and here I'm, just saying use Debian 9 as a base image I'm having my base JRE generated in that directory then have using the Java or the JVM from that JRE is invoking my jar file. What is that Vantage? Well elastic alone,.

A

The biggest and what is that you get as a customer, your image size shrunk from literally 490 megabytes, 262 megabytes. So basically, you are cutting down to less than a third of your image size, so always pay attention to. What does it that matter to you? How can you cut down your image size because that's what needs to be lean? It's not about this story is not about memory storage.

A

It's about how fast you can download that image on your local machine and how efficiently you can bootstrap that image once you've done that then now you're looking at okay, how do I test this file? How do I test this darker image in a kubernetes line? Well, the way you think about this is the way there is darker file, which kind of defines how the darker image needs to be generated. You need to look at sort of a kubernetes manifest you need to understand.

A

You know, go to kubernetes, don't I Oh excellent piece of documentation over there. What different, kubernetes resources need to be created. It talks about part services. Replica sets deployments all the basic concepts that you need to understand and then what you need to do is you need to really create your kubernetes, manifest that generate those resources for you. Ok, now you write the manifest files. You need to provide those manifest file to and kubernetes cluster, which will basically create those resources for you. So how do you start a kubernetes cluster?

A

Well again, there are multiple ways for that. If you are running, but my preferred way is to offer desktop because I have running on my machine, that's how I create my images so I just literally go to docker desktop.

A

So, let's go here: I go to the Opera, desktop and I, say preferences and with the click of button Oh later I go on this kubernetes tab here and I just say able kubernetes, and that literally gives me a single node kubernetes cluster on my machine very easy to get started with that's one option and I've seen that customers use a lot.

A

In addition, you can of course, use mini cube and there are lots of other flavors and varieties are available. That gives you a single node, kubernetes cluster, but pretty much by far, and maybe you are the two most prominent officers that you have seen now once you have cooperate, is running on your machine, then you need you have the manifest file. What you need is you need to be able to talk to that kubernetes cluster and that's where cube cotton, which is a CLI to manage your kubernetes cluster comes in.

A

So you download the CLI simple brew, install on a macbook and then you say: blue install, kubernetes, CLI and then you say cube Caudill, take this manifest file, deploy it to a cluster and then it will create the resources. What is the biggest advantage, if you think about it from the kubernetes perspective of developers perspective? Let's say you say, create four replicas of my application. Okay, it's gonna, create those four replicas and the biggest advantages is going to make sure that those four applicants stay up.

A

If an instance goes down, kubernetes has built-in mechanisms that bring. Is that replica back up again? So that's the important part from a developer perspective, because then I can focus on my application. As opposed to like is Maya planning, is it meeting the quality requirements you still have to do casting and all those things, but you know you are differentially revealed everything is done by kubernetes, for you that's the advantage.

A

The other part you want to think about is when you are building a micro services based application. It typically is not just a micro service. It's probably a bunch of micro services is the matter of fact. You know it's not uncommon to have application with. You know, ten to twenty to even 5200 micro services.

A

That's very common actually now think about, and each of those micro services are scaling up and down so step back for a second now you're building your application java application lots of different concepts over there, plus the different darker files over there lots of different maven plugins over there. The different docker containers over there. How do I manage them as a single application? That's where help you get set so kubernetes sound is a way by which you define and declare kubernetes application very easy to use.

A

Actually so, just like you know, you install a cube, cut' and CLI. You can install brew, install kubernetes self, then you at least until he'll to the architecture is this architecture is changing and evolving in the next version of help, but for now at least you can install a client component of health and a server component of how and that server component is sitting in the kubernetes cluster.

A

Once you have install that, then you write up a hell chart as they say, and then in the health chart is where your product or the qubit is manifest, will go. So, let's take a look at our health chart here.

A

But before that, let's take a look at our deployment manifest. So this is my deployment manifest, and it's saying this is apps Persian v1. This is a deployment manifest I can give it some meta data, I'm, saying a single replica. Putting some more metadata kubernetes creates what we call as pod. It is as opposed to using container as the basic primitive it uses pod as the basic primitive now. The whole idea of pod is part can have multiple containers but typically correlated, and what, by that, what I mean?

A

Is you don't mean that you know you create your front-end and the backend, because they're related into the same pot, but like a sidecar container, you have one main application container and then maybe other sidecar containers. There are other things doing other things where you say: logging or security or service mash, which we'll talk about in a second. So that's another idea now, in this case my pod has a single container. Okay,.

A

Now I'm, looking at okay, I got one manifest, but I really want my application to have multiple of such manifest. So how do I manage that? How do I template eyes that application, so that you know I have some sort of a consistency across all of those manifests in terms of labels in terms of naming so go to my app here, and this is basically my helmet art. What I'm showing you here is the top-level chart or animal here which kind of gives a metadata.

A

The chart itself then I, go to templates and in here is where my greeting deployment, which is my manifest sitting over there okay. So this is my greeting deployment at Yahoo and you can see I'm using some templating mechanism here once I use a templating mechanism that can allow me to actually bring that consistency across all of my service manifests.

A

Another important aspect to understand: is you generally more commonly do not create the part by itself? What you do is you say, deployment is a kubernetes artifact, create a deployment, and then the deployment will create the part for you, because you wanna say how many replicas of the part you want, because then deployment make sure that that many replicas of the pod stays up. Okay, now parts are ephemera.

A

What that means is part starts up on a worker node and it is given an IP address, but that IP address is not a reliable in the sense. If thought ties and comes up on a different world, cannot you cannot refer it back and maybe give it the same IP but may not, if you wanna say might be so that's where what you do.

A

Is you create this thing called as a service which is a kubernetes service and that kubernetes service says I'm gonna have a very loose mechanism for this, because a part is gonna, be given a label and service is going to look for those labels. So if I go back here, I have my breeding service here and my greeting service up saying a selector tag, which is saying pick a release: name greeting and if I go to Bellini deployment in here, I'm saying app really is named greedy, so so very loosely coupled mechanism.

A

But that makes it that much more flexible. So essentially, you are leading a deployment, a service, a health card, and here you go. You got your application infrastructure up and running.

A

So then, the next thing is really to get started with kubernetes on the cloud. As I said, you know, a lot of different cloud providers provide their managed services. We, of course our customer fuels, have a zonie chaos all the time what amazon Eakins gives you is a managed control plane and you bring the data plane and connect to it. Now, if you look at our official documentation, it could like wow. This is daunting.

A

This is too challenging for me, but what we recovered to our customers, we works is a very good partner of AWS and is also a CN CF member, and they have written a tool called as ETS cuddle. Just like cube Caudill, it says ETS, column and ETS. Cuddle really allows you to create ETS cluster, with literally a single line. So let me show you how seamless it is so if I go to say, kubernetes cluster. Any cares on AWS.

A

Here is a command that I'm using here. Ok, so just that single line of code and regions is also optional, I can just say: eks cuddle, create cluster, give it a cluster name and how many nodes you want by default. It goes to you as well too, but depending upon the region that you are not deployed in it's totally alright. So that's the way you wanna look at it. You know I mean there are all there.

A

You can see available like about twelve plus regions across the world, so pick the region that is closest to you or closest to your customers, where you want it to be provision and you can create an amazon ETS cluster very easily. Now the cluster is created and your app is running on the local desktop. How do you switch the context from your desktop to the clock? So remember we talked about cube, Caudill cube. Caudill can be considered to talk to a cluster because it has the security credentials, certificates, etc.

A

That are needed to talk to that class. So you can say a cube, Caudill config. Let me show the command here actually.

A

So I can say you cuddle config, get context. It will show me all the contexts that are running or contexts that are available on my local environment. I may have configured multiple clusters so may I may have the aquifer desktop I may have a eks cluster I may have an ETS cluster another one in a different region and I would name them differently. So literally moving your application from desktop to the cloud is very seamless. All you say is cube. Caudill use a different context.

A

You change the context and then you just read apply the same application and that is very seamless and enabled by all of these OC icd tools that are available in the market. You know I mean, of course we have a managed service of code by play but feel free to use the tool that works. For you know, Jake is work for you go for it, j-rok works for you go for it. Circle CI, whatever tool works for you, whatever gives you that ability to just literally add those commands in there are automated and scripted.

A

That makes it super easy.

A

The last point I want to quickly talk about, is well. We didn't get to talk about the debugging part of it, but one of the things, if you recall, is if I look at my dog or file in the darker file. I have these things called as environment variables, okay, Java options if I fire up my IntelliJ and I want to say I want to connect to a remote JVM.

A

It will give me those options, so I literally dropped those options in my docker file and that's what allows me to debug my kubernetes pod in production and development. Whatever way you want, and the last thing I wanna talk about quickly is service mesh. Well, what is the relevance? I mean subbu talked about here today, the relevance of app mesh, but what we are looking over here is I.

A

Have all these micro services deployed I want a certain micro service to be able to a certain micro service. I want to have the ability by which I could introduce a new version of a micro service and send only I'm percent of the traffic over there now cooperate is by itself. Does not provide that capability? Typically, what you do is you have a deployment running. We say five replicas, but now you wanna say only five percent of the traffic.

A

Now five percent of the traffic, the minimum replica that you've got to run is one and one of five is twenty percent. So you can really scale it up to twenty, and then you can have this deployed as one and then you can accommodate for that five. So that makes it a little bit difficult for companies and that's where exactly the concept of service comes in service really relies upon the concept of my proxy, which is again a CNC F graduated project and envoy, is injected as a sidecar, as we discussed earlier to your part.

A

Now. Imagine you have those hundreds of micro services spanning across your cluster and for each part you have the envoy sidecar injected. So that becomes your envoy data plane. You need to manage that data plane and that becomes your history of control plane, because you want to be able to say which service talks the with service.

A

Now, if I want to do a canary, if I'm going to do a B, if I'm going to do control groups, whatever you want to do, you can control that using the is to your resources, so I think we may be running out of time, but I would definitely recommend taking a look at this and on this particular repo here.

A

Also, as I said earlier, take a look at the LinkedIn recording it's a deep dive course exactly kind of walks you through the poor samples and how it could be done, the last, but not the least, set pull requests. You know this is an open source world. We would love to see you pull witness from you. Thank you, I'll be outside for several time. If you want to talk, I would love to talk and hear thoughts and feedback thanks.