Cloud Native Computing Foundation KCD Italy 2021, 20 Dec 2021

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: Knative goes beyond serverless | Alexandre Roman

Description

KCD Italy 2021* - The video of the talks are now available!

Serverless is a good pattern when it comes to saving infrastructure resources: why should you run apps when there’s nothing to do? The open source project Knative is often used to run functions as serverless apps in Kubernetes clusters.

In this talk, you’ll see how to leverage Knative for Kubernetes apps, not only functions. Check out how to apply serverless patterns to an existing Spring Boot / Nodejs app (backend / frontend) with a live demo.

🤓 SPEAKER: Alexandre Roman, Tanzu Senior Solution Engineer, VMware
👉🏼 Here the slide deck: https://www.slideshare.net/KCDItaly/knative-goesbeyond-serverless-alexandre-roman

______________________
* KCD Italy 2021 is the first edition of a community-organized non-profit event that will take place on November 17th and 18th and will gather adopters and technologists from open source and cloud native communities for education, collaboration, and networking.

📯 Follow us to be notified about the upcoming MeetUps: https://community.cncf.io/kubernetes-cloud-native-milano/

A

So it's it's absolutely great. uh The next speaker it's going to be alexander hermann. He is a tanza senior solution, engineer at vmware and his talk about k native, I'm sure it's going to be super interesting, so alexandra, your state stage is yours and please we are.

B

A

To have you bye, great great.

B

Great to be uh to attend the uh kcd event, um hello, everyone, my name is uh alexandroman, so I am a solution engineer working at vmware tenzu and in this talk we are going to talk about k native.

B

So for those of you who don't know about k native, there was um an interesting tweet a few days ago from matt moore, one of the uh founders of uh k native.

B

So, if you know nothing about k native and how great it is for any application and not such and not only serverless functions, uh I think you might be interested in in this talk. So let's see why uh keynative is great for your kubernetes application.

B

So a few reminders before we start when you want to deploy your application to kubernetes. What do you need you need to to have some pods? The pod is responsible for running your containers in in your cluster.

B

Actually, you don't have to deal with pod. You would create some deployment which will manage pod instances and will is responsible for launching as much respect as that. You that you need- and you would then you need to expose and to create some kubernetes service so that you can select and use the pods that you want to access uh if you have your application, but at this point the the application is not available outside of the con outside of the cluster. So that's why you need to either use some load balancer.

B

So the idea is to create uh this object so that you can uh get to your application from outside the cluster. Maybe you you might want to use some ingress controller and create a route so that you can use the application from outside the cluster. So most of you already knows about that. uh You know that uh there is. This is a lot of files.

B

This is a lot of yaml that you, you need to to write in order to deploy your application, and I think everyone can agree on the fact that this is too much yaml uh kubernetes was um created uh for operators uh as a way to you know uh having a great abstraction layer over the infrastructure, so it's great to consume cpu network storage uh resources, but when it comes to deploying applications, you know that there is a lot of work that you need to do as a developer.

B

Until you have your application running on on your cluster and as a developer, we know that what developer really wants they yeah. They want to have fun right, but this this also they want to to write code. This is uh what they want to do. uh They want to write code, and I think we can all agree on that. Right.

B

Writing infrastructure code such as yaml files, the load balancer, the ingress, is it's not fun as a developer, and you can imagine that if you have a lot of applications, you have a lot of yaml files that you need to write that you need also to maintain, um because uh this is a lot of files uh that you need to uh to deal with as soon as you want to deploy your application to kubernetes.

B

Fortunately, uh there is this thing: uh key native canadian is a kubernetes framework. um You may know about canadia.

B

It is famous for uh serverless to run functions as a service, and uh but it is much more than that- and we see in this talk how you can leverage canadian for a lot of applications, not just serverless ones, and a few days ago uh there is. um There was this joint announcement from ibm vmware and google native went ga with this 1.0 version, so it is now a product which is ready for prime time that you, you may want to use for your next application on on kubernetes.

B

So how great k native is for your application. So let's have a look at this definition. Yeah! That's right: if you want to deploy some application to kubernetes, you can have this yaml file. You know that it fits in a slide. So that's good news for everyone such who are afraid of writing too many yaml files.

B

That's all you need! That's all you need to get an application running on kubernetes and what you get as a result. You get pods of course running in your cluster these clusters.

B

These pods, will be managed by your deployment and replica assets. So at this point it is no different that a regular kubernetes application. You have pods replica, set and pods running in your cluster. What you get for free is horizontal product scaling uh native, and we will see that in a moment you get. You can create new instances for your pods, using uh using key native and, what's great, also with canadian, is as soon as you start using this definition with your cluster for your application. Then you can access.

B

uh You can get an url uh to connect to your application and you don't have to deal with ingress with service. Everything is managed by a native, uh and I think it's really great to have that if you compare this um way of doing things uh with the legacy way uh on the left, you see uh everything that you have to write in order to deploy your an application, get some pods with the deployment. Then you have to to write some service and ingress. You may want to use uh auto scaling with horizontal power.scaler.

B

You know that you can see here in less than 10 lines of yammers. You have the same result and it is way much better to maintain and to create at for your for your application. So let's, uh let's have a look. uh I have this definition here, so uh you just have to push this file to your cluster and as soon as you do that you, you would get a canadian service created in your cluster.

B

So if you use the clear native cli you'll see that I have an application, I have a native service with some url. I can go further and show additional details about my application, so I know that my application is ready. I can access, I can get the the url. So let's do that.

B

And as soon as I am using that endpoint, a new pod would be created for my application here, you see it is a spring boot application here you see the pods running on my cluster just 10 seconds ago. So if you go deep, if you go further, you see that it is just a regular pods running in my cluster wait a minute and this uh this application is ready. For you know you can access this application anywhere using the the url.

B

We are running a live demo, so, okay, you, you may want to access this application. So let's have a look at the at the logs. You see it is a regular spring boot application, I'm not using a functions framework, I'm not using uh um any fancy framework. It is just uh the the regular springboot framework that you may want to use for your application. Of course, you may want to use any framework as soon as your application is a stateless using http.

B

You see the application started on the on my cluster and around around the three or three seconds to go to to get to the to the application running up and running. On my cluster and at some point, the the pods would be destroyed, so yeah we can see the pod is about to be terminated by k native, because I am not using the application. There is no traffic, so the application is about to be uh destroyed.

B

uh This way, I am not consuming resources on my cluster and I got this for free as you can see, and I don't have to to write a lot of jumbo files to in order to get uh that result.

B

So let's go uh deep uh into the k native features and then we will run uh two demos uh that will show you how to use this feature for real. So we have just.

B

We just saw the auto skating feature available in k native, so I think I believe that everyone understand what we are talking about when we talk about auto scaling uh when you have requests uh for your application, key native would be responsible for starting as many pods as um required in order to you know, uh get the uh the job done and when there is no traffic, and we saw that kenneth was able to destroy the uh remaining pods so such so that you don't consume resources if you are not using the application and as soon as you have a new uh request coming then, and the uh k nhf serving uh component would take care of this uh request and would recreate the the pods so that you can have an application running.

B

You can also have some advanced rollout support. It is a neat feature that you may want to use. Let's say you have v1 and v2 uh using connective and connective. Only you don't need to install style or any other framework.

B

You can then send the traffic between the two versions of your application. So let's say you want to send uh most of the traffic on the v1 uh version of your application.

B

You can set up a root with k native so that you, you send uh some part of the traffic with the to the to the new application. With uh this advanced rollout support, there is also uh some there is a great eventing model available in k native. So the idea is to connect part of your system, part of your I.t with these k native services, so that, as soon as you have some event which is fired in your system, then the native service would be triggered and would be would handle this.

B

This event, it is quite handy if you want to you, know, leverage some external services such as github or or s3. So let's say you have a new issue created on github. It could be a new event that your creative service uh could uh manage could end all. uh Let's say you, you create a new file in some s3 bucket.

B

uh You could um you could take this file and uh this a pod, a connective service, would be then responsible for um you know, reading the file and do things with this. With this file and the nice thing with this eventing model, uh you can also uh get you also get integration with uh message brokers.

B

So if you in case you have a kefka or habit mq, then you can build a complex uh eventing model and you can integrate with existing queues and existing events, inventing architecture in your system. So this is, you can do a lot of things with k native. So let's see what we can do for real with uh live demos.

B

So, uh first of all, I am, um I'm gonna show you an existing application, so my application is, you know, nothing, nothing, fancy! I have a to do application, so this application is made of two components. You have a front-end container with view.js and we have a back end built with a spring boot.

B

This application is using, you know, regular kubernetes definitions. So if we go to the deployment you see the deployment you you see here, the front end with the liveness plug the readiness probe. You have the the same information for the back end, a few configuration in order to connect to my database.

B

This is something that you already know, uh and it is running on my application on my cluster, so let's say hello and I I have created a new entry in the database using the using the back-end, so nothing fancy here.

B

Let's see how we can convert this application, this deployment to a native service so that we can leverage the auto-scaling feature available in in canadia. So I have this this application. The name is k native to do application.

B

The idea is to copy past really copy past the part of your deployment that you need and I need some image for my front end. I need some. uh I need the image for my back end. I need a few configuration details. This is this are the same configuration entries that I am using with the kubernetes application, the all the all the other configuration files are not required.

B

I don't need to deal with ingress. I don't have to deal with services. Everything is managed by kubernetes and k native, so that I have my. I have my application running, so this is the regular application. You see, one part for the back end and one part for the for the front end. Let's switch to the connective uh version of this application and in this case you see that we only have uh the database running in my application and I I have converted the existing kubernetes files to canadian services.

B

So you have the front end and the back end, which are now k native services. So, let's see what happens as soon as you start using this application. So here you have the application available.

B

If I hit refresh, then you see a new front-end container is created and since the front-end is uh using some api to connect to the back-end, then you see a new instance for the back-end. So I can create and say hello, pcd.

B

Okay, everything is managed uh by the back end, all right and as soon as I stop using the kubernetes the application, then the pods would be destroyed, as we saw previously. So now, let's see what happens if we uh start to, uh if we see if we send uh some traffic, so I have this uh this script.

B

I'm about to send a lot of traffic to my application. I want to to to see what happens with k native, so I'm about to start this script and, as you can see, uh new pods will be created as as required for my my application. Hopefully it should it should start.

B

In the meantime, we should see new parts created by canadian services. Okay and this uh these pods are automatically uh managed by k native. uh I don't have to make any modification to my existing connective services, so in the in the in the real life, maybe you want to tune the um configuration of autoscaling.

B

You may not want to create too many instances. So that's why you see this annotation available in the connective service, and it this is something that you may want to use to. uh You know put some constraints on the on the native service, and if we go to the logs of my application, let's see okay, so this is the same application made with spring boot and if we look at the uh starting, you see this application was created in a matter of milliseconds.

B

uh Why? That's because, uh in this version of my connective application, I am using a draw vm with native image support. So that's why my pods uh for my back end uh would really would start really fast, but you are you don't have to to uh to use uh this uh vm um feature with springboot. You may want to use the the the regular uh you know java uh framework runtime uh to to start your your application.

B

It would work the same with with k nature, so it is a way to showcase the you know the differences between uh canadian and kubernetes and, as you can see, there is uh much less files, much less yammer that you need to write in order to to run your application.

B

So back uh to the back to the demo, I have another fencing demo and I want to show you uh so this one is about eventing with kennedy. You can build a great event-based applications.

B

In this case I have an application which is using machine learning and so that we, if you put some image in some s3 bucket um as a png file, uh this um this installation would listen to the to the event coming from aws s3.

B

Then, at this point, uh this event would trigger the creation of a tensorflow server, which is responsible, for you know running the machine learning stuff that we need in order to recognize the uh car uh plates uh in this image and as a result, uh these uh these plates. If they have been um read successfully, you see the result in some google sheet document that I have here. So the idea now is to connect to the s3 bucket. I have a few png files, so the idea is to put these files to the s3 bucket.

B

And as soon as this file, these files are loaded in the bucket. So I need to upload the file all right and in the meantime, we are going to switch to the right, kubernetes name spaces. So you see a new name. A new pod uh is created for in order to manage the event and um in a few seconds as soon as the plate is read by the tensorflow and the machine learning stuff running in my cluster, you see, uh and it can be. Quite uh you know, resource intensive.

B

You know that running gpu running running machine learning may require some gpu may require a lot of cpu available in your cluster. So you don't want to run that stuff every you know every day always and you see the results as soon as you have the the plate.

B

You see the image which has been successfully read and you can, of course, if we go back to s3, you can put many more plates and the system would create as many pods as required in order to read these png files, and you see that it is a great way to you know, connect existing parts of your uh infrastructure of the systems that you may want to use in order to leverage the the pods and the cognitive services available with this framework, all right, so in a few in a few seconds we are, we will see the the new results available in this in this document, all right so back to the presentations.

B

We have, uh we saw some uh great use. Cases for k native so is uh here is my my recap: let's go to the next slide so for your next application. Maybe you want to use k native uh if your application is stateless. Remember that kennedy will kill your pods. So if you have no traffic, uh let's say you have some user session stored in your application.

B

If the pods is destroyed, because there is no traffic, because there is no nobody using your website, then the pods would be destroyed by canadian and you may lose the user session if your application is not stateless, so make sure that you stored everything outside of your application and there are many solutions available in order to implement this capability, you may want to rely on eventing.

B

We saw that you can leverage brokers such as kafka and rabitanku in order to connect a different part of your infrastructure with canadian services.

B

Remember that k native is a cross-platform serverless framework, so this means that you can deploy your functions as a service on to any kubernetes distributions. You may want to use canadi for your kubernetes running on premises. You may want to use any cloud provider just everywhere. You have everywhere. You have some kubernetes cluster.

B

You can use key native, so we we saw in the presentation that you may want to use also the proportion of phase releases so that you can implement features like a b testing and, of course, you can optimize the resources uh with auto scaling uh you, you would only consume the resources uh which are required uh for your application. As soon as you stop using the application, then the resources are freed.

B

So this means that you may use you may deploy more application to your cluster because you have resources available when you don't use the the application- and you know one of the most important feature for me. Last but not least, you want to write less yaml that that could be a good reason. You know to use k native uh as a way to write way less yaml files.

B

You know these days with kubernetes. There is yaml everywhere. Any solution that uh can allow me to uh write less yaml files uh is awesome uh to me. So uh this is uh this could be. You know the uh a killer feature for uh 4k native. Okay, so as the conclusion um you may want to try and learn creative uh you may want to go to learn.tensor.io.

B

You will find free workshops so that you can use canadia with live kubernetes clusters- you you you, you will see that uh there is uh some tutorials step by step. uh We go. We are going to uh show you how to use canadian uh with your kubernetes clusters.

B

If you are looking for tips and tricks there is this great website available canadian.tips, and if you want to go deep dive with the source code of my application, there is my github so feel free to go to the to the github repo and see how to implement canadian for your application and, in the meantime, if you have any questions, feel free to ask in the chat or feel free to reach out to um to me with my twitter handle, and thank you guys. Thank you, everyone uh for attending this session. Thank you. Thank you.

C

Alex it was a great talk. Thank you. Thank you for contributing. uh We have some time for questions. So, let's start, is it possible to use k native on kubernetes managed clusters.

B

Yeah, um as I said uh as soon as you, the only prerequisite is kubernetes. So no matter, if you have kubernetes, uh if you are using, you know a ks. If you are using a gke or any kubernetes, any managed kubernetes cluster.

B

You are free to deploy a key native and use the auto scaling, for example, on with this cluster.

C

Okay, there is another one: how does k native, compare to other framework for serverless.

B

Okay, that's a great one. There are some frameworks available out there to to implement serverless.

B

The nice thing with uh connective um is the fact that it is a community project and which is backed by uh some big companies out there. So you you see that you have ibm vmware, google and many many companies um uh which are contributing to this uh to this project, so it is quite mature, so every of course this project will continue to evolve over the time, but uh it is a safe bet.

B

If you want to use canadia for your serverless for your next serverless application, and especially when you you take into account the fact that you can deploy canadian everywhere as soon as you have a kubernetes clusters available, yeah.

C

Okay, thank you alex. We finished our time, so thank you very much for participating for contributing to the quesadilla italy. It was a pleasure to have you here. Yeah.

B

Great pleasure, thanks for having me bye-bye.

C

Bye: okay, italiano.

C

D