Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2020 - Virtual, 15 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Sponsor Demo: Rancher - K3s and Raspberry Pi to Go - Portable and Practical

Description

K3s is Rancher's CNCF-certified lightweight Kubernetes distribution designed for resource-constrained environments like IoT and the edge. It's fast and versatile, and to demonstrate, Sheldon Lo-A-Njoe from Rancher Labs has built a 6-node Raspberry Pi cluster into a Pelican case, complete with a video display and programmable LEDs. In today's demo he'll give a quick walkthrough on installing and using K3s with an introduction to Rancher for managing it. He follows this with a deployment of the chaos engineering tool KubeInvaders that uses the LED array to show the creation and deletion of Pods while shooting down an attacking alien horde. It's everything you can imagine - cool, awesome, informative, and fun, and you'll walk away from it with a smile and an understanding of not only how pervasive Kubernetes will become in the near future, but also for how ready Rancher and K3s are to meet you there.

A

Hi everyone, my name, is sheldon lawanyu and I'm a field engineer here at rancho labs. I will be giving you a virtual demo today of k3s our lightweight kubernetes distribution. Kfirst itself is very simple: to set up and use so I'll try to create a fun demo illustrating its ease of use and its potential.

A

I hope you enjoy now. Let's get started before we get started. Let's have a quick look at what kts is and how it works. Ktris is our cncf certified lightweight kubernetes distribution built for iot and test computing. However, we are seeing it being used in other environments as well, including the cloud and data centers features is packed as a single binary and contains everything it needs to set up a community's cluster by rule of thumb. The minimum you need is a modern linux. Kernel that supports c groups.

A

K3S contains everything else needed to set itself up, including for starting out with a fully functional kubernetes cluster with ingress service load, balancer network plugin storage, provisioner, container runtime and much more by default. Kvs uses embedded sqlite for a single master cluster, but can also be integrated with either mysql postgres or mariodb for an ha setup with an external database and of course, you can also use hcd no need to set up all components manually, as you only need one binary to create.

A

Your servers and agents k3s by default also uses container d, but you can also use docker or choose an alternative, cri implementation. Another useful aspect is that agents don't need to open up incoming ports to the nodes. The agent connects to the server to the tunnel proxy that sets up the tunnel, which is then used for bi-directional communication.

A

Okay, that's enough! For now, let's jump into the demo, let's set up k3s in its simplest form. All you'd have to do. Is this, but we're going to add some parameters, so it better fits our environments with my command I'll, be adding docker as a container runtime setting the correct cube content, permissions, binding and advertising on the correct ip and configuring the api server to more frequently check if a node is down.

A

This is just for demo purposes, running the default command and this one will set up a single node cluster, so the server will also be able to run workloads.

A

This is now downloading the latest binary setting up the system, the unit files creating symlinks, and it has even created an uninstall and killer script, so we can easily clean up everything. Ktrs has installed and configured when it's done doing its thing, we'll have a fully functional single node kubernetes cluster to play with, and there we go, let's check if everything worked as planned and there's our single node cluster now ready to be used.

A

Now that we've got our server set up, let's set up our agents in order to do that, we're going to need a token that has been generated. This token is located in this location, let's copy that and then we'll export it and paste it. In my other nodes, as you can see, I've got one control, plane and five worker nodes. Now all of them have that token, and I will now try to add them to my existing kubernetes cluster.

A

This is the command that we're going to use to initialize our agents, as you can see, it's going to get the same binary, but now I'm also adding. In that token, I just exported and I'm pointing it to our server.

A

This will automatically let k3s know that this is only an agent and it won't deploy the server aspects. Also, since I changed the server to more frequently look if the node is done, I'm going to upgrade the frequency at which the node is going to send its updates.

A

As you can see, it's doing almost exactly the same thing. Only now, it's only installing the agent and not the server component in the meantime, you'll see that the ones that are already ready are going to start coming up as soon as they're available.

A

Now, let's start deploying the application, but before I can do that, I need to first create a registry secret, so my application can access my private docker hub repository to do this. We will run qctl create secret and this will be a docker registry secret with the name recreate then I'll, add in my username, but we'll fill out my password and email off screen. Of course, we can then validate the secret creation by executing cube, ctrl, git secrets, and there we go there's my secret.

A

So far you haven't seen the infrastructure. Yet, however, this will be crucial for the application I'm about to launch here. You can see my 6 node raspberry pi cluster, the 8 pixel led strips, represent a node and 8 free spots on that node for running applications.

A

Now, let's launch the application. By going to the location where I've cloned my git repo. Here I have a pixelcontroller.yaml and it contains configuration to deploy a daemon set in my cluster. This daemon set will run on every node and will provide the controller for enabling and disabling those leds whenever the application is deployed or is deleted from one of the nodes.

A

The controller will run in the pixel controller namespace. So let's first create that by running cubectl create namespace pixelcontroller.

A

Why don't I just have that in the same manifest well good question now, let's also watch the parts as they come up and create the resource by running cube, ctl, create dash f for file, pixel controller jammon.

A

This could normally take a few minutes, but of course, with the power of video editing, I've sped it up a bit. So there we go. Everything is done. This took some time because the raspberry pi's had to download pretty big images.

A

Now, let's go ahead and install the actual application itself, that's going to be activating well triggering the activation of the leds, let's watch, while those parts come up as well.

A

Let's go to the singlepixel.yaml file here you can see it's just a deployment with a single replica that will deploy basically a single pixel application, which will activate one of the pixels of the eight leds.

A

Let's run it by doing a qctl create dash, f, single pixel, and here you can see it just spins up one replica, just added that so I can see on which node it's actually running now, let's scale up the amount of replicas by doing cube, ctl scale, deploy and then the name of the deployment, of course, there's replicas and then we'll add in a couple of more then maybe replicas.

A

So, while that's spinning up, let me show you what's actually happening on the raspberry pi cluster here you can see while the application is actually starting up. The leds are also activating on the raspberry pi.

A

Now that you've done a bunch of manual terminal work, let's manage and use the cluster in an easy way, as you might want to manage many of these types of clusters. For this we'll be using rancher our multi-cluster kubernetes management solution.

A

We can easily import the raspberry pi cluster by just going to add cluster and then import luster. Now, let's give it a name and hit create here we can get the cube ctl command we need.

A

So all we have to do now is paste that on our server as you can see, a bunch of resources have now been added to this cluster and the ranger cluster agent and node agent have been spun up now. If we go to rancher, we can see that the cluster is ready and available to use from the range management interface.

A

We can see the metrics and go to the nodes available in our cluster. Here you can see the taint I added to the control, plane, node and all of my available workers.

A

If I click on the control play, node you'll see that we get a bunch of information about the node itself. Of course, architecture versions, things like that, but also the actual command that we ran gatries with. So we can go back and see how any imported ktvs cluster was actually run. Of course, there's a bunch more information. We can see the configuration for the tools that have been used in the environment.

A

And we can do that for the workers as well of course, so of course, the agent node had a lot smaller command, but we can also see that from here as well,.

A

If I go back to my cluster and then to the project, then you'll see we have the actual workloads running in that environment, and here you can see this, the single pixel deployment with the zero replicas and if I start scaling up from it in ranger the same as doing it from the terminal. We see that the application gets launched on the raspberry pi cluster on one of the worker nodes and we see the pixel or the led actually activating.

A

If I scale up the amount by editing the deployment and not just pressing on the plus button, you see the same thing happening again. Of course, this is just a visualization of basically any application deployment.

A

As you can see, the deployments are almost completed and if we go back and try to bring down the amount of replicas using the ui exactly what we expect is happening in that rancher is scaling down my replicas for me and now I'll show you another thing that of course, normally you can do in rancher and that's of course manage your storage as well. Here you can see you could create your own persistent volume. If you'd like, but as you can see in your storage class, you already have a local pod storage class available.

A

This is there because, as I showed you before as well, kgs installed the local path provisioner, and this allows us to then also start using dynamic storage creation. However, this just uses the local disks of the machine itself.

A

Of course, we can also go in and look at all of the namespaces available. As we can see the pixel controller namespace we've got our daemon set running. I want to move that to the default project, so I can see the actual resources in that project as well. Of course, I could have created one with a pixel controller project, but I'll just put it in default. For now, as you can see now we see the single pixel and the pixel controller resources in our environment.

A

As you see the daemon set on every worker node in my environment, so you can watch all of these resources and manage them from within rancher, including things like secrets certificates.

A

You can see the registry credential that I just created as well and, of course you can also deploy applications to that raspberry pi cluster.

A

Now, let's have some fun well, I hope some more fun. Let's set up cube invaders and execute some chaos engineering on our, not so helpless getrios cluster.

A

In order to do that, we'll first need to set up a service account and cluster role binding for the cube invaders application. Then we need to get the token belonging to that service account.

A

Now, let's copy that token and open up the dot cube in file in our home directory in this file, we can configure a kubernetes endpoint credentials and which namespaces we want to be able to wreak havoc in now. Let's save this and go back to the cluster.

A

If you haven't played around with cube invaders, I would highly recommend checking it out. You can just do a google search for cube invaders and you'll find it on a kubernetes dot. Io blog, the author is eugenio marshall from sourcesense and it's fully open source, so you can also get the code and tweak it to your liking.

A

I have not deployed it into my cluster, as the images aren't built for my architecture, but thanks to eugenio, I can also just download the mac application and point that to any cluster. So that's exactly what I did as you can see. I can now play what seems like a fun little nostalgic game, but in the background I'm actually killing pods in my cluster's default namespace.

A

Currently, it's set to automatic and of course it has been sped up, but I could also manually play around with this, which I will this nicely illustrates how kubernetes actually responds, or at least should respond when applications are going down, the more resiliency you build in the less chances there are for you to get called in the middle of the night because of some app that just randomly fell over.

A

I hope you've now seen that, though ktris is mostly used for the edge and iot the possibilities are endless.

A

Any embedded device with one cpu and one gig of ram can be turned into a kubernetes server and or agent.

A

Well, that's it for today. I hope you enjoyed watching this demo as much as I loved shooting those aliens. Thank you so much for your time. If you have any further questions for us, please drop by our virtual booth and ask for one of our representatives we'll be happy to answer any questions you might still have have a great day and enjoy the conference.