Cloud Native Computing Foundation Kubernetes Forum Seoul 2019, 10 Dec 2019

Previous Meeting

⏯

youtube image

►

From YouTube: Understanding the Cluster-API Structure Through the Openstack Provider - Jaesang Lee & Esther Kim

Description

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

Understanding the Cluster-API Structure Through the Openstack Provider - Jaesang Lee & Esther Kim, SK Telecom

The Cluster API is a Kubernetes project to bring declarative, Kubernetes-style APIs to cluster creation, configuration, and management. It provides optional, additive functionality on top of core Kubernetes. The Cluster API allows us to automate the deployment of Kubernetes
in a more advanced way. In this session, we'll use the CAPO(Cluster-API Provider OpenStack) to build Kubernetes on OpenStack VMs and learn about how Cluster-API works and how to implement it.

A

Yes, now I'm going to start the presentation. Today I unintentionally took on the last topic. Everyone worked hard. I'll try to do the last topic with a simple and simple mind.

A

A

Today,. We will find out this file structure of the cluster through the capo cloud. Api probio phone stack that can build Kubettes on Open Star, and let us know about the application cache for it. We will inform you later.

A

The time was not that long,, so I was worried that the time would be shortened a lot if I just did it with the code base, but fortunately, after the lunch break, in the first class,, someone gave a presentation about the cluster api. Those, who have heard it once understand a little more I think it would be easy to do.

A

Hello,. We are working at sk, telecom, and now.

B

A

A

Next, we will learn about the capo of the provider of the OpenStack application cluster leaf. At, the end of the announcement, Yo Capo is actually deployed to the off-stack. We are using, and we will show you how the vm's Cubates is built.. I'll show you.

B

4 Then I'll talk about the emergence of the cluster api,, the first topic of motivation. The.

B

Most representative way to build a Kubernetes cluster is the command line using Cube adam,, as shown on the representative. Site. There is a way to install it with this method. This method is the easiest and can be installed without any other knowledge,. If you know only Guo Netis.

B

When it is,. It is not actually possible to.

B

Use it,, so these solutions and various solutions have appeared.

B

If you have only this knowledge,. You can write a script for this and connect to only one node and run this script without having to connect to all nodes,. So you can easily install a Connetis cluster However,. There is a limit to this Provisioner. Once. You have to install Kubernetes for installation,, not only Kubernetes,, but Defender c for a separate Provisioning language is added.. I have to.

B

Study about you, separately, and I have to do all the installation process with scripts. So. It's not code,, but it's not as difficult as code,, but separate management is also required.. You need.

B

To write all the scripts, and since all of the installation, scale, upgrade, and uninstallation above are one-time operations,. You cannot constantly monitor the server status or give an alert to the situation when this node is down, or the user can not recognize it.

B

And. You can create and delete instances based on apr in public cloud or private cloud environments that are used a lot these days,, but these.

B

Color creation and deletion cannot be done with api, and all users directly access the aws dashboard before creating instances If you are using a giveaway or bare metal,. You need to manually install all the defender seals necessary for the base,, such as os installation or menu.

B

B

As, a covernetes style api,. It is an api that can create, manage and configure clusters. It provides functions to manage the life cycle of aucubitus cluster.

B

The picture is very similar,, but the worker writes a computation for the cluster and now gives commands to the bootstrap cluster. This bootstrap cluster is now a very small, minimal cluster, with only 1 master. A controller or card for deployment is installed.

B

Pass the cluster compaq through this cluster,, and if the infrastructure type written in this compig, follow the instructions or aws learn what metal, openstack,, etc. I will issue the command,. Then I will actively create an instance with the api. According to this infrastructure, and the instance will now run the cover 4 test, bootstrap, step from the creation to the instance.

B

So now, before, I manually created the instance directly If. You do not have to create or install the 5-day liquor directly in the bare meta,. If you just write the cluster comfrey well,, you.

B

Can already create it actively at the cluster level without being very passive.

B

Here. Until now was the concept for the cluster api. Now. Let's soak in the intro to the epi project. As always,, the billion cluster api project started in 2017 as a sub-project of Mr. Cluster, Life, Cycle and.

B

Has been in progress for about 2 years., The goal of this project is this The answer has not yet been voiced,, but now the goal is to use declarative APIs to manage the overall life cycle of components, from creation to scale, upgrade and destroy,.

B

And we aim to support all of the Um Prem public cloud and the default Since. It provides free management and the implementation may vary depending on the type of infrastructure,. The interface makes it possible to support those who want the same,.

B

And one of the goals is to create an infrastructure capable of self-healing.

B

As, if the deployment is warm at 8:00 or if the deployment is warm,, it goes to self-healing. Even within Kogan's test cluster,, when a node becomes warm or an instance becomes warm and ready,. Another instance is added and dynamically manages the pluster health. We plan to support up to 4 such functions. That do.

B

And next,, it is not a pro of the project. The goal of the project is not added to the Kubernetes default. Api, and components not related to the cluster are not managed. And. You are already using a lot. Management tools such as ck or ak seks are going to have no effect. Now. It.

B

Cannot be created with the cluster api.

B

So, following the cluster api,, the actual components will now be explained.. The first component is the infrastructure.

B

Provider. This is the source required for installation. And secondly,. You need a bootstrap provider. The bootstrap pro fighter is used for bootstrapping. The cover netis cluster.

B

It is a cube adm that is used a lot with the command. Line. I modified it a bit more and created a separate project. Now in operator, terms. I am using the operator cube a dm. The.

B

Third component is crt. It.

B

Is a function that can be customized according to the lead of . For, the billion cluster api project,. We created four city, IDs. To, explain in detail about the Lee ID,. It is now the first cr ding cluster. As. You can see from the name,. Now it is cluster 1 for the entire cluster. The configuration for the dog is now done with a 5-part object, and in this, you can specify 4 details such as the network settings of the cluster or the version of the cluster.

B

Next,, it is a machine. This machine is a configuration for one node of the cluster So. Here you can put a comp like q, black version or bootstrap comp. The.

B

Third cid is a machine set,, which is the same concept as a replica color.

B

So,, the machine has a shape for the group,, so the replicas are called Fives. In. This example, Fives are called Dance 0. Five machines with the same shape are created.

B

Lastly,, it is Machine Deployment. This is Kuber. It is the same as the basic resource n deployment in Netis, so it helps you to create and manage a set of machines to deploy what you see, and to do scale updates and grade rolling updates.

B

So here is an example. This is a very similar specification to the deployment you are using in the past by giving a strategy and a rolling update, strategy.

B

So cluster api providers are divided into projects depending on the type of infrastructure provider,. So the type of provider brought here as an example or aws is cloud open, stacked., In addition, there is a project called Metal Cube for bare metal, and talker So, I.

B

Would like you to note that the basic framework is the same depending on the infrastructure,, but the actual implementation is slightly different in the frame entity.

B

Ica phone creates a vm using Openstack, and configures a cluster cluster with that vm. And. It has a separate cid., The controller implementation. According to.

B

3 will be explained as a picture of the actual capo in operation. The. First thing you need before installing the capo is the cluster city. It.

B

Is a basic tool used by all providers,, as well as all pros,, in addition to .

B

Now create a drug file for the cluster machine and control plane that will be washed next to the rare in this cluster city and pass it as a factor Now. If there is a file to be done and the cluster is Terry,, pass it as an argument and give the command Create as a command to create a boot staff, pluster. It.

B

Will be installed, additionally. So here,. In the case of capo, there is a capo controller, and controllers called capa, and cap bbk are created. When.

B

They are created,, it recognizes this and issues commands to Openstack. So, each of the key ston 1st villages in Openstack now requests the creation of volumes, bootstrap images, network settings,, etc., and v.

B

After creating the vm,, you will initialize the master of the master Kuber Netis like this through the bootstrap provider,.

B

And then create an additional vm and join it as a worker node to use it, so that 1 cover netis A cluster is created. Now,. If the above cluster was an Ubuntu cluster in the written example,, if you change the number of times that only the image version is written in the water file,, the Ubuntu version of the cluster in OpenStack and the cluster in the centos version of the example and It can be configured together.

B

This was an overview, and the real demo is now going to be given by the boss.

A

Now,, there are many parts that can be felt while typing,, so we will record a video and show you what kind of setting files are applied and how they are executed..

A

It's, cute, again, and I 'll explain it by dividing it into large built-in pieces. First of all, I'm, going to create a config file for cluster stakes like 50 billion clusters or something.

A

When, the bootstrap cluster is created,, first, create a vm on the Opus side, run the master node with the master node and its comment. S, and finally add a worker node that can communicate with the master node. I 'll show you.

A

Before using this capo per day,, this source for applying the openstack, a capo must be defined in advance.

A

Black Tube Steel has been downloaded in advance,, so Capo does not automatically install this package. Because. It was developed assuming that these packages are all images,. The image must be created in advance.

A

In, the case,. There is some cloud in version: 28.3 installed,, but again,. This cloud in file used by the cloud aplaster api is a bit damaged,, so you need to install 19.3 separately. I.

A

Used the disk and set it up so that I can connect to the talent network with my public ip.

A

Do the work first. Yes,, so after downloading, you will need to know the open stack information. First., So, now, write the open stack information in the tile called cloud field.

A

I'm going to write down what it is.

A

Now I gave it the name: Cloud's Taco Postic. So,, with this name, Taco Openstack, I'm. Now.

A

Doing the Gino Rating. I gave the.

A

Command to do this for the channel with the settings as a single in the taco open stack in Cladoya water and the out folder.

A

Then,, the script is downloaded and a skeleton file is created..

A

Now,, the skeleton files to be set have been created.. We will change the settings of the j skeleton file to suit our environment. It.

A

Has the name by default now,, so I can't change it to the name. Taco Cluster. Then I changed the network band to be used by the cluster.. You can just change this. According to your environment., I wrote down the ip,, wrote down the network id, and also wrote down the name of the private network to be used for my normal communication.

A

This is how I set up the cluster. File. It contains the settings for the node, so when connecting from outside or when workers connect,, you need to set the endpoint ip. Now. We will assign a floating ip in Openstack and write down the information in the floating ipad.

A

And now, in the cluster The version of its version to be used in p&I is also now specified.. I set the version now to install the 1 point: 15.3 version liquid process as a test version.

A

Write down the version information.

A

And use it in the vm. Now, we have applied something like a pre-appropriate flavor and something like a window image floating ip information or something else, such as a key pair name or security group.- So, in the case of Citigroup, Because. The language basically needs to communicate with Kuvas now,. You.

A

Need to create a secret group that has 443 ports,, 643, ports, and ports. Open. You have to open it. Together. So. You have to set these things in advance.

A

So,, like the cluster file, I, first change the test cluster to the taco cluster,.

A

Then I change the user,.

A

And then I change the version of the test. R. Since the file is provided,. You can change it to suit your environment.

A

Now, I will change the file for the amount of provider component.

A

Finally, now there is a water file to use add-ons. Whether. It is cli or csi,. It can now additionally contain information about the plug required for the process,, and now it contains settings related to Calico. So,. The iklan map used by Calico I made the changes,.

A

But now the third second step will show you, the third opportunity. Now, that the configuration file is over, I'm, going to create a bootstrap cluster.

A

A

Confessed that it is an independent JIA firmware that will literally do the work of bootstrapping. So here,. The container-based Cubates called Kind is now installed.

A

Communicates with and openstack communicates, and then the master vm is created. So, when the master snake is created,. The three controllers currently shown in the bootstrap cluster are installed on the master vm,, so the purchase test is installed. There. Move, on., So,, I use the word p see,, but after the booting process is finished,, the life cycle is now over in the bootstrap cluster,, so the dog ends its role. Here, and the master node takes over and performs the work after that.

A

First of all, I've looked at the cluster body, and I can get it standing on the 4th floor. Then, I'm, going to use the crate cluster option and from now on, I'm going to make a str cluster.

A

Cluster Parallel flight So, I use a type of v.23 called Kind.. Cluster Club does not do that. And. What are components, machines,, providers, and ocean files?.

A

Uh Junior there's a file called 2.5 kube capo I like the option to leave a comp, p, post, Then, I, say Prepare Len, bootstrap cluster and now with Kind I will install Cuba, Tess So I said cayenne. Pluster You can check that the cluster api is now being installed with the command.

A

Then, in the upper left window on the right of this jig, I now sent a command to monitor the dog and teeth. It.

A

Will be shown in. I'm.

A

Still doing pre-pairing.

A

4 Something is now printed.

A

Only the vm was created,, but when the bootstrap cluster is all up now,, you can check that the enemy vm has been added. The.

A

Running ball is in a state, and when this is all running,, the bootstrap cluster will now start communicating with the off stat.

A

I'll show you.

A

Yes, so when I'm in the running, state, I've seen the log. So,. If you look at the bottom of the capo log,, the air conditioner's cluster flight success play comes out and it's crating.

A

So,, looking at the off screen again,, you can see that it is already being created in the area called cluster control, plane, zero. Then,, a pro tip, was also added. You.

A

Can see the log of how it is booting on Eve. So, if you hit love, it just boots through the first map, cloud, Dew.

A

So, finally, I'm just getting the image pool. So,. It will take some time to receive the image,, but after that time, it will now change to Oz Garrett. Across the stand girl, I'm waiting for the process. So on the top,. The monitoring screen appeared again on the Ride cluster side. Now,, since everything is not set up,. The main log.

A

Appears like this.

A

Then now the point has already been activated: Then p Voting cluster is written at the very end, Then badly I climbed the cluster master and brought it up again. So. Let's move the red bean in the bootstrap pluster So. Now the bootstrap cluster now plays its role So,. Now you.

A

Can see the terminator's lobe on the top left, and the newly generated screen on the top right. Uh uh,? It was amicable. So,. If you change the Cube computer and send the log, yes,.

A

You can check that it is now checking the machine status.

A

It's all over,, so from now on, I'll leave a message: saying, "Use it as Yo Cube Compaq,", and it's.

A

Over. Now, finally, I'll add the worker load, yesterday. I'm.

A

Not using the actual cluster,, but since the master node is already prepared,. I fly right after the cube, body.

A

Look at the node information.

A

Well, next time, I'm, going to blow machine deployment bunt with cube cpl cube, 10. I 'm, going to accrue. Yes, I have replica information First! Yes, Now set the 1 a la flipper replica value to 1 and run 1 war. Color, then now change it to the world and show you how to increase the two again. I'm going to.

A

Lose. I'm going to fly with a cube, body. Hey,. What files are being treated.

A

Then,, the logo of the capo controller's Zec Rating Machine, is now running, and you.

A

Can see that the vm corresponding to the 2 worker load is created on the opus tab, screen. If, you look at it now,. You can check that the fisherman Yemi has been created, called the taco cluster md material.

A

So,. If you look at the log of this vm now,, you can now try to destroy the running screen by doing a cloud initiate a cube cube.

A

Tess Now,, you can see that the Calico union on the screen, I'm monitoring, is running.

A

A

If he is now all of them and is running,, then the silver user will be in an environment where he can upload his pod of the Walker Rose. You.

A

Can check additional things. So right now,, since it's a cd player, comment, I 'll, edit this, and I'll change the replicas from earlier to March for this.

A

So,, if you look at the 5% picture,, you.

A

Can see that one more cluster nd park, cluster mg node has been added, and now a third worker note is starting to be added. Sequentially. Now,, as you saw earlier, as the first node was added, q. Flash is started internally, and now q, black and q, black start, and as the 24 starts,, the calocutt pato runs.

A

Now, you can see that it is very easy to add. I'll.

A

Show you, the stepmother, here, and now I'll organize something.

B

And also, because it supports various types of thin, as well as open stack with various providers,. Its reputation is very wide.. It.

B

's a process,, so you can use the control cushion at any time, and you can try it a lot. And, we'll show you what we're doing at sk telecom for a moment, and we're now.

B

Using the Kubundis cluster api for multi-site bare metal to install clusters. So, in the cluster called Central Eyes Comprol Control Plane, as an actor, Metallo installs, an auburn cluster in a lot of regions and.

B

Puts an open stack on top of it. You can configure the netis cluster again. So uh,. In order to go to the process of using the cluster api,, we are currently working on the r version two,. So next year, we will share with you and receive opinions so that the project can be developed.. I wish I had a lot of reference links,. So here's the cluster apr c-level link here and now, sk telecom also manages this project as an open source,. So you can see it as open source if you click on the other 100.

B

B

It was a presentation. Thank you.