Cloud Native Computing Foundation KubeCon + CloudNativeCon China 2018 (Shanghai), 21 Apr 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Cluster API Deep Dive With a Tencent Case Study - Feng Min, Google & Zhiguo Hong, Tencent

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

Cluster API Deep Dive With a Tencent Case Study - Feng Min, Google & Zhiguo Hong, Tencent

The Cluster API is a Kubernetes project to bring declarative, Kubernetes-style APIs to cluster management. It provides optional, additive functionality on top of core Kubernetes. It tries to address the fragmented infrastructure layers in the current stack. In this joint talk, Feng Min will go over the state of infrastructure today, and see how cluster API fit into the picture. Feng will also look at the structure of cluster api for things like Machine Set/Machine Deployment, and check out how they can help out the common cluster operation like upgrade. As a case study, Zhiguo Hong will talk how tencent cloud support get integrated, and how cluster-api can be applied to manage k8s cluster in tencent cloud. The audience will walk away with a clear understanding of cluster-api architecture, and how current cluster management can be transformed into cluster-api world.

To learn more click here: https://sched.co/FuJg

A

A

A

To create to care to them manually.

A

Reduce working cost.

A

B

B

A

What the for maintenance.

B

A

Another like the.

A

Special commands resource.

A

Do not have such process.

A

We use a series of.

C

A

To build the cluster management architecture.

A

A

Including the class data.

C

Self-Assurance.

A

At mission deployment.

A

Can help others to achieve the common.

A

We have three three types.

A

B

A

Specific configuration.

A

Related information.

B

A

That's a control.

B

A

Well, maybe the control plan is only for master.

A

Degree, it is not, but also Antonio young controller.

A

C

A

Series of tools like.

B

A

Mention that the clustered control.

C

A

The constitutionally all for the to do them poster strategy. The demonstration of other functions.

B

Resources like a Martian.

A

Or machine resources.

A

Reconsideration, this kind.

C

Of resources, they.

A

Have to state the controller will choose which.

B

A

Is suitable, the.

B

Reckon system is.

A

B

Say the bootstrapping.

A

B

A

Machine what the changes had made to the pastor, if we.

D

When this platform's realized the describes interfaces quite easy to realize as well, so the decision will be calling into the GCE adaptor and we'll use some GC api's to.

C

D

C

D

That was about class API structure architecture. What is it for and next I'm going to explain with an example? Why is an API or is it effective.

D

C

D

Very common occasion.

C

D

C

D

Their own realization for the upgrades of clusters, so let me use one example: if we use class API, all of the relay realization will be unified. You just need one realization in this example. If our cluster has three notes, we have machine deployment. There are three notes and the version is V. 1 point 8 point 3.

C

D

Want to upgrade it I want to upgrade it to 1.9.3.

D

Of course, you can also use a utility to operate it, but it can be as simple as mo file just need to add the yellow file and just to modify it as 1.9.3 and apply for the Cooper upgrade and.

C

D

Under the hood so in machine deployment, this is the initial status. There's a machine set now three knots and version numbers. One point.

C

Eight point three.

D

So as step one, so when a machine deployments controller it will can create a new machine set. Of course there are zero notes, but the version is 1.9.3.

C

D

Of all a change is the version two one. So for the machine set controller realizes the replica is one so there's a change of status and then it will be creating a new machine resource and the machine controller will create such a new node of.

C

Course afterwards, the.

D

New machine set will change the nose number two and create a new nose. So it is a new machine set and for.

B

The old machine said: there's a replica as you can see, the replica is already.

D

The one is deleted and the third one third step we changed is node number two sweet, so we've got a newer version and.

C

D

The old node we only got.

C

D

And we change the replica to zero and then in this case we deleted all the old nodes and for the old machine sets, etc they can all be deleted. So, of course this is simplified process. The time and actual is not really related to the structure or the cloud platform. You can abstract the time as long as you realize the adapter, the implementation or the others can be shared. So.

C

D

Was an example of the cluster updating and next I'm going to talk about the bootstrapping of the cluster cluster.

C

D

Bootstrapping is a very interesting topic.

C

D

C

D

All nodes in the cluster even master knows we want to create that through API, you create a resource machine resource and the machine is a master, and in this case we have a saying that whether you get the chicken or the egg first so who is created first, you have cluster first and then you deploy the machine controller to the cluster or you have the machine controller running somewhere. First before you create the master nodes. So that is the question and so for plan one for the master notes. We have some special treatment which.

C

D

Have some side-channel to create the master nodes? You can use the api on the platform.

C

D

In this case, you have the old problem. There are. A lot of codes can be copied repeatedly, so you create them where you again, and secondly, we're trying to use the kubernetes to manage the kubernetes.

C

D

C

Restricted abstracted for.

D

The tool we have a local cluster- it can be any cluster, it can be mini cube, it can be any local. Cluster can be darker.

C

A monument so for the local.

D

Cluster you can deploy. You can use the controller as a part to declare a to deploy. It.

C

D

C

D

Can create two resources, including the cluster resource, a machine resource.

C

When the controller has a resource, it.

D

Will create a real cluster and in this one, for this resource, definitions, etc. You can also copy from one ethnicity to another. One.

B

D

Afterwards, after you copy all this information of these, and you can deploy the cluster API components on the new, a new.

C

D

For the old cluster, you can delete that.

C

D

The new cluster will be up and running.

C

D

Here you can see how the API is there from zero, and next one is the.

C

Deployment of the cluster, the.

D

First one is in cluster or not so it's a plan, one. We call it in clusters plan in this cluster. We put the Machine controller and cluster controller and other controllers.

C

D

Are deployed as normal part, so they are deployed inside the cluster. The benefit is the realization is easy and is easy to understand, but there will be problems with Willa reliability. For example, if note is broken and I want to replace.

C

C

D

Is not master note if it's a master note is down I need to replace it.

C

D

The parts are running on a master note when you replace the master note, there's no place.

C

For the controller so.

D

It will be not very convenient. You can also use other approaches, like H, a configuration etc, but this will make it more complex. There's another deploy deployment code.

D

C

D

Cluster so in the manager, cluster and child cluster. So for the.

C

Control panels.

D

They are deployed on a manager cluster, so they are just seen as a normal part in this case, if we are going to upgrade the control API and the API or the others, you just need to upgrade the part as normal and it doesn't come to doesn't influence the running of the child cluster as well. Since the controller is running on the manager cluster, it is also responsible for creating some resource for the child cluster.

C

D

Our project is starting from last December. We actually have made a tremendous progress for those mainstream cloud platforms like public clouds or private clouds. They actually join the support, providing the support. So we would love to this opportunity to introduce it to you and you can see if you can use it in your own project and also contribute more to the development of the whole ecosystem. And next my partner Joe core, will share with you the application of cluster API on Tenzin's.

B

C

D

Like to introduce the.

C

D

Api's support of the with the TKE and some operations.

C

D

For this one, this is a github page for custom API. On this home page introduction, there are cloud providers, realization links. This is the Tenzin cloud is a test and clouds cluster API provider, realization.

C

D

The code here is quite simple:.

D

So this one sells says: if you want to compile how you're going to run it, and there is a release here.

C

D

C

D

Are mainly yama and.

C

D

So now I'm going to use Yama's and examples in release to demonstrate to you. How do you create a cluster and add or reduce machines in cluster and delete machines and cluster afterwards.

B

D

Go back to the slides.

C

D

For using cluster API, there are two steps step one: we need a bootstrapping cluster.

C

D

Cluster api-related controller needs to be running on this cluster and second step. You can use cluster api.

C

D

Now I'm going to demonstrate how to change a regular TPE cluster to a bootstrapping cluster enabled by it.

C

D

Here, I have three steps to show you. The first is to install the cluster API CRD.

D

And you can see, there are facilities installed. Cluster machine is in state machine deployment.

C

D

One is to install secret, so let me introduce this one to you.

C

D

In the release we have, the secret llamó is incomplete.

C

D

Here we've got to fill in the secret ID and secret key of your own testing cloud account. So here we got the link for you to get your account ID and key. So here, I'm using my own accounts to filling the secret ID in key, so I directly create the secret here.

B

D

For this secret.

C

D

Is the T key is key API controller image to use a tenant API, so in the operation you will need to use a secret here and for step 3.

C

D

This generic controller and a cloud provider specific providers, you need to random.

C

D

They are both operating in a format of deploying deployment, and here I created three deployments, and one of them is generic controller, and then we got the Machine set controller and machine deployment controller and the other two is the TKE realized realization.

C

A

If it's created, let's check the status of it.

A

Vanessa the network section of this.

A

To achieve the conversion of.

A

Is creating is being creating I.

C

A

It will take one extra minute so how you, this time to introduce to introduce other things.

C

A

A wave they can manage the Tiki clusters through.

B

A

For inside of the tension cloud, since we management over 2,000 clients.

B

A

How can we manage that? Actually, we use the cluster API well.

B

A

Bit different, it's slightly different from the community version different in the format, but the basic process.

A

Ideas remain the same.

C

A

Moment you before, when using the cluster API to manage the cluster.

C

A

The original client information in distributed.

B

The database now.

A

With cluster it behind, we can store this kind of information.

D

A

Api server or a TCP here.

C

A

Way can bring the following advantages:.

C

A

It is a declarative, API.

C

So it is good for.

A

Diagnosis and the family recovery- and it has clear, coracle, logical.

A

Besides, it can reuse the pattern or achievement or original code from the community process, a practice.

B

A

Back to let's check the status of this node reloaded.

A

Here we can see the research note how's. The cloud is created.

A

Let's do another operation I.

C

A

Delete this kind of ocean deployment and it automatically.

A

Deletes at the hosta cloud.

A

About to the management panel reload, it.

A

Operation or delete the whole cluster.

A

And then this resource is deleted and we found the cluster.