Cloud Native Computing Foundation KubeCon + CloudNativeCon North America 2022, 11 Nov 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Production Practice For Large-Scale Financial Application Platform In...- Jiahang Xu & Jianbo Sun

Description

Don’t miss out! Join us at our upcoming event: KubeCon + CloudNativeCon Europe in Amsterdam, The Netherlands from April 17-21, 2023. Learn more at https://kubecon.io The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

Production Practice For Large-Scale Financial Application Platform In China Merchants Bank - Jiahang Xu, China Merchants Bank & Jianbo Sun, Alibaba Cloud

Speakers: Jianbo Sun, Jiahang Xu
China Merchants Bank(CMB) is one of the biggest bank in China which adopts cloud native technology for its rapidly evolving business needs. The challenges we meet are quite different from the practices of the internet industry, banking demand higher stability and security, at the same time, more complex historical architecture. For example, we need to handle the migration for diverse workloads from bare metal to serverless to provide a consistent experience. We must provide progressive rollout with traffic management to keep the stability without breaking any business continuity. We need to have insight for the application delivery and management process to meet the business SLO by observability, further more, to reduce the cost of resources and improve utilization. In this talk, we'll share our practices on building a modern banking cloud native platform, that mainly leverages CNCF projects such as KubeVela, KubeVirt, Envoy, Opentelemetry and others, to serve diverse workloads and solve all the above chanllenges.

A

Hi everyone um from China, Merchants, Bank I, will give a brief introduction for myself. I have a 13 years software research and development experience across Telecom Automotive Financial industry. Recently, here I focused on quality of resourcing knowledge and take some in practice. Currently I'm responsible for both Cloud native application plan for in China Merchants, Bank I'm glad to have a chance to share our Enterprise practice experience six. So, let's start okay.

A

Today's topic is about the production practice for large-scale financial application platform in China mattress bag, and this step I will appear with these Jim bosson. A staff engineer. Alibaba clouds is by also is my friends. um Most of us are the cool Villa maintainer.

A

Today's agendas includes four parts backgrounds. Practice in CMB zombie is a short name of my company in a Villa Cuba Villa in central feature, introduction and we'll show a demo about the feature. The first two parts I will charge the part three and part four uh jumbo will take on charge. Okay, let me give a brief introduction about China Merchants Bank Channel Merchants Bank is a commercial best bank ranking as 14th of the World Bank. In last year's 2021, uh there are two merger, um mobile apps, App Store, a mobile here check and CMB life.

A

Both over apps has large skill customers globally until end of June mobile Bank, app has already 1178 million customer Global 7B live has 100 um 32 million. Also so uh China Merchants Bank has developed a reputation for being a technology folks, banking, globally, where I'll wait to Cloud native.

A

We really change in application domain. um In recent year we has already established a large-scale private cloud in our organization from the high level private Cloud architecture diagram. We can notice. There are Apple wider virtual machine, kubernetes serverless.

A

Hadoop and some decades runtimes and huge infrastructure in the we provide diverged foreclosed and the middleware service in corresponding domain, a corresponding on time.

A

Applications couples with runtime and infrastructure. It's really complex and challenge for us. Devils cannot be agile and cannot be high efficiency.

A

So both application engineer and the plan for engineer have some some issue to be resolved from application engineer size.

A

They want to decouple application from wrong time or infrastructure, and also one application and support Academy delivery to make the penis stable ability and the security and one application observability for maintenance and the plan for engineer site. How can we solve them by Cloud native ecosystem at first? We think we should it's important most important to decouple application from runtime and influence, so we follow the application open or application model in the OEM implementation. Cuba Villa wait design our application model we divided and Conquer, and abstract and castibly encapsulates our application.

A

Workloads use case for the components: definitions domain with the Define each workloads. uh What component definition each one mapping unique application use case, for example, front end for static sites. Web storage for HTTP API server vmso is for virtual machine service task for scheduled job API Gateway, my circle of link and the more this kind of workloads. The usage is very obvious as the name and secondly, we Define many components and trees. Definitions are part of OEM specs, their application, oriented Concepts, it's easy to understand and use application engineer will be happy.

A

Of course, 10 4 Engineers should provide each definition by Q templates with scuba Villa.

A

We can achieve concern Separation by OEM and Cooper villain concerns separation to the couple application from runtime, since we can Define our application model in the describe. It declare declare it so we can extend and Define many useful feature in this way. The same way now we can Define calorie calorie delivery as a treat. The left is a example. There is a four components: APM.

A

And Trace dependency in the in one Sprint front end and web service are related.

A

Comments should be released end to end. At the same time, the issues, integration and release end to end are once those front-end and web service components, version 1 to version two.

A

So the step two as a most important step for Traffic Control calorie validates with a damage traffic step and the user can config zero percent to 100 gradually to calorie instance, showing in this school diagram for validation and in this step. There are three reports very important, sweet, three point in this step: we need shoes. We need to choose first, jump components as the entry to tag the traffic traffic for.

A

Control, moreover, it is less history during root text traffic only to calorie instance, version 2., especially both only from front end and web service, can be infected. Other components in these Trace cannot be effective affected.

A

Now we can validate both Canary version, 2 instance with tech to traffic. It end to end in step. 3 is we have a two choice? If the calorie version is qualified, app calorie become can become to Baseline version, two version, one instance will be removed if the calorie version not qualified one hundred percent, the traffic will come back to version. One carry version, 2 instance will be removed, base 9 no version change.

A

We can in the right in the right. There are two Snapshot illustrations. We can implement the feature with Cuba Villa workflow engine and the calorie routes as streets definition plug-in and void then make traffic control there. So the the the there is a uh the file yaml file that is describe and declare the workflow and the calorie routes. Trade. Illustration: okay,.

A

um In the same way, we can also integrate f observability and the SOS trade, as the diagram shows their demo demo app the same name app. They will generate log Trace metrics of the APB, so.

A

This Observer observability data will be encoded as Hotel and connect it to time, Siri database change, database and colon storage for allies and the dashboard. If one queue can choose connected instance and connection rules, we can Define metric tricks applied to corresponding components. Components will be observed with open Telemetry f1q support SRE for components. We can Define also Define, also Define SL ice plugin to integrate open so specs.

A

We can config service level objectives, SLI metric alert policy, there are metric, and so it's simple in the in the in the in the in the file, the snapshot.

A

Okay, I just choose some egg, typically practice, or example, to prove the Cuba Villa can be as the application uniform control plan. In Enterprise we trying to use Google to build Cloud native application platform to achieve this, achieve it and look at the diagram. The platform is add-on-based. Architecture includes three main parts: a workload layer, three layer and OEM engine I work, late workload, a layer includes all components, definitions and controllers.

A

Trade layer includes all Trace definition and controllers, and the Villa has the OM engine. So we can see all type attacked by around. We have the the yes the, especially in the in the trades layer. We have support in the ax meta, F, config, app release, approach, uproot F, observed F dependency, F calorie and F elastic, and we can support more treats okay with application uniform control plan. We can achieve concerns separation, separation to reduce complexity with transparent runtime. Yes, that's wrong time.

A

So my part already finished so next jumbo will introduce Kubo, Villa important feature and progress in the demo line.

B

Okay, thanks to jahang, he has introduced the large-scale practices in China materials bank and now I will introduce some kubella Essentials. That I will give you a quick demo at the last. My name is Tiempo I come from Alibaba Cloud I am the maintenance of kubavela. The most highlight of cubavera is extensibility. The expensibility follows om model. The end user just needs to write the component type. It will reference to a component definition reduce choice, and the registry follows the oam model.

B

The definition has a consistent API interface and it can be platform agnostic that it can be implemented by different projects. For example, if we want to deploy a Helm chart, we can both use ago CD or flag CD uh to implement it and they are the capability providers and users can glue definitions and the projects into an artifact. The effect in covala world is called add-ons.

B

We have built lots of out-of-box capabilities and we have Fable equivalent control plan and it's a core part of kubavel and the lots of add-ons, including Flag City, Fargo City and the Veda ux. It's a UI console of kubavela. Also, we have a cross-plan terraform, Integrations, open class management for multi-class deployment and opening up for iot Edge and more than 50 add-ons. So we have a lightweight installation called valid. You don't need to rely on any kubernetes clusters for control plan. Now I will give you a quick demo about how it works.

B

You can find the scripts of all the demo in the link virtual machine in My. Demo is Ubuntu installed, you will get Verity here and we will install the Covelo control plan and finds a public IP for demo.

B

Now it's installing. You can also install kubela on any existing kubernetes clusters. You just need to install it with helm. Okay is a folder of my demo, okay, it started now and actually uh the vanity leverages k3s as a kubernetes cluster. So we will also have a kubernetes config here we can read it and you can also copy it in your local machine for convenience to export to config.

B

We can also install some add-ons. For example, we want to see the UI console of value. X and Isabella D provides the air gap installation, so it can install very fast. So we use a command here, so all the images have already installed locally. Okay. Now we can visit it. It's this dashboard and the initial password is- and let's use it, change the password email.

B

Okay, now we have the Google dashboard. It's have two languages supported. This is one application installed. The value, ux add-on is also one of covela applications. We can see the components and the underlying resources. We also have the service endpoint. Let's see all cover, ecosystem capabilities are built as add-ons now we already have more than 50 add-ons here. The most interesting add-ons are the observing add-ons, let's in enable them. This is a promise use add-on.

B

Okay, I have enabled it. We can check the status here and it will bring up Prometheus instance, let's enable other add-ons, such as nude exporter, and let's also enable Coupe, State, Matrix and export, and we also enable the graph definition. It provides us a crd operations on graph Now API. This is very important capability that we can make a graphene dashboard as code and in Cuba. This atom brings up grafner server, let's change it as new report and enable it okay. Now we have several add-on enabled.

B

Okay, it's already running. Let's check the graph now dashboard here. Password is Provena.

B

Here are four dashboards that have been added automatically.

B

One is the covella system metrics. It provides a very detailed, equivalent system state, so we can check the system health status and it also provides what component we have. What trait we have and other what workflow steps here. Okay and another important dashboard I want to show you is Google application panel. It gave application details and you can see what components in one application and what what are the back end resources and you can also check the details and we also have some kubernetes API server. Matrix covella has lots of component types.

B

When you install some add-ons, it will also.

C

Provide the component types and I will show you the web service era. Web.

C

As you can see,.

B

The web service add-on has several parameters you can deploy it here or use command line, use command line. You can use Vera, show web service.

B

To learn what are these arguments.

B

You can also use Bella, show web service.

B

We can also see some examples of how to use it here are the parameters and details. I'll show you how to build your own definitions to extend the capability of kubavela and now, let's start to create a definition, and here is a java definition and we just want to deploy what the using Tomcat. So it's a very basic definition. We just run two objects, one is a kubernetes deployment and another one is community service and we have defined some arguments. Let's deploy it relative apply.

B

Law definition.

B

For just now we don't have this component type. We don't have this one, so we can refresh the website. Then we can see the Java component type here and if we, if we thank you look, we can deploy somewhat. For example, we use we use a single bar here.

B

Deploy it great and deploy it, you can also deploy the this application. Yaml.

B

Now it's creating and you can see the logs.

B

You can also see the instance status.

B

You can also see the topology okay, it's running now we can visit and it has some suffix simple.

B

Okay, we have succeed.

B

Okay, we have succeeded in show you uh how to define a definition and then, let's, let's build an add-on, uh I will choose, the click calls add-on and I was just find a click house operator.

B

So I will use the clickhouse operator. It was the first time I saw the clickhouse operator. I just got the installation step that is yamo here. Yes, they all install in one yamo. So if we want to build an add-ons, the easiest way is to copy the yamo bundle here and use a ref objects of Cuba, and we also add some trade to scrap the Prometheus Matrix. That we also provide this uh dashboard is a dashboard.

B

I found is just an image, so we use web service to deploy the image and provide some service count to give the capabilities to access the Community Resources, and we also I also find it provides some grafner dashboard. When a user deploy the clickhouse operator, it's very hard to integrate the grafner, dashboard and Promises, while Covelo has a powerful capability to do them, we just need to copy the dashboard Json here then we can install this add-on.

B

We can install the add-on by Bella add-on, enable this folder. This folder is clickhouse error. Oh no, sorry, let's enable it with service type nude port. So we can check the click house.

B

Okay, it's installing.

B

We also add one component type called click house, so we can deploy apps with click Hall stability, it's actually using the clickhouse installation. Crt kubela also provides the capabilities that you can Define what the resources behind the clickhouse crd, okay, we have already installed it. The the dashboard is here. We can see the dashboard it's provided from the clickhouse operator and we also able to ins, create house component with several parameters. Let's create one clickhouse instance: choose clickhouse.

B

Bind to default environments install it and with new the port, okay, create it.

B

Deploys covella has the power to check the underlying State, faucet and port. How could cover do that in the add-on? We can also Define the resource topology. Here we can Define the.

B

The topology is the click house. Installation has some child resources, such as state faucet, service resources,.

B

Okay, we have finished, deploy, The Cliff House application and we can check it here. So the big house has a suffix play. Okay, now we have deployed small. We have grafner dashboard of Treehouse. Let's go back to the grafner dashboard.

B

The magic power is here, we have automatically installed the clickhouse operator dashboard.

B

The dashboard is provided by the clickhouse operator Community. We just copied that and automatically glue this dashboard here thanks everyone, we have just finished our demo, because time is limited. We still have many powerful Covenant features that didn't show you, such as the programmable workflow multi-clustered application, delivery and management.

B

At Last I want to share the recent roadmap of chlorella to you, but we want to make our end users to have many more choices to use Google such as kubernetes custom resources, API or command line, or UI console and along with guitars with more powerful observability. But we will enhance the Vera core to make it more extensible easy to use and flexible again. We will enhance the security and the sustainability.

B

We also will enhance our add-on ecosystem with higher quality add-ons and easy to build add-on toolkits to make your own add-ons more easy, and we will also build some interesting add-ons to make covala ecosystem has more powerful features and scenarios. We hope all of you can join the kubella community to experience the powerful features of rubella thanks. Everyone hope to see you in the covella community call bye.