Get started REPORT ｜ today, talk about the microservice architecture currently used by Tencent

Since last year 10 After Sogou was officially merged by Tencent in January , I've always wanted to tell you about some of the technology stacks currently being developed and used inside Tencent , I think it has high learning value for students . But there are clear regulations within the company , It is not allowed to share and publish unpublished information privately . Once found , Dismissal of high voltage line ！ So I haven't done it for a long time .

Recently, I was in Tencent TechoDay During the technology open day, I learned about the core Polaris of microservices being used internally （Polaris-Mesh） And one-stop microservice development framework Spring Cloud Tencent Have been opened to the outside world .

The north star Github：https://github.com/polarismesh

Spring Cloud Tencent：https://github.com/Tencent/spring-cloud-tencent

Then I can combine them today to give you a brief talk about Tencent's internal micro service system .

One 、 Why embrace microservices

I am a 2011 Entered Tencent in , The technology stack used at that time was mainly LAMP（Linux + Apache + MySQL + PHP）. At that time, another popular technology stack in the industry was MVC（Spring + iBatis/Hibernate + Tomcat）. Whether it's LAMP still MVC, They are all designed for single entities to share the architecture . The developed services are also deployed on physical servers , Generally, it is deployed by the operation and maintenance .

stay web The era of web pages , The above two sets of technology stacks have been really popular for a long time . But with the explosion of mobile Internet , There are more and more services to be developed , Functions are becoming more and more complex , Team developers are also expanding , More and more frequently . The traditional single service architecture gradually presents these problems .

Inefficient Deployment Services ： Traditional single services are deployed directly on physical machines or virtual machines , The software that the service operation depends on must be manually installed and configured by the operation and maintenance department .

Machine resources are difficult to reuse ： Modern server configurations are very high , But few single services to run can put CPU、 Memory and other resources are fully utilized , There will be a lot of idle . But in order to ensure the isolation between services , They often dare not deploy multiple services they don't want to do together .

The cost of teamwork is high   When there are multiple people in charge of the same service , It is inevitable that the submitted code will affect each other , Release also needs to be discussed . In addition, the testing work before each release also needs to be very comprehensive .

Later on , Slowly, the concept of service has been derived . Split the single application into several different services for independent deployment and release . For example, I was responsible for Sogou mobile assistant , In the background, it is divided into application recommendation lists 、 Apply check updates 、 Search for 、 Accurate recommendation 、 Cloud control 、 Address location and other independent services come out .

stay 2014 Year begins , Docker Containers became popular , The degree of virtualization is compared with that before KVM With a qualitative leap . A container image can package all the software environments that the service depends on , In the smallest case, it's only a dozen M That's it , It's very free and convenient to distribute . It can be started in seconds .

The industry has carried out a series of evolution and iteration around container technology , Gradually evolved today's Micro service architecture . The popularity of microservices is 2017 Suddenly broke out years later , Major first-line Internet companies have also introduced this technology and implemented it in their actual business , Tencent is no exception .

Two 、 Tencent's microservice engine

Early on, various departments within Tencent began to test water micro services to varying degrees . stay 2020 In, we integrated many previous microservice frameworks within the company , Born the present with tPRC As the core of the microservice framework . Now most of the services and interfaces have almost switched to the container based microservice architecture .

Tencent is also good , The industry . Although the microservice architecture will differ in the details of technology selection . But basically, it can be represented by the following microservice architecture diagram .

In this microservice architecture , The left and right sides are the infrastructure of microservices , In the middle is the development framework . Among these basic components , I think the most important and core is the one on the left of the above figure gateway 、 Name Service Polaris 、 And the configuration center . This is also Tencent's external cloud micro service engine （Tencent Cloud Service Engine,TSE） The ability to open to the outside world .

So I'll talk about these three next .

2.1 Cloud gateway

In the era of monomer application , The practice in the access layer is generally to apply for an Internet domain name , Then in the access layer, the request is distributed to different business servers according to the domain name .

In the microservices architecture , Gateway is still the entrance of traffic . But in the era of micro Services , The number of services is much more than before . It is no longer possible to directly configure one domain name and one service , And in the microservice scenario, routing 、 The demand for monitoring, configuration management and other aspects is stronger .

Therefore, a microservice gateway is needed , As all API The main entrance of the interface . The gateway encapsulates the back-end microservices , adopt API The gateway exposes services uniformly . Provide unified security in the gateway 、 route 、 Flow control 、 Monitoring and other services . When traffic arrives , After some preprocessing, the gateway can distribute traffic according to different service names .

2.2 Service Governance Center

Tencent's service governance component is called Polaris . The project was originally a unified name service component jointly built by Tencent's internal open source collaboration , For resolution RPC Called service registration 、 Dynamic routing 、 Load balancing and fault tolerance .

Now Polaris is not only deployed to Tencent cloud to provide external services , And also settled in the open source community . The project has been widely concerned by the industry since it was launched .

Github Address ：https://github.com/polarismesh

The project is not only to open source all the source code , Moreover, detailed usage documents are configured .

Next, let me introduce this open source project Polaris Mesh Problems solved and capabilities provided .

In the microservice Architecture , The first problem to be solved is how to find the service you want to call . In a big company , There may be thousands of services running online at the same time . So when your service wants to call a service , How to correctly find the corresponding service and access it ？

This problem is through Polaris Polaris Mesh Service registration and discovery mechanism . The service provider will send Polaris Mesh Register your service name and address . When the caller needs to call, it is embedded in the process Proxy Agent first Polaris Mesh Get the service address to call , After finding it, directly launch the call .

Generally, service providers have multiple instances , So which one should I choose to visit ？ There are two different usage scenarios .

One use case is grouping , For example, your service callers and service providers may be deployed in multiple places , stay Linux Network performance 15 Two optimization suggestions ！ In the article, we mentioned that it is best to call services in the same machine room . Avoid calling across machine rooms . Because it will take too long . Another example is that you may need to deploy the test environment separately , And grayscale environment .

Another application scenario is load balancing , Even in the same environment and machine room , It is possible that each service provider's CPU And other configurations are different , It is also necessary to reasonably schedule the request traffic .

These two practical application scenarios need to be solved through reasonable routing rules .

Let's talk about current limit . In the responsiveness of services, there is a word called avalanche . Avalanche refers to a natural phenomenon , There is a lot of snow on the snow mountain , But they are stable . But when the last snowflake fell , The whole snow mountain collapsed . There is a similar phenomenon in services , If your service can be handled normally every second 1000 A request , But if it comes 1001 A request , The whole service will crash , even 1 A request can't be handled normally . This is the avalanche in service .

In order to eliminate this phenomenon in service management , A current limiting function was born . You can configure the maximum processing capacity of a service instance , When this processing capacity is exceeded, no new requests will come later , To ensure the normal operation of the service . There are two ways of current limiting , One is rapid failure , Refuse the extra request directly . Second, line up evenly , Let the service handle it slowly . It's like the flow restriction bar in the subway station .

These two current limiting capacities are Polaris Mesh Can be easily configured in .

In addition to the above abilities ,Polaris Mesh It also includes circuit breaker rules 、 Access authentication 、 Observable and so on . In terms of access mode, it not only provides Java / Go / C++ / Node.js / PHP Of SDK Access , And support Spring Cloud / gRPC 、 Service Grid 、K8s Service governance and other mainstream clients . The complete access method is shown in the figure below ：

On the use , I think the configuration tools provided by Tencent cloud are more convenient than those we use internally . Routing rules 、 Fusing rules 、 Current limiting rules can be configured directly on the interface . It's very convenient to use .

2.3 Configuration center

In the era of traditional single service , Configuration is often written in a thing called configuration file . But in the era of microservices , Due to the explosion in the number of services , A lot of configuration items , All kinds of irregular modification requirements will make the whole work process seem chaotic . Too scattered configuration files increase the difficulty of management .

So in the era of micro Services , Configuration centers have become standard . The idea is to put all the parameters in the project 、 Configuration information such as switches , All in a centralized place for unified management , And provide a set of standard interfaces . When a service needs to get configuration , Call the interface to pull . When there is a configuration update , The configuration center will notify each service period to load dynamically .

In the industry , There are many common solutions , Include Zookeeper、Eureka、Nacos、Consul、Apollo wait . Tencent's cloud TSE All of the above mainstream configuration center solutions support , And the function is enhanced , Provide a visual console 、 Service management 、 journal 、 monitor 、 The alarm 、 Authentication and other functions that the original scheme does not have .

On deployment , You no longer need your own operation and maintenance team to deploy . One click deployment supports multiple activities 、 Highly available configuration center clusters such as persistence .

In the visual console , It is very convenient to manage the data of the configuration center 、 Log view 、 Operation monitoring and other daily operations .

The north star Github Address ：https://github.com/polarismesh

2.4 Development framework

about Java For developers , It can also be used directly Spring Cloud Tencet Development framework . The framework implements Spring Cloud Standard microservices SPI, It relies on Polaris polaris, Implement various distributed micro service scenarios .

Spring Cloud Tencent All components have been uploaded to Maven The central warehouse , Just introduce dependencies .

Spring Cloud Tencent：https://github.com/Tencent/spring-cloud-tencent/

3、 ... and 、 summary

Microservice architecture is a popular technology . Building a set of microservice infrastructure often requires assembling a series of microservice middleware , For example, service registry 、 Configuration center 、 Log center, etc . The initial cost investment is very high . Generally, large companies have special human and financial resources to build these basic wheels needed by microservices .

For small and medium-sized companies , Even for individual developers , If all from 0 Deploy these basic components that microservices depend on , Deployment alone is enough to toss for a long time . Moreover, there are many technology options in microservices , For example, the configuration center has Zookeeper、Eureka、Nacos、Consul、Apollo And so on . The difficulty of learning and operation and maintenance is much more complex than the traditional development mode .

But as a small and medium-sized company or individual developer , If you don't touch the microservice architecture, it will slowly be out of touch with the times . Fortunately, now public cloud manufacturers have provided microservice infrastructure . I am here Techo Day The feeling experienced on , It's just too convenient , Open the box .

Recall that I just started to toss Zookeeper、Consul When , Compilation and installation alone waste a lot of time . Now one click deployment can be used , It's really convenient . And I feel that Tencent's external product experience is better polished , In fact, many places are easier to use than our internal tools .