Archsummit: evolution of the underlying framework of cherished microservices

1、 Preface

This year's ArchSummit Theme “ Architecture upgrading under digital transformation ”, Main focus ： Cloud native 、 Research efficiency improvement 、IoT  System architecture 、 Microservice architecture 、 Low code system 、 Sea going business structure 、 Artificial intelligence and machine learning 、 Enterprise digital transformation 、 front end  Serverless  R & D system 、 Digital transformation in the financial sector 、 Big data practice and Application .

The author is engaged in the Internet industry 4 In more than , This is the second time to participate in ArchSummit Global architects Summit , The main focus is “ Microservice architecture ” The technology sharing of all teachers under the module , Here for zhenai CTO Teacher pengwanliang 《 Cherish the evolution of the underlying framework of micro Services 》 Make some summary , Welcome to join us for additional exchanges ~~

2、 About Instructor

Mr. Peng is a very senior technician , In the Internet industry, there are close 20 Years of experience , At present, I am working as CTO.

3、 Cherish the chronicle of microservice framework

Evolution of zhenai's microservice framework , It is the result of gradual evolution ：

• 2018 In, I experienced the splitting of business micro Services ： Decouple business and solve complexity problems

• 2019 Completed the transformation of the data layer ： Solve the pain points of business data layer

• 2020 Container deployment and DevOps pipeline Design ： Improve the system's scalable and highly available cloud native capabilities

• 2021 Realize more business in different places （ Double cloud and double life ）： It improves the system disaster tolerance ability and service security

4、 Cherish the evolution of microservice framework

4.1  Facing problems

Teacher Peng introduced , since 2018 In, the business scale of zhenai.com exploded , There are four problems in the R & D team ：

4.2  Breakthrough direction

So their team began to focus on building infrastructure , And make concentrated breakthroughs in four dimensions , namely ： Service Architecture 、 Basic components 、 R & D process 、 R & D specifications ;

Cherish for as long as 4 On the way of architecture evolution in , The development team has experienced micro service splitting 、 Link optimization 、 On the cloud 、 Containerization 、 Double cloud and double life , And have corresponding thinking in each link , Exposing problems repeatedly 、 After solving the problem , Finally, a set of cherished and characteristic microservice governance architecture is precipitated , That is to form the final set of underlying framework , This framework has the following characteristics ：

• Integrating open source frameworks , Combined with business secondary development

• Build a set of technical specifications

• Unified technology base

4.3  Evolution process

4.3.1  Micro service splitting

4.3.1.1  Business split

Although the business of each enterprise is very different , But the methodology of micro service splitting is roughly the same , So teacher Peng under this topic , Not too much expansion ; But from the introduction PPT in , We can still clearly understand that zhenai's idea of splitting micro services is , Cut according to the business model , And vertically split into multiple microservices , It includes ：unify-pay、core-user、verify wait ;

4.3.1.2  Microservice framework

Precious micro service splitting , At the technical level, it has the following four characteristics ：

Mainstream Java Commercial projects , with SpringCloud Open source framework as the underlying support , The cherished microservice framework is ingenious SpringCloud and Dubbo The two frameworks are integrated together and used , This should be a highlight of cherish framework .

4.3.1.3 DDD With cost reduction and efficiency increase

Besides , Teacher Peng also mentioned a key point of technology implementation ：

be based on DDD Business split of domain model , On deployment , The original api Layer and the  provider Layers are merged on one node , It can increase the call overhead and save hardware resources , To realize “ Authors efficiency ”.

So how does treasure work ？

By introducing DDD Domain model , Business API Reduced coupling ; This involves vertical splitting （ Design from bottom to top ）, Realization of each business area   Code organization in the way of layered subcontracting , Structure is clearer , More reasonable .

• adapter： The external interface implementation of each field , The interface implementation of each business at each layer is different （OpenApi、 client 、web End etc. ）

• application： Business mix

• domain： Domain layer

• infrastructure： The support of the most basic services

2022 year , One of the most popular topics of Internet enterprises is -- Cost reduction and efficiency increase , And cherish the landing practice of the R & D team , It just provides the industry with a research and development idea ： In the DDD Domain split , Combined with business granularity to achieve the ultimate , It can not only make the boundaries of business areas clearer , It can realize cost reduction and efficiency increase by optimizing the deployment , Reduce server maintenance costs .

4.3.2  Microservice gateway layer transformation

The transformation of zhenai microservice gateway layer is a very unique project .

• One side , It's based on SpringCould-Gateway Open source framework to expand , The ability of gateway layer foundation is realized ： Request validation 、 Black and white list 、 be based on sentinel Fusing and current limiting （ Secondary development ）;

• On the other hand , Treasure gateway provides a lot of capability support for the business side , such as ： Merge request 、 Retry optimization （Connection abnormal , Services that are not idempotent cannot be solved ）、 Elegant shutdown 、 Start the warm-up 、 Access log reporting .

4.3.2.1  Merge request

gateway MergeFilter Without changing the original interface , Allow clients to compose requests , Merge multiple requests into one , Its purpose is to reduce the problem of too many requests during page initialization .

Its functions are as follows :

• Split request : Resolve the custom agreement , Split the merge request into actual request groups .

• Parallel execution : Execute the actual request group in parallel , Set listening return .

• Merge response : rewrite response, Respond with the downstream service of the merge request group .

• Demote and retry : Configure timeout and abnormal degradation , And retry the failed request .

• Request monitoring : Monitor request groups with large time differences , To optimize the layout .

4.3.3  Data layer improvements

4.3.3.1  Facing problems

The most vulnerable bottleneck of complex business system is database , Again ,DB It is the most critical link to cherish the stability and performance of business systems , Therefore, cherish the cloud with the help of business , Yes DB Launched various optimization .

4.3.3.2  Data splitting

First step , Sub database and sub table ： The team follows microservices “ One application, one database ” Principles , To the original DB To break up .

It was not done before because of the high cost of database management , After the business goes to the cloud, cloud database is used , The improvement of infrastructure capacity , Help to quickly achieve the target architecture .

4.3.3.3  Remove the data middle layer

The second step , Remove the middleware Mycat, Apply direct database ;zhenai-framework（ Cherish the underlying framework ） Manage public configuration : Database and table algorithm 、id Generate rules 、 Routing tools, etc .

The overall structure is as follows ：

1） Sub database and sub table implementation

Cherish the bottom through  sharding-jdbc  and  sharding-proxy  The introduction of two components , It realizes the ability to divide databases and tables after going to the cloud .

The key to the above step is ：

sharding-jdbc It is to solve the problem of dividing databases and tables at the code level , You need to configure the strategy of dividing databases and tables by yourself ;

sharding-proxy Is a kind of database middleware , Solve the problem of dividing databases and tables from the database level

2） Configure isolation

Readers are prone to doubt at this step ： Why are these two components introduced at the same time ？

Teacher Peng answered this question for us ：

Because the formal environment requires non-stop splitting of large tables , At the same time, in order to DB Configuration and other high-risk configurations are managed centrally （ Used git System save ）

.

In a formal environment , System data layer and business layer configuration , There are different risk levels ;

• Business layer configuration ： Page component configuration ;

• Data layer configuration ：DB Account and password 、 Connection timeout, etc ;

3） Connection switching without shutdown

Because of the need for a formal environment, there is a requirement of non-stop , So we need to consider “ Implementation of database connection switching ”, Here, teacher Peng also carried out detailed steps to split ：

Particular attention ：

here , You need to think about your business scenarios , Because treasure is not a big order transaction, this scenario with a huge number of connections , So I finally chose to use this double connection .

4.4  Framework capability upgrade

Treasure continues to upgrade in the following four aspects ：

• Link level fuse degradation

• single IP Fuse

• Graceful shutdown of the whole link

• Start the warm-up

4.4.1  Fusing the drop

Service monitoring needs to be improved , Otherwise, service degradation cannot expose problems , Because in the upper business ,HTTP The corresponding is normal , But there are exceptions at the bottom .

4.4.2  Graceful shutdown of the whole link

Precious full link graceful shutdown , On the whole, it is divided into 3 A step ：

• Service offline

• The rest of the business logic is processed

• Destroy resources （ Thread pool 、 Scheduling tasks 、 Database connection 、 There are already requests for processing, etc ）

Mr. Peng gave us some experience in making elegant downtime ：

• Open source components have achieved a set of elegant downtime , such as Dubbo, Then this framework can reuse their implementation directly ;

• If you self encapsulate a set of precious interfaces , So you need components that shut down gracefully , All implement a certain underlying interface , Then there is a set of unified specifications to ensure that elegant downtime does not affect the business .

4.4.3  Service restart flow control

4.4.4  Bytecode enhancement - Log monitoring

Agent Use Java Byte code non intrusive acquisition http Access log ,dubbo Access log , Business log , adopt UDP Report to collector The server ; then collector The server writes data to kafka、es、HBase, Implement log aggregation .

4.4.5 Redis encapsulation

The framework provides API To the business layer , Learn from it  MyBatis Mapper Thought , When defining, first declare the storage structure and storage object type .

4.4.6 web Layer capability encapsulation

5、 Architecture empowerment and research efficiency improvement

5.1  Facing problems

Cherish the R & D team , There are also some problems ：

• Inadequate resource isolation

• Parallel demand joint commissioning / It's difficult to test

• Too many people participate in the deployment process

• Multiple environments are difficult to manage

5.2  R & D assembly line construction

5.2  Containerized plug-ins

zhenai-dockerx-plugin It is a self-developed maven plug-in unit , Make the application access transparently docker、k8s.

Its function has the following six points ：

• Check
  : Branch check 、k8s ns Check

• Hit the mirror
  : Use zhenai-dockerx-base As the foundation docker Mirror image

• Structure layout yaml:
  The simulation template renders k8s deploy/svc Of yaml

• Deployment launch
  : call k8s Interface deployment

• Multi-environment management
  : Arrange files to distinguish the environment

• Component management
  : according to mvn Parameter determines whether to deploy the corresponding component , Such as nginx, It's mainly used to develop 、 Test environment

5.4  Request dyeing test

5.5 DevOps The whole cycle

5.6  Double cloud and double life

zhenai-framework Encapsulate data access and adaptation middleware , At the same time, the middleware also supports the dual cloud mode .

5.6.1  Double cloud advantage

6、 Architecture experience sharing

Ma Yun once said , Thousands of successful experiences , There are only a few reasons for failure . In the whole process of technology sharing , Miss Peng often mentioned , The pit that their team of architects stepped on during development , Let's sort it out ：

• Do architecture in business enterprises , Cost effective approach is ： Combine business pain points to find a framework , A framework that can solve business problems is a good framework  ;

• The architect team does scheme design ： It is recommended to give priority to open source components , Avoid making wheels repeatedly , Don't over design ;

• The best effect of the business framework is ： After the framework is combined with the business , Can solve 90% Business problems ;

• Business development ： Encountered 80% The problem is the database ,20% The problem is application deployment ,10% It's something else ;

• The team of architects ： Solving existing problems is not the end , The technical architecture must be forward-looking ;

7、 summary

Finally, I will conclude this summary with Mr. Peng's architecture evolution theory , Business architecture must be constantly improved with the development of business , The early stage may be a problem constantly exposed during the continuous development of the business , Followed by specific breakthroughs , Finally, the adjustment to the framework level .