Understand microservices bit by bit

History of architecture evolution

As the number of users increases , Our project is in a Tomcat Containers 、 One MySQL It's hard on the database , Users complain about the slow response speed , We observe on the server CPU Share soared , Insufficient available memory . This centralized structure is shown in the following illustration , It can also be called “ Single machine or single structure （Monolith）”.

It's certainly a good thing that the number of users has increased , For our technicians, the challenge is even greater . First of all, we can think of some simple and direct methods , Upgrading bandwidth 、 Upgrade the server hardware configuration —— Pretty good , In addition to these horizontal methods , There is also the method of vertical dimension , And this is deeply combined with software , It is not as simple as simply modifying the hard configuration .

In the beginning Web The server 、 database 、 File resources run on the same machine , The processing capacity of a server is limited , First separate them ：Web The server is independent of one machine 、 The database is independent of one machine 、 File storage is independent of a machine —— obviously , Files of static resources, such as pictures 、 Video and so on do not need dynamic processing , So using Apache or Nginx that will do , need not Tomcat, Or deploy in Alibaba cloud OSS Specifically optimize the storage of files , Even join the network optimization CDN Mechanism .

This structure can be called distributed （Distributed） Structure is shown in figure 2 Shown , Of course, the current practice is roughly divided according to the software form .

meanwhile ,Web The program adds various cache Services , for example JSP Dynamic page static , That is also a caching strategy , Of course, there are also special ones for caching Redis or Memcache Service for .

Of course, the database is also related to Web Server separation . If the performance bottleneck is found in the database , Take master-slave database （Master/Slave）、 Strategies of read-write separation and database and table segmentation . At the same time, we should continue to optimize SQL, Join indexing and other work . At this stage, the specific architecture shown in the illustration is finally formed .

After the number of users reaches a certain order of magnitude , It is found that the above optimization methods are ineffective , What to do ？ The above is a machine doing one thing , A machine alone cannot do it , Many machines are needed to support it , So it's still a pile of machines , But the strategy is obviously different . This way we call it “ Cluster structure （Cluster）”, As shown in the illustration .

Services originally run on a single machine , Whether it's Web Server or database server , Take it as a node （Node） Make multiple copies , It forms a cluster , Is multiple instances of a component . For example, at the same time 100 Please come here , The backend is ready for 10 platform Web The server , Then distribute it equally to each set 10 A request to process . It can be seen that there is a key role in scheduling how these nodes respond to requests , be called “ Load balancing server ”, All requests from the user are given to it first , Then implement the distribution according to the load . Through a certain algorithm , Let idle nodes （ Low load ） Priority response , The faulty node is not involved .

The benefits of more machines go beyond the improvement of performance , It also makes disaster tolerance possible , When a machine hangs, it hangs , The cluster machine allows some machines to hang up and the whole system is still available .

Please experience distributed （ As illustrated in the illustration 7.2） With clusters （ As illustrated in the illustration 7.4） Subtle differences , The reason why distributed tasks are distinguished A、B、C…… Represents different businesses that have been split , For example, an order business in the mall can be divided into order initialization business 、 Payment business 、 Financial business 、 Logistics business ; There is no distinction between the tasks of the cluster A、B、C…… Yes means that the task is an overall business , Not split and isolated from each other , Replication is performed on multiple servers . in other words , The task under the cluster is still a single machine application , It includes multiple business capabilities , And no communication with other nodes , The task is completed in the process of a single node from beginning to end .

Clustering and distribution do not conflict , In practice, they are often used together .

When the business grows to a certain extent , It is found that the above optimization methods are ineffective , What to do ？ Although the advantage of the cluster structure is that the system horizontal expansion is very easy , But it has to be said that the cost of stacking machines is very high . Besides, this problem is not linear , For example, the second kill promotion in the e-commerce scene ： Hinder the integration of stand-alone applications ,“ seckill ” Are deployed in one WAR The inside of the bag , In that case , When a large number of requests rush in at a high speed , Unable to identify which traffic is seckill , Which is the traffic of normal pages , It is likely that promotional activities have affected normal network access . Consider the situation , We have to put more detailed granularity into the system to think , specifically , It is necessary to extend a module separately , Not as a whole “ In general ” Expand （ Although it has been clustered ）. Gradually, a kind of “ Business split ” The technical means of , Use independent processes to host some business logic , It can be expanded in more detail , More scalable . Such a single business perspective can be understood as a further deepening of distributed applications .

A distributed task can be regarded as a service （Service）, Each server node only completes this kind of service , More focused . This is not only the requirement of performance optimization , At the same time, it is also an urgent need to rationalize the architecture . Just imagine , A system has forums, mailboxes and chat rooms , It involves user login and user center , Because they are deployed separately , Not only resources are repeatedly wasted, but also management is not easy . So in order to overcome the above problems , Various ideas and methodologies such as distributed service have been introduced into large-scale projects .

Microservice style architecture

The deep distributed architecture is characterized by business splitting （ Service split ）, Form a stronger connection than the cluster network 、 Closer network . In the past, our learning and working environment was to put all the code in one project , Package when deploying WAR Upload the package to the server , This is the single machine structure . Even if it evolves into a cluster structure , The actual work is to deploy several more servers, but the code is basically the same , There is no need to make too big changes . But after entering the current distributed structure , The old code has to be modified according to the communication problems between several independent processes .

Suppose it is still an example of a mall , According to the idea of microservice , We need to split it into multiple independent services according to functional modules , Such as ： Customer service 、 Product service 、 Order service 、 Background management services 、 Data analysis services and so on . Each service can be independent Web Container or database , Even these services can be written in different languages （ Such as Java/PHP/Go/Node.js etc. ）.

In the past single machine structure , One module calls another module , It's as simple as a function calling another function , No violation and no affectation , Cross process calls are rarely made , There is no service registration only for microservices 、 Find out . But after the service-oriented transformation of business splitting , More questions to consider , Distributed network topology 、 Network reliability 、 Communication mechanism and other contents have become the topic of our attention . inevitably , Services need a mutual communication mechanism to communicate , This is RPC（Remote Procedure Call） Called .

The benefits of microservices are ： Say goodbye to strong coupling 、 chaos 、 General large-scale system , Dissolve into one “ Small service ”, Be able to expand and maintain a module in a targeted way , For example, the example of the above mall , Just increase the number of nodes in the order module and the seckill module before promotion , And the background system can maintain its original ability ; In addition, because of decoupling 、 So the modules can be easily replaced . The benefit of microservices is the reuse of modules with the structure , That's understandable , It's like splitting independent functions , Code reuse can be realized . This is not due to performance , It's the benefits of sorting out after business service .

In order to solve the problem that cannot be solved by cluster, a distributed microservice is proposed . Generally speaking, there is no silver bullet level panacea for microservices , Is still “ Divide and rule （Divide and Rule）” Philosophy of . On the other hand , In order to solve one problem, more problems are introduced to overcome . Is it worth it ？ Once the microservice transformation or implementation is unsuccessful , Often not “ Change numerous for brief ” It is “ Simple things become complicated ” 了 , Do more harm than good . This is not a logical paradox , Or the eternal golden rule ： In short, it depends on the choice of needs , When clustering is really not enough , Just go to wechat .

When it comes to microservices, I have to mention that they existed more than ten years ago SOA（Service -Oriented Architecture）, There are many overlapping concepts with microservices , but SOA The disadvantages of this method are as follows ESB（ The service bus ） and WebService The burden of subclasses is heavy . Microservices , seeing the name of a thing one thinks of its function , It is characterized by “ minute ” Each sub service constitutes a large system , Relatively lightweight .