6000 + words to help you understand the evolution of Internet architecture!

Click on the official account ,Java dried food Timely delivery

author ： Small M
source ：https://cnblogs.com/xiaoMzjm/p/5223799.html

Preface

We use javaweb For example , To build a simple e-commerce system , See how the system can evolve step by step .

The function of the system ：

• User module ： User registration and management

• Commodity module ： Commodity display and management

• Trading module ： Create transactions and manage

Stage 1 、 Build a website on its own

The beginning of the site , We often run all our programs and software on a single computer . Now we use a container , Such as tomcat、jetty、jboos, Then use it directly JSP/servlet technology , Or use open source frameworks like maven+spring+struct+hibernate、maven+spring+springmvc+mybatis; Finally, choose a database management system to store data , Such as mysql、sqlserver、oracle, And then through JDBC Connect and operate the database .

Load all the above software on the same machine , The app is running , It's also a small system . At this time, the system results are as follows ：

Stage two 、 Separation of application server and database

With the launch of the website , The number of visitors is on the rise , The load on the server is slowly increasing , Before the server is overloaded , We should be ready to , Improve the load capacity of the website . If our code level has been difficult to optimize , Without improving the performance of a single machine , Adding machines is a good way , It can not only effectively improve the load capacity of the system , And it's cost-effective .

What are the additional machines used for ？ At this point we can put the database ,web The server is split , This not only improves the load capacity of a single machine , It also improves disaster tolerance .

The architecture after the application server is separated from the database is shown in the figure below ：

Stage three 、 Application server cluster

As visits continue to grow , A single application server can no longer meet the needs . Assuming that the database server is not under pressure , We can turn an application server from one to two or more , Distribute user requests to different servers , So as to improve the load capacity .

There is no direct interaction between multiple application servers , They all rely on databases to provide services to the outside world . The famous software for failover is keepalived,keepalived Is a similar to layer3、4、7 Software for the exchange mechanism , It's not the exclusive product of a specific software failover , It's a product that can be applied to all kinds of software .keepalived Match up ipvsadm It can also do load balancing , It can be called a artifact .

Let's take the example of adding an application server , The added system structure is as follows ：

The system evolved here , There will be four questions ：

1. Who will forward the user's request to the specific application server

2. What's the forwarding algorithm

3. How the application server returns the user's request

4. If users visit different servers every time , How to maintain session The consistency of

Let's take a look at the solution ：

1、 The first problem is load balancing , Generally speaking, there are 5 Kind of solution ：

1、http Redirect .HTTP Redirection is the application layer request forwarding . The user's request has arrived HTTP Redirect the load balancing server , The server requires the user to redirect according to the algorithm , When the user receives a redirect request , Request the real cluster again

advantage ： Simple .

shortcoming ： Poor performance .

2、DNS Domain name resolution load balancing .DNS Domain name resolution load balancing is in the user request DNS The server , Get the... Corresponding to the domain name IP Address time ,DNS The server directly gives the server after load balancing IP.

advantage ： hand DNS, We don't need to maintain the load balancing server .

shortcoming ： When an application server hangs up , Can't inform in time DNS, and DNS The control of load balancing is in the domain name service provider , The website can't do more improvement and stronger management .

3、 Reverse proxy . When the user's request reaches the reverse proxy server （ It has reached the website machine room ）, By the reverse proxy server according to the algorithm forward to the specific server . frequently-used apache,nginx Can act as a reverse proxy server .

advantage ： Simple deployment .

shortcoming ： Proxy servers can be a performance bottleneck , Especially a big file upload .

4、IP Layer load balancing . After the request reaches the load balancer , The load balancer modifies the request by IP Address , So as to realize the request forwarding , Load balancing .

advantage ： Better performance .

shortcoming ： The broadband of load balancer becomes the bottleneck .

5、 Data link layer load balancing . After the request reaches the load balancer , The load balancer modifies the requested mac Address , So as to achieve load balancing , And IP Load balancing is different from , After requesting access to the server , Direct return to customer . Without going through the load balancer .

2、 The second problem is the cluster scheduling algorithm , Common scheduling algorithms include 10 Kind of .

1、rr round-robin scheduling . seeing the name of a thing one thinks of its function , Polling for distribution requests .

advantage ： Implement a simple

shortcoming ： Regardless of the processing power of each server

2、wrr Weighted scheduling algorithm . We set weights for each server weight, The load balancer dispatches the server according to the weight , The number of times the server is called is proportional to the weight .

advantage ： Considering the different processing power of the server

3、sh Original address hash ： Extract users IP, From the hash function, we get a key, Then according to the static mapping table , Investigate and deal with the corresponding value, The target server IP. Overload the target machine , It returns null .

4、dh Destination address hash ： ditto , It's just that what we're extracting now is the target address IP To make hash .

advantage ： Both of the above algorithms can realize the same user accessing the same server .

5、lc The minimum connection . Priority is given to forwarding requests to servers with few connections .

advantage ： Make the load of each server in the cluster more even .

6、wlc Weighted least connected . stay lc On the basis of , Weight each server . Algorithm for ：（ Number of active connections *256+ Number of inactive connections ）÷ The weight , Servers with small calculated values are preferred .

advantage ： Requests can be allocated according to the capabilities of the server .

7、sed In the short term, we hope to delay . Actually sed Follow wlc similar , The difference is that the number of inactive connections . Algorithm for ：（ Number of active connections +1)*256÷ The weight , The server with small calculated value is preferred .

8、nq Never in line . The improved sed Algorithm . Let's think about the circumstances under which we can “ Never in line ”, That's the number of connections to the server 0 When , So if there are server connections 0, The equalizer forwards the request directly to it , No need to go through sed The calculation of .

9、LBLC Minimal connections based on locality . Equalizer according to the purpose of the request IP Address , Find out what to do IP Address recently used by the server , Forward the request , If the server is overloaded , The least number of connections algorithm .

10、LBLCR Minimum connections based on locality with replication . Equalizer according to the purpose of the request IP Address , Find out what to do IP Address recently used “ The server Group ”, Be careful , It's not a specific server , Then select a specific server from the group with the minimum number of connections , Forward the request . If the server is overloaded , Then according to the algorithm of the minimum number of connections , In the cluster Not Servers in this server group , Find a server out , Join this server group , Then forward the request .

The latest interview questions have been sorted out , You can Java Interview library applet online brush questions .

3、 The third problem is cluster mode , commonly 3 Kind of solution ：

1、NAT ： The load balancer receives the user's request , Forward to a specific server , The server processes the request and returns it to the equalizer , The equalizer returns to the user .

2、DR ： The load balancer receives the user's request , Forward to a specific server , After the server comes out to play the request, it directly returns it to the user . Need system support IP Tunneling agreement , It's hard to cross platform .

3、TUN ： ditto , But there is no need for IP Tunneling agreement , Good cross platform , Most systems can support .

4、 The fourth question is session problem , Generally speaking, there are 4 Kind of solution ：

1、Session Sticky .session sticky That is to put the request of the same user in a certain session , All assigned to a fixed server , So we don't have to deal with cross server session Problem. , Common algorithms are ip_hash Law , That is, the two hash algorithms mentioned above .

advantage ： Implement a simple .

shortcoming ： When the application server is restarted session disappear .

2、Session Replication .session replication It's replication in a cluster session, Make sure that every server has all the users session data .

advantage ： Reduce the load balancing server pressure , There is no need to achieve ip_hasp Algorithm to forward requests .

shortcoming ： When copying, broadband costs a lot , If you have a large number of visitors session It takes up a lot of memory and wastes .

3、Session Centralized data storage ：session Data centralized storage is to use database to store session data , Realized session Decoupling from application server .

advantage ： comparison session replication The plan , There's a lot less pressure on broadband and memory between clusters .

shortcoming ： Need to maintain storage session The database of .

4、Cookie Base ：cookie base Is to put session There is cookie in , There is a browser to tell the application server my session What is it? , It's also implemented session Decoupling from application server .

advantage ： Implement a simple , Basically maintenance free .

shortcoming ：cookie Length limit , Low security , Broadband consumption .

It is worth mentioning that ：

nginx Currently supported load balancing algorithms include wrr、sh（ Supports consistent hashing ）、fair（ I think it comes down to lc）. but nginx As an equalizer , It can also be used as a static resource server .

keepalived+ipvsadm More powerful , Currently, the algorithms supported are ：rr、wrr、lc、wlc、lblc、sh、dh

keepalived There are ：NAT、DR、TUN

nginx It doesn't provide session Synchronization solution , and apache It provides session Shared support .

Okay , After solving the above problems , The structure of the system is as follows ：

Stage four 、 Database read-write separation

We always assume that the database load is normal , But as the number of visitors increases , The load on the database is also increasing . Then someone may immediately think of the same as the application server , One copy of the database is the second load balancing . But for databases , It's not that simple . This MySQL Database development 36 Rules ！ I suggest you look at .

If we simply split the database in two , Then the request for the database , Load separately to A The machine and B machine , So it is obvious that the data of the two databases will be inconsistent . So in this case , We can first consider the use of read-write separation .

The structure of the database system after the separation of reading and writing is as follows ：

This structural change will also bring about two problems ：

1. Data synchronization between master and slave databases

2. Application selection of data sources

The solution to the problem ：

1. We can use MYSQL Self contained master+slave To achieve master-slave replication .

2. Third party database middleware is adopted , for example mycat.mycat It's from cobar Developed from , and cobar It's Alibaba's open-source database middleware , Later, development stopped .mycat It's better at home mysql Open source database sub database sub table middleware .

Stage five 、 Use search engine to relieve the pressure of Reading database

If the database is a reading database , Often unable to do fuzzy search , Even if the separation of reading and writing is done , This problem has not yet been solved . Take our trading website as an example , Published items are stored in the database , The most commonly used function of users is to find products , Especially according to the title of the product to find the corresponding product . For this need , We usually go through like Function to achieve , But the cost of this approach is very high . At this time, we can use the inverted index of search engine to complete .

Click on the official account ,Java dried food Timely delivery

Search engine has the following advantages ：

• It can greatly improve the query speed .

The introduction of search engine will also bring the following costs ：

• Bring a lot of maintenance work , We need to implement the index building process ourselves , Design total / Additional building methods to meet non real time and real-time query needs .

• Need to maintain search engine cluster

Search engine can't replace database , He solved the problem of “ read ” The problem of , Whether to introduce search engine , Need to consider the needs of the whole system . The system structure after the introduction of search engine is as follows ：

Stage six 、 Use cache to relieve the pressure of reading

1、 Cache of background application layer and database layer

As the number of visitors increases , Gradually, many users access the same part of the content , For these more popular content , There's no need to read from the database every time . We can use caching technology , For example, you can use google Open source caching technology guava Or use memcacahe As the cache of application layer , You can also use redis As the cache of database layer .

in addition , In some cases , Relational databases are not very suitable , For example, I want to make a “ Limit the number of password errors per day ” The function of , The idea is probably when the user logs in , If login error , Record the user's IP And the number of mistakes , So where is the data to be put ？

in addition , The latest database series interview questions have been sorted out , You can Java Interview library applet online brush questions .

If it's in memory , So obviously it will take up too much content ; If it's in a relational database , Then we should establish database tables , And a resume java bean, And write SQL wait . And analyze the data we want to store , It's just like {ip:errorNumber} In this way key:value data . For this kind of data , We can use NOSQL Database to replace the traditional relational database .

2、 Page caching

In addition to data caching , And page caching . For example, use HTML5 Of localstroage perhaps cookie.

advantage ：

• Reduce the pressure on the database

• Greatly improve access speed

shortcoming ：

• Need to maintain cache server

• Increased the complexity of coding

It is worth mentioning that ：

The scheduling algorithm of cache cluster is different from the application server and database mentioned above . It's better to use “ Consistent hash algorithm ”, In this way, we can improve the hit rate . Let's not talk about this , If you are interested, please refer to the relevant information .

Structure after adding cache ：

Stage seven 、 Database horizontal split and vertical split

Our website has evolved to the present , transaction 、 goods 、 The user's data is still in the same database . Despite the increased cache , The way of separation of reading and writing , But as the pressure on the database continues to increase , The bottleneck of database is more and more prominent , here , We can choose to split data vertically or horizontally . Want to be an architect , This The architect's Atlas suggests taking a look , Little detours .

7.1、 Data split vertically

Vertical split means to split different business data in the database into different databases , Combined with the present example , It's about trading 、 goods 、 The user's data is separated .

advantage ：

• It solves the problem of putting all businesses in one database .

• More optimization can be made according to the characteristics of the business

shortcoming ：

• Need to maintain multiple databases

problem ：

1. Need to consider the original cross business transactions

2. Cross database join

The solution to the problem ：

1. We should try to avoid cross database things in the application layer , If you have to cross databases , Try to control... In your code .

2. We can use third parties to solve , As mentioned above mycat,mycat Provides rich cross Library join programme , Please refer to mycat Official documents .

The vertical split structure is as follows ：

7.2、 Data horizontal split

Data horizontal splitting is to split the data in the same table into two or more databases . The reason for data level splitting is that the data volume or update volume of a business reaches the bottleneck of a single database , At this point, you can split the table into two or more databases .

advantage ：

• If we can solve the above problems , Then we will be able to do a good job of data volume and write volume growth .

problem ：

1. The application system of accessing user information needs to solve SQL Routing problem , Because now the user information is divided into two databases , You need to know where the data you need to operate is .

2. Primary key processing is also different , For example, the original auto increment field , We can't simply continue to use .

3. If pagination is needed , That's the trouble .

The solution to the problem ：

1. We can still solve the third-party middleware through , Such as mycat.mycat Can pass SQL Parsing module for our SQL To analyze , According to our configuration , Forward the request to a specific database .

2. We can go through UUID Guarantee uniqueness or customization ID Plan to solve .

3. mycat It also provides a rich paging query scheme , For example, do paging query from each database first , Then merge the data to do a paging query and so on .

Structure after data horizontal split ：

Stage eight 、 Split of application

8.1、 Split application

As the business grows , More and more business , More and more applications . We need to think about how to avoid making applications more and more bloated . This requires taking the application apart , From one app to two or more . Or take our example above , We can put users 、 goods 、 The deal is split up . become “ user 、 goods ” and “ user , transaction ” Two subsystems .

The split structure ：

problem ：

1. After this split , There may be some of the same code , Such as user related code , Products and transactions need user information , So in both systems, we keep the same code for operating user information . How to ensure that these codes can be reused is a problem to be solved .

solve the problem ：

1. By taking a service-oriented route to solve

8.2、 Take the road of service

In order to solve the above problems after splitting the application , We split up public services , Form a service-oriented model , abbreviation SOA. The latest micro service interview questions have been sorted out , You can Java Interview library applet online brush questions .

Adopt the system structure after the service ：

advantage ：

• The same code will not be scattered in different applications , These implementations are in various service centers , Make the code better maintained .

• We put the interaction of database in each service center , Give Way ” front end “ Of web Application pays more attention to the work of interaction with browser .

problem ：

1. How to make remote service call

resolvent ：

1. We can solve this problem by introducing message oriented middleware

Stage nine 、 Introduce message middleware

As the website continues to grow , There may be sub modules developed in different languages and sub systems deployed on different platforms in our system . At this point we need a platform to deliver reliable , Platform and language independent data , And it can make load balancing transparent , It can collect call data and analyze it during the call , Guess the growth rate of website visit and so on , Make predictions about how websites should grow . Open source message middleware has Alibaba's dubbo, Collocation Google Open source distributed program coordination service zookeeper Realize server registration and discovery .

The structure after the introduction of message middleware ：

Ten 、 summary

The above evolution is just an example , Not for all sites , In fact, the evolution process of website is closely related to its own business and different problems encountered , There is no fixed pattern . Only serious analysis and continuous exploration , To find the right architecture for your site .

Spring Boot After the scheduled task is started , How to stop automatically ？

Work 3 My colleagues in didn't know how to roll back the code ！

23 Design mode and Practice （ Very comprehensive ）

Spring Boot Protect sensitive configurations 4 Methods ！

Goodbye, single dog ！Java Create the object's 6 Ways of planting

Why does Ali recommend LongAdder？

A new technical director ： No code writing with headphones ..

blockbuster ！Spring Boot 2.7 Official release

Java 18  Official release ,finalize deprecated ..

Spring Boot Admin Born in the sky ！

Spring Boot Learning notes , This is so complete ！

Focus on Java Technology stack, see more dry goods

obtain Spring Boot Practical notes ！