Is there anyone who plans to open source an industrial "seckill" system architecture?

Want to figure out a set of architectural solutions , The most fundamental way , Is to go practice it .

But , Most programmers , There is no such business , I can't touch such a scene , What shall I do? ？

A friend set up a set automatically , can Let everyone instantly experience and tune the seckill architecture with high concurrency , Share with you ！

For seckill business , On the system How can it be optimized ？

In the direction of , There are two main points ：

First of all , Try to intercept the request upstream of the system , Don't let lock conflicts fall on the database .

The reason why the traditional seckill system hangs , It's because the requests are pushed to the back-end data layer , Data read-write lock conflict is serious , High concurrency and slow response , Almost all requests time out , High traffic , The effective flow of successful orders is small .

second , Make the most of the cache .

Second kill buy tickets , This is a typical business scenario with more reading and less writing ：

（1） Train number inquiry , read , Large amount of ;

（2） Check the remaining tickets , read , Large amount of ;

（3） Order and pay , Write , Small quantity ;

Great for using caching to optimize .

Second kill business , How about the common system layered architecture ？

Second kill business , You can use a typical service layered architecture , How should these four layers be optimized ？

One 、 Request interception on end

JS level , You can restrict users to x Only one request can be submitted in seconds , So as to reduce the system load .

Voice over ： Submit frequently , Friendly tips “ The frequency is too fast ”.

APP level , You can do something similar , Although users are crazily shaking wechat to grab red packets , But in fact x Seconds to send a request back .

Voice over ： That's what's called “ Try to intercept the request upstream of the system ”, browser /APP Layer can intercept 80%+ Request .

Two 、 Site level request interception

How to resist programmers writing for Cycle call http Interface , The first thing to do is to identify the unique identity of the user , Block frequent users .

What to use as the unique identification of users ？

All ticket businesses need to log in , use uid Can identify users .

At the site level , To the same uid Request count and speed limit , for example ： One uid,5 Seconds only pass through 1 A request , This will stop 99% Of for Loop request .

One uid,5s Only through a request , What about the rest of the requests ？

cache , Page caching ,5 Other requests to the site layer within seconds , All return to the last page .

3、 ... and 、 Request interception of service layer

Concurrent requests have reached the service level , How to intercept ？

The service level is very clear about the inventory of the business , Very clear about the database's ability to withstand pressure , According to these two, we can cut the peak and speed limit .

for example , Business services know very well , A train has nothing but 2000 Tickets , At this time, transmission through 10w Request to database , It doesn't make sense .

Voice over ： If the database can only resist 500 Write a request , Just pass through 500 individual .

With what to cut the peak ？

Request queue .

For write requests , Make a request queue , Only a limited number of write requests are transmitted to the data layer at a time （ Place the order , Pay for writing business like this ）.

Only 2000 Train tickets , Even if 10w Please come here , Only through transmission 2000 Go to the database ：

（1） If all previous requests are successful , Put down another batch of ;

（2） If the previous request is out of stock , Then all subsequent requests return “ It's sold out ”;

For read requests , How to optimize the ？

cache resist , Whether it's memcached still redis, Single machine anti per second 10w There should be no problem .

Voice over ： Expand the cache horizontally , It's easy to expand linearly .

Such peak clipping and current limiting , There are very few requests to write , And very little read caching mis The request will penetrate to the data layer , And then there is 99% Our request was stopped .

Four 、 Database layer

After the first three layers of optimization ：

（1） The browser blocked 80% request ;

（2） The site layer intercepted 99% request , And do page caching ;

（3） Service layer according to business inventory , And the database's ability to withstand pressure , Write request queue and data cache are done ;

You'll find that , Every request that penetrates to the database layer is controllable .db Basically there's no pressure , walk .

Voice over ： This kind of business data volume is not large , There's no need for sub Treasury , Make a highly available database .

According to the optimization scheme above , In fact, the most pressure is the site layer , Suppose that the number of real valid requests is... Per second 100w, How to deal with this part of the pressure ？

There are two ways to solve this problem ：

（1） Site level expansion , Expand capacity by adding machines , A resistance 5W,20 It's done ;

（2） service degradation , Abandon the request , For example, abandon 50%;

The principle is to protect the system , Can't let all users fail .

How to experience the seckill business , Build a set of micro service layered seckill system architecture ？

（1） First , Buy hardware resources on the cloud ;

（2） next , Build the underlying foundation components ： The storage system （MySQL、MongoDB、TiDB etc. ）、 High performance cache system （Redis etc. ）、 Highly reliable message system （RocketMQ、Kafka、Pulsar etc. ）、 Search engine system （Elasticsearch、Solr etc. ） And many other basic components ;

（3） The third step , Build a microservice management platform ：Docker Containers 、Kubernetes Management platform and other highly concurrent flexible container management platforms 、CI/CD And so on ;

（4） Step four , Build a micro service layered architecture ： Low code microservice seckill business architecture design 、 Code landing 、 Test verification ;

（5） Step five , Optimize the three high design ： Implement high performance 、 High availability 、 High reliability and other design guarantees , Ensure the robustness of seckill system ;

（6） Last , Set up a visual performance management background ： It is convenient to see the throughput of the whole link of the system through three-dimensional and multi-dimensional visualization , Flow ;

Voice over ： Put aside hundreds of thousands of machine costs , The whole stack 20 People build teams , One year can almost build a prototype .

Is there a quick way to experience ？

In order to make more children's shoes experience quickly “ seckill ” System architecture , My friend @ Xuanjie Founded Naixue , And lead the team through 2 Years of research and development , Created an exclusive set , Industrial grade “ Second kill architecture ” laboratory , It integrates all the above functions , The user only needs to open a web page , You can experience and code tuning “ Second kill architecture ”：

（1） left ： Spike architecture knowledge points ;

（2） On the right side ：terminal Direct connection server , Can operate ng,web-server,service etc. , You can modify the code directly , Tune the architecture ;

（3） meanwhile ： There is also the background to manage the architecture , And visual monitoring .

Voice over ： Details are shown below .

Figure 1 ： left - Architecture diagram , On the right side - Practice environment

Figure 2 ： left - Operation instructions , On the right side - One click automation architecture , Check the log

Figure 3 ： left - Architecture description , On the right side - View and modify the code

Figure 4 ： Visualization platform

How to apply for experience ？

1 Meta scan code , Apply for the experience now , Only 100 Accounts

1 Meta scan code , Apply for the experience now .

meanwhile , Free gift ：

（1）6.27-6.29, Explanation of secsha core architecture ;

（2） Industrial grade “ Second kill architecture ” Introduction to laboratory functions ;

（3） Source code and courseware ;

Explain , Due to engineering grade “ Second kill architecture ” The server resources required are large , Limited account , Front only 100 people .

Read the original , Get involved right now , Pro automatic handwriting code optimization , Actual industrial grade “ Second kill architecture ”.