Evolution of e-commerce development

 We    System = Application service + Database storage 
 What kind of business does our system have   What kind of service , How the corresponding data is stored , How a simple business uses higher technology 
 When many of our users come to visit , How to achieve higher scalability , How to achieve high availability of the system 


 How do we rate website size 

 Our service , External is a website ( for instance   JD.COM   TaoBao  ), There are various internal micro services   The underlying business is complex , Consider the capacity of these websites at this time , How to expand , When we use this technology   Why use this technology   What are the advantages and disadvantages of this technology 
 For example, the following 

 Based on Taobao 10 Flow chart for , Our application started like this  , Directly package our project into war  Put it in tomcat 
step1
 We write an application , He can connect db   External services {
    bs framework }  It includes many functions   For example, register to log in 
 All his functions are in one jvm  The following implementation   There is nothing in between rpc  call   even to the extent that db And the project are on the same server 

 When we have a lot of traffic   This architecture has to change 

step2  We can expand the capacity , Deploy 2 individual tomcat , utilize Nginx  Do the forwarding , In this way, we can achieve   Increase the throughput of the whole system   But this brings problems   It's distributed session  Session problems 

 For example, your users A, according to nginx  Fall on   machine 1 On   Well session Left at the node 1   Next time according to nginx  Fall on   machine 2 On  , Such node 2  It's not session  Of   You need to log in again 

 be based on   Distributed ession  The conversation problem    There are several solutions 
 First of all nginx  Load balancing   configure ip hash , According to the user's ip Conduct hash  Then we can see that it falls into that node 
 The problem is that   A single point of failure   my Ip unchanged    I will always be at this point 
{
     Suppose I this tomcat  Hang up   I can't place an order   Because now my   The service collapsed. }
 There is also a waste of resources 
 There is also the neglect of the processing capacity of the machine   for instance   I want my machine to have good performance   Handle more requests   , Poor performance handles fewer requests  
 This is impossible 

 The way 2 session  viscous 

..   Now suppose our db The solution also solves ,  There are still problems with this scheme ,   When the traffic goes up, I have to add services 
 For example, the previous one tomcat  node   Now? 2 individual ,  It will cause a waste of resources 

 Than talking about double 11  Crazy orders , I just need to expand my single node ,  I don't need anything else, like logging in , register , Expansion of commodity modules 
 Join me. It's all taken apart   Words   I just need to add an order   Nothing else to manage 
 Many business visits are not high   But he also takes up resources 

 Next, I will split our single node service , for instance   account number  , Member services   In this case , Dangshuang 11  When placing orders concurrently ,  I only need 
 Just expand the order node , And in every upgrade version , When the expansion goes online , I just need to package and launch the order module   I don't care about anything else ,   all   Service is a single responsibility 

 And then we split vertically according to functions , Some single services   Put it on the next floor , Provide some business information , In this way, the current structure will be gradually formed , By business   For example, member middle office , Trading Center  ,  It can provide us with more intelligence   Such application scenarios as business talk 

 such   There are more services , More services ,1 A into N individual   Between services   Call dependencies also increase 
 For example, I place an order   Check inventory , Check commodity   Look at the marketing strategy    Make payment    I just order ,  I will call 4 A service , I used to follow the process from top to bottom ,  Not now   All his services are remote   Maybe the inventory is in Beijing , The goods are in Shanghai    We have to use telecommunication , Telematics   for instance dubbo,feign, Include http, Since it is telecommunication   There will be a probability of packet loss , Because the network will jitter 


 Because I am calling synchronously   Check the inventory first   I'm checking the goods   Look at marketing   Finally, call payment to make payment 

 Can I make asynchronous calls   for instance   I'll pay first    Then other inventory   goods   Marketing can be done slowly 
 Then there will be an asynchronous call    I'll pay first , Then operate slowly ,

 Also, distributed transactions   For example, using seata  Or use mq   Use seata  The bottom floor will be locked , Distributed transactions cost a lot of resources 
 It is possible that your throughput will not go up  , We usually use mq To achieve final consistency ,
 Generally, we will place orders asynchronously   This way of message oriented middleware     


 We use the message center   He can solve these problems   But he also brings about these problems    One is message repetition , Lost message   idempotent    There's a backlog of news 
 Message middleware   Solved   asynchronous   decoupling   Flow front elimination 

 The reason for network jitter is that there is no retry mechanism   The message may be lost 

 The problem with idempotency is that I call you repeatedly   The results you return are the same 

 For example, when I scan the code to pay    Because of my network card ,  I swept it 2 Pass code   How do I do   Let the first successful consumption   The second   You don't succeed 
 If a message is lost, it will be retried    Retries can occur many times   Suppose we have already dealt with   The second time we should not deal with 

 We can set a unique for each message Id  For example, there is an order when placing an order Id, Now all kinds of mq  There is also a message for sending messages Id
 I handled it successfully   I insert a  , When I insert again , because id Is the only one that cannot be inserted repeatedly ,  I won't insert it the second time 

 The same logic as the de duplication table   It can be used redis do 


 Status code   I handled it successfully , I will change the status to 1 , The next time you insert it   Must be in db Medium is 0  Words   To insert    If it is 1  You won't insert any more 


 There's a backlog of news , I process a piece of data and I put it in a table Insert  A piece of data   If this message records   Every day  mq  Send hundreds of   over a thousand   What do I do    I'm on my watch 
 Insert thousands of data 

 With the development   There will still be a compatibility problem , We can go through feign,dubbo These make remote calls , It can be used redis,es For storage , however dubbo  It's a company ,redis  It's a company . Different companies make different products , But in the case of micro Services   It's a whole set of solutions （ Include redis, dubbo  these ）
 We need container deployment  Devops docker k8s   Continuous integration 
 Our microservices depend on more , If we use container deployment, the images we get can be used directly , You don't have to build it yourself, for example redis  These services 
 Cloud native is a micro service + Operation and maintenance + Containerization   Continuous delivery   I will not affect the normal use of the website   I can often go to iterative development