Select auto increment or sequence for primary key selection?

   

First of all, the design of different databases is different , Some are self increasing , Some have sequences , How to use it? ？ Used many years ago Oracle When we do the public security system every time we pass a car , No matter whether it is a motor vehicle or a non motor vehicle , Places like Shanghai are adding new data every day 1 Up and down .300 More than million local motor vehicles plus some non Shanghai motor vehicles , Less than 400 ten thousand . Normally a private car drives out , Drive back , after 10 It's very common to cross a road , So it is recorded every day 20 Time . So not to 400 ten thousand X20 many times . About 100 million . The situation is similar across the country .

    So how to store these data , Ensure efficient query ？ First of all, we check the vehicle license plate , Then the license plate must be an index . But our vehicles are mainly photographed and stored with photos . Our photos don't exist in the database （ That's too big , But in the early days, it was stored in the database ）, At this time, the primary key and the name of the image should correspond to each other . This requires the uniqueness of the primary key . How to ensure efficiency without conflict .

    Suppose we installed in Jiangsu Province 1 Million pieces of equipment from Xuzhou to Suzhou （ The premise is that we have a big job in the whole province ）. Each device controls 4-8 A driveway , Almost every second 1k-1w The equipment （ You can think about becoming the Internet of things ） Write the collected data to the database . The actual database writes are in the range of hundreds to thousands per second .

    The primary key method we used at that time was administrative division + Equipment serial number + Vehicle Lane + Mm / DD / yyyy HHM / S .

    Administrative division is understood as postal code , We mail letters in the mainland 6 Bit number can be distinguished into a town .

    Equipment serial number , It can be installed at a township level 100 The devices are already saturated . But we kept it 4 position , Can install 9999 Devices . Even Beijing Tiantongyuan, a community that spans three subway stations , That's enough. . The equipment in each administrative division is from 1 Numbered starting .

    Each device has 1-16 A driveway .

    Each lane can only have one car at most in the same second .

  Therefore, the equipment in Tianning District, Changzhou City, Jiangsu Province is 213000+1234+5+2022 year 6 month 1 Japan 10 spot 01 branch 30 second   namely 2130001234520220601100130 This is the primary key . When we look at this number, we can also know where the data comes from .

It can be seen that we are useless Oracle Sequence . For the following reasons ：

1、  Take the sequence competition , We have to process hundreds of thousands per second . There is no competition in design , There is no need to redis. Write fails in case of conflict , When the car passes, it passes , It is impossible to resend . The request cannot be delayed , You can't lose .

2、  Natural numbers have no meaning , No practical use .

3、  What to do if you encounter the biggest problem ？ Show me

When the a The maximum of the stopped , This is suitable for human intervention to continue to expand , But when is the end ？

If circular reuse is adopted, please see the following . Technically, it can , But if the sequence is a primary key, then there is a conflict .

So we don't use sequences , Not a bad sequence , It's better that we design it this way . So there is no problem with billions or tens of billions on a single watch .

 

So are other databases the same ？ Let me see PG Of . Grammar and Oracle It's a little different

 

create sequence a increment by  1 minvalue  1 maxvalue 5 start with  1;

create sequence b increment by  1 minvalue  1 maxvalue 5 start with  1 cycle;

 

Experimental results and Oracle Is exactly the same .

Respectively, stop at the maximum value .

 

And cyclic reuse .

 

 

What if there is no sequence ？ Some databases do not have sequences but have self augmentation , such as MySQL. Please do not mention here from redis To take （ At least in my business flow, I will never design like this ）, The reasons are all said , In addition, what is the use of meaningless storage ？

that MySQL What is the use of self augmentation ？ because Oracle It's a pile of watches ,MySQL It's a clustered index table , Sequential storage is the most compact . that id Self augmentation is to give you compact storage . Never have business implications , Put the business meaning in the second column to create a unique . A few days ago, I encountered several databases to be merged , At this time, it reflects the pain of disobedience . They are all self increasing id Having business meaning or relationship . Now the table data will be merged , Everyone is mutually exclusive . Brain damage . I wish I had listened to you .

So this self increase ID How to choose ？ It's usually int,int Related to scope , It's not about length . give an example tinyint Can only save 256 Data , If you take negative, you'll only get to 127 了 .

So here we are 127 in the future , No more , Add more or 127. If you want to continue, you will report primary key conflicts . This type can only be changed .

    So compatible MySQL Of Tidb How to deal with it ？

 

The result looks like Oracle The sequence of reaches the maximum .

    Different databases , Different scene designs are different . Don't use it wrong .