Cache consistency solution - how to ensure the consistency between the cache and the data in the database when changing data

Write it at the front

We all know , Caching is to improve the reading speed of data , The scenario to deal with is the scenario of reading more and writing less .

When reading data, we usually read the cache first , If the cache does not read the database again, then update the cache .
From query database performance optimization to redis cache - Talk about cache penetration 、 An avalanche 、 breakdown

When the cached data needs to be modified , To modify the cache , You have to modify the database , How to ensure that the cache is consistent with the data in the database ？

How to cache data consistency

Double write mode

The double write mode is to write the database first , Write cache again ：

disadvantages

High and low , Due to Caton and other reasons , Causes write caching 2 At the front , Write cache 1 rearwards , This leads to the inconsistency between the cache and the database data , That is, there is dirty data .

however , This kind of dirty data is only temporary , Depends on the expiration time of the cache , When the cache expires , Reading the database again will get the latest correct data . in other words , The latest data is that there may be a certain delay , But the final result is consistent .

Failure mode

disadvantages

In the case of high concurrent read and write , Write db2 The deletion cache of is executed in read db1 In front of the update cache of , This will cause inconsistency between the cache and the database , That is, there is dirty data .

however , This kind of dirty data is only temporary , Depends on the expiration time of the cache , When the cache expires , Reading the database again will get the latest correct data . in other words , The latest data is that there may be a certain delay , But the final result is consistent .

analysis

Whether it's dual write mode or failure mode , Will cause cache inconsistencies . That is, multiple instances （ database 、 cache ） Updating at the same time will cause dirty data . What do I do ？

1、 If it's user dimension data （ Order data 、 User data ）, This concurrency rate is very small （ The action that the user clicks won't be fast ）, Don't think about it , Even if there is dirty data , Cache data plus expiration time , Trigger the active update of read at regular intervals .
2、 If it's a menu , Basic data such as commodity introduction , If the demand for real-time is high , You can also use canal subscribe binlog The way .
3、 Cache data + The expiration time is also sufficient to solve the cache requirements of most businesses （ Business is not about heart data , Allow temporary dirty data to be ignored ）.
4、 Ensure concurrent read and write by locking , Line up as you write . It doesn't matter to read . So it's suitable to use read-write lock .
5、 Locking will make the system bulky , So can we not add it or not , But the requirements for data consistency are very high , That can only be locked .

summary

1、 The data we can put into the cache should not be real-time 、 The requirement of consistency is very high . So when caching data Plus the expiration time , Ensure to get the latest data every day or after a period of time .
2、 We should not over design , Increase the complexity of the system , For distributed locks , Don't add if you can ; For cached dirty data , If you can tolerate dirty data for a period of time, you don't need to deal with , Otherwise, the system will be too heavy to maintain .
3、 When it comes to real-time 、 Data with high consistency requirements , And write more scenes , You should check the database , Even if you slow down .
4、 Reading more and writing less , If you have high requirements for data consistency , You can guarantee concurrent reading and writing by locking , Line up as you write . It doesn't matter to read . So it is suitable for Read-write lock .
5、 Use canal subscribe binlog Can solve the problem of data consistency , But new middleware has been introduced .

The way we usually use

1、 All cached data will be added with expiration time , When the data is out of date, the next query triggers active update .
2、 Reading and writing data , Plus distributed read-write locks .

expand ： Use canal subscribe binlog

binlog yes mysql Binary log , For statements that operate the database , All in this form .Canal It's Ali MySQL database Binlog Incremental subscription of & Consumer component . database-based Binlog It can monitor the changes of database data and then be used for data synchronization and other services .

In short , Is to subscribe to the database binlog, If the database is modified , Get this modification event and modify the cache at the same time , Ensure cache consistency .

This method introduces new middleware , Make the system more complicated . But perfect ensures cache consistency , Usually it only needs to be maintained once , Subsequent maintenance is not required .