How MySQL and redis ensure data consistency

---

Preface

How to ensure the consistency of database and cache double write , Here are a few ways , Let's discuss their advantages and disadvantages .

---

One 、 Update the database first , Update again redis

This program , No one should adopt .
If the database is successfully updated first , Then update redis Failure , Then the data will be inconsistent , So this method abandons

Two 、 To update redis, Updating the database

This scheme is similar to the first , It also has the same problem
If updated reids success , Database update failed , It will also cause data inconsistency

3、 ... and Update the database first , And then delete redis

This program , It will also cause data inconsistency , But compared with the previous two schemes , If he sets key The expiration time of , Then the final consistency of the data is guaranteed . If you delete after updating the database redis Failure , It is not set redis Expiration time . Then the data will be inconsistent .

If the thread A Update the database , Preparing to update redis when . Threads B Updating threads A to update redis I got redis Data in , Then the data obtained from other data is still old data , If you remove redis Failure will also cause data inconsistency

Solution ：
1.mysql and redis Set up transactions , Rollback data when an exception occurs
2.redis Set retry mechanism , Enter retry mode after deletion failure

Four Delete first redis, Update the database

This program , There is also a certain probability of inconsistency , But it is close to the optimal .

If the thread A Deleted redis, Preparing to update database . Threads B The query redis No after , Query the old data of the database , And write it down redis. After that thread A To update the data successfully , Database and redis The data are inconsistent

Solution ： Delay double delete

Threads A In the delete redis And after updating the database , After a period of sleep , Delete again reids Data in . This sleep time must be greater than the time of a query .

---

summary

There is no absolute guarantee of data consistency , Or lock at the expense of performance , Or serial . At high level , These schemes can only be optimized