Isolation level

problem

When multiple transactions in the same application or multiple transactions in different applications are executed concurrently on the same dataset , There may be many unexpected problems , These problems can be divided into the following three types ：

Dirty reading (Drity Read):

Two transactions are known A and B, A Read has been B Updated but not yet submitted data , after ,B Roll back the transaction ,A The data read is dirty data .

scene ：

The company paid , Leaders 5000 Yuan to Tom On your account , But the transaction was not committed , and Tom Just to check the account , Found that the salary has arrived , There are too many accounts 5000 element , I'm very glad to , But unfortunately , The leader found that Tom My salary is not right , yes 2000 element , So the transaction is rolled back quickly , After modifying the amount , Commit a transaction ,Tom When checking the account again, I found that there were only more accounts 2000 element ,Tom I'll have a good time , From then on, I was depressed , On the road of no return ……

analysis ：

The above situation is dirty reading , Two concurrent transactions ：“ Business B： Leaders give Tom salary ”、“ Business A：Tom Check salary account ”, Business A Read transaction B Data not yet submitted .

It can't be read repeatedly (Non-repeatable read):

Two transactions are known A and B,A Read the same data multiple times ,B stay A In the process of multiple reading, the data is modified and submitted , Lead to A When reading the same data multiple times , Inconsistent results , Example ：

scene ：

Tom Take the wage card to spend , Pay the bill at the cashier after drinking and eating , The waiter told him about this consumption 1000 element ,Tom Give the bank card to the waiter , The waiter inserts the bank card POS machine ,POS The machine reads that the card balance is 3000 element , It's just Tom Dawdling to enter the password , His wife hid her ears in a flash Tom Salary card 3000 Yuan transferred to his account and submitted the transaction , When Tom Enter the password and click “ confirm ” After button ,POS The machine has detected Tom I don't have any money for my salary card , Deduction failed ,Tom I wonder , It's clear that Cary has money , So I doubt POS There are ghosts , Fight with the little cashier sister ,300 After the round, he died because of his serious injury ,Tom My wife is in pain , unhappy , Since then, we have embarked on the road of no return …

analysis ：

The above situation is non repeatable , Two concurrent transactions ,“ Business A：POS Machine deduction ”、“ Business B：Tom My wife transfers money online ”, Business A Read the data in advance , Business B Then the data is updated and the transaction is committed , And the business A When you read the data again and deduct , The data has changed .

scene ：

Tom When you take a paycard to spend , once POS The machine reads the salary card information （ That is, the transaction begins ）,Tom Even if the wife transfers money , stay Tom Enter the password and click “ confirm ” After button ,POS The machine has detected Tom The balance on the payroll card has not changed , Finally, the deduction is successful .

analysis ：

The above situation is called repeated reading

Fantasy reading (Phantom Read):

Two transactions are known A and B,A Read data from a table , then B Some new data has been inserted into the table , Lead to A Read the same table again , There will be more lines , In short , A range of records are read successively in a transaction , But the number of records read each time is different , Call it phantom reading , Example

scene ：

Tom My wife works in the banking department , She often checks through the bank's internal system Tom Salary card consumption records .2019 year 5 Some day of the month , She inquired Tom The total consumption of the salary card in the current month （select sum(amount) from record where card_id=‘6226090219290000’ and date_format(create_time,’%Y-%m’)=‘2019-05’） by 80 element ,Tom My wife was very surprised , assume “ My husband is so frugal , It's good to marry him ！”, and Tom At this time, I was just outside and paid at the cashier , consumption 1000 element , That is, a new one is added 1000 Yuan of consumption records and submitted transactions , The wife immersed in happiness inquired Tom Salary card consumption details of the current month （select amount from record where card_id=‘6226090219290000’ and date_format(create_time,’%Y-%m’)=‘2019-05’） To find out , The result of the search turned out to be one 1000 Yuan consumption ,Tom My wife was instantly furious , Takeout ordered a large durian , The evening fell ,Tom Living in deep water , I only feel the pain of needle pricking in my knee …

analysis ：

The above situation is called Unreal reading , Two concurrent transactions ,“ Business A： Get transaction B Records of consumption ”、“ Business B： Added a new consumption record ”, Business A Get transaction B There is one more piece of data in the consumption record .

scene ：

Teachers' A perform SQL Statement to change the grades of all students in the database from specific scores to ABCDE Hierarchy , The SQL After the statement is executed, before the transaction is committed , Teachers' B Insert a record of specific scores and commit the transaction , Teachers' A The transaction commits and executes the query SQL sentence , At this time, the teacher A I found that there was another record that had not been changed , It's like an illusion .

Summary ： It's easy to confuse unreal reading with unreal reading , Unrepeatable reading focuses on revision , Unreal reading focuses on adding or deleting . To solve the problem of non repeatable reading, we only need to lock the line that meets the condition , To solve unreal reading, we need to lock the watch .

Solution

According to the actual demand , By setting the transaction isolation level of the database, you can solve the dirty reads that occur when multiple transactions are concurrent 、 Non repetition and unreal reading , The database transaction isolation level is from low to high Read uncommitted、Read committed、Repeatable read and Serializable And so on . Databases are different , It supports different transaction isolation levels ：MySQL The database supports the above four transaction isolation levels , The default is Repeatable read;Oracle Database support Read committed and Serializable Two transaction isolation levels , The default is Read committed.
1、Read uncommitted（ Read uncommitted ）： Dirty reading possible 、 No repeated reading or phantom reading .
2、Read committed（ Read the submission ）： Avoid dirty reading , But there may be unrepeatable reading and unreal reading . The default level of most databases is Read committed, such as Sql Server Database and Oracle database . Be careful ： This isolation level will only lock the corresponding rows when writing data .
3、Repeatable read（ Repeated reading ）： It can avoid dirty reading and non repeatable reading , But there may be unreal reading . Be careful ：①、 When the transaction isolation level is repeatable , If the search condition has an index （ Include primary key index ） When , The default lock mode is next-key lock ;②、 If the search criteria are not indexed , When updating data, the entire table will be locked . A gap is locked by a transaction , Other transactions cannot insert records in this gap , This can prevent unreal reading .
4、Serializable（ serialize ）： Avoid dirty reading 、 No repeated reading or phantom reading , But the concurrency is very low , Rarely used in general . Be careful ： This isolation level will lock the whole table when reading and writing data .
summary ：

Transaction isolation level Dirty reading It can't be read repeatedly Fantasy reading
Read uncommitted（ Read uncommitted ） √ √ √
Read committed（ Read the submission ） × √ √
Repeatable read（ Repeated reading ） × × √
Serializable（ serialize ） × × ×

explain ：√ Indicates that there may be ,× It means that there will be no
Be careful ： Higher isolation level , The more data integrity and consistency can be guaranteed , But the greater the impact on concurrent performance .
MySQL Transaction isolation level
see ：MySQL Database support Read uncommitted、Read committed、Repeatable read and Serializable Four transaction isolation levels , The default is Repeatable read, You can view it through the following statement MySQL Database transaction isolation level ：
select @@global.tx_isolation,@@tx_isolation;

modify ：MySQL The modification of database transaction isolation level is divided into global modification and current session modify , The specific modification method is as follows ：
1、 Global modification
①、 stay my.ini Add the following configuration at the end of the configuration file ：

# The optional parameters are ：READ-UNCOMMITTED, READ-COMMITTED, REPEATABLE-READ, SERIALIZABLE.
[mysqld]
transaction-isolation = READ-UNCOMMITTED

②、 restart MySQL service
2、 At present session modify , Sign in MySQL Execute the following command after the database ：

set session transaction isolation level read uncommitted;