Deeply understand the mvcc mechanism of MySQL

undo log

undo log It's a rollback log , It does two things ： Provides rollback operations and Multiple line versioning (MVCC).

When the data is modified , Not only did it record redo, The corresponding undo, If the transaction fails or rolls back for some reason , With the help of this undo Roll back .

undo log and redo log Recording physical logs is not the same .undo log The main record is the logical change of data , It's a logical log .

Can be considered as delete When a record ,undo log A corresponding insert Record , vice versa , When update When a record , It records an opposite update Record .

When executed rollback when , You can start from undo log The logical record in reads the corresponding content and rolls back .

Multiline version control （MVCC） When , through undo log To achieve ：

When a read row is locked by another transaction , It can come from undo log Analyze the previous data of this line record , To provide the version information of this line , Let the user realize non lock consistent reading .

MySQL Transaction isolation level

except redo log and undo log outside , The other you are studying MVCC What I have to know before , Isolation level . Isolation level is quite basic knowledge , So let's review quickly here .

 

notes ：MySQL The default isolation level for is repeatable-read Level

And in MySQL in repeatable-read Level can also handle unreal reading , This is a MySQL Unique next-keylock Realized .

Different database vendors have different requirements for SQL standard The four isolation levels specified in support are different ：

Oracle Just support read-committed and serializable Isolation level .

MVCC Only in read-committed and repeatable-read Working at two isolation levels , The other two isolation levels ：

read-uncommitted, Always read the latest data lines , Instead of reading the data row of the current transaction version .

serializable, Will lock all read lines , and MVCC Are not compatible .

MVCC principle

Version chain

MySQL Each row of records of is actually a linked list in logic .

MySQL In addition to recording business data in line records , And hidden trx_id and roll_ptr

• trx_id： Represents the... Of the recently modified transaction id , Every time a transaction changes a clustered index record , Will take the matter Business id Assign a value to trx_id Hide columns . When adding a transaction ,trx_id It will increase , therefore trx_id Be able to indicate the order in which transactions start .
• roll_pointer： Point to the address of the previous version of the line , Every time a cluster index record is changed , Will write the old version to undo journal in , Then the hidden column is equivalent to a pointer , It can be used to find the information before the modification of the record .

update user set name = ' Yuanbao  2 ' where id = 1 

 

Of course , This linked list exists in undo log in , Different from the latest version of data .

After each update , Will put the old value in one undo log in , Even an old version of the record , As the number of updates increases , All versions will be roll_ptr Attributes are connected into a linked list , We call this list Version chain , The header node of the version chain is the latest value of the current record .

in addition , Each version is also included in the generated version Business id（trx_id）, This information is very important , We'll use... Later .

ReadView

That's it undo log, Say again ReadView. We said earlier ：MVCC Only in read-committed and repeatable-read Working at two isolation levels , and read-committed and repeatable-read The difference between them is that they generate ReadView The strategy is different .

For the use of read-committed and repeatable-read For isolation level transactions , You must ensure that you read the transaction modified records that have been committed , That is, if another transaction has modified the record but has not yet committed , You can't read the latest version of the record directly .

The core issue is ： We need to judge , Which version in the version chain is visible to the current transaction .

So ,InnoDB Came up with a ReadView The concept of , This ReadView There was a id list trx_ids To read and write transactions currently active in the storage system , That is to say begin Not yet commit or rollback The business of .

Continue to use the previous example to understand ReadView and trx_ids.

Submit trx_id yes 2 After recording , Then there's a trx_id by 3 The business of , modify name For Yuanbao 3, But the business has not yet been committed .

update user set name = ' Yuanbao  3 ' where id = 1 

Then the version chain is ︰

obviously , At this time trx_ids by [ 3 ]

 

If another transaction queries id by 1 The record of , because trx_ids Currently, there are only transactions trx_id by 3 The business of , and trx_ids The meaning of is to record unfinished transactions . ad locum , Transaction not completed , So this record is invisible , Continue to query — strip , The result returned to Yuanbao 2.

At this time, I put trx_id by 3 The transaction of has been submitted , And a new trx_id by 4 Also modify id by 1 The record of name= Yuanbao 4, And don't commit the transaction .

update user set name = ' Yuanbao  4 ' where id = 1 

At this point, the version chain is ︰

 

read-committed —— Before each query of data, a ReadView

trx_ids Update to [ 4 ], The version chain passes trx_id The result of comparison is Yuanbao 3.

repeatable-read —— Generate a ReadView, All subsequent reads reuse the previous ones .

There will be no reconstruction ReadView , trx_ids still [ 3 ],MySQL Think of affairs 3 Hang in the air , therefore select The result is Yuanbao 2. The first 2 Time select The result and the first 1 It's the same time , So it's called repeatable reading .

This article references from In depth understanding of MySQL Of MVCC Mechanism - cloud + Community - Tencent cloud (tencent.com)