MySQL gap lock

Study Mysql, There will always be a mountain that can't go around , That's the lock . Theoretically , There are more patterns in the lock , It can't go beyond the categories mentioned in the operating system class , however Mysql The lock tipped me over .

stay Mysql The granularity of middle locks can be divided into ： Table lock , Row-level locks , Clearance lock Three . Table level lock and row level lock are not too difficult to understand . Only clearance lock, I can't understand its design intention accurately , And the phenomenon I've tested makes me feel weird .

So why is there a gap lock , According to most of the information available , Clearance lock is introduced to solve the problem of RR The problem of unreal reading at isolation level .

Here's an example , First create a Table:

Create Table: CREATE TABLE `foo` (
  `uid` int(11) NOT NULL,
  `age` int(11) NOT NULL,
  PRIMARY KEY (`uid`),
  KEY `age` (`age`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;
insert into foo values(1,1),(4,4),(7,7),(9,9);

then , Open two mysql client (M1,M2), The order of execution is as follows ：

M1: begin;
M1: select * from foo were uid > 1 and uid < 5 for update;
M2: begin;insert into foo values(2,2);commit;
M1: select * from foo were uid > 1 and uid < 5 for update;
M1: commit;

If in M1 First execution select Statement, only add line lock , Then all that's locked is uid=4 This business . stay M1 Second execution select When the sentence is , because M2 Insert a (2,2), So we will find one more （2,2) The record of . This will produce unreal reading .

mysql The solution is ： Use clearance lock , take uid In the gap interval （1,4),(4,7） All locks , So when M2 stay insert Row data （2,2） even to the extent that （6,6） Will be blocked by lock to prevent M1 Phantom reading .

If it comes to a perfect end , Then I won't roll over , Look at the other three sql sentence ：

M1: begin;
M1: select * from foo were age = 4 for update;
M2: begin;insert into foo values(6,6);commit;
M1: select * from foo were age = 4 for update;
M1: commit;

If you do it manually, you'll find M2 It will be blocked by a lock , It's because he's interested in age With a clearance lock （ Locks are added to the index ）.

because Locks are added to the index , According to my first reaction , Direct pair age=4 This index lock solves the problem , Why do you need a clearance lock ？

I checked for a long time , Just found a very important point that few people mentioned Primary key stored in secondary index , Will participate in secondary index sorting , such as age When the index is sorted , Actually, it's (age,uid) To sort . And the reason why I use uid I think most of the reasons for participating in the ranking should be B+ The same value is not allowed to be stored in the tree . Use age,uid After splicing, all secondary indexes can be guaranteed , stay B+ The value in the tree must be unique .

let me put it another way , We can't just lock age=4 This condition , Because there might be （age,uid）= (4,1)/(4,2)/(4,5) Wait for any index .

Secondary index in stitching , because age before uid After , therefore age To some extent, the value of represents the entire index value . That's why the clearance lock can be locked age=4 This condition .

In order to verify the correctness of the above statement , Let's look at this sql:

M1: begin;
M1: select * from foo were age = 4 for update;
M2: begin;update foo set age = 2 where uid = 1;commit;
M1: select * from foo were age = 4 for update;
M1: commit;

Let's start with a brief analysis ： 1. age Is not unique secondary index 2. The secondary index is implemented internally by age,uid After splicing, they participate in sorting 3. The gap is locked (age,uid) = (1,1) ~ (4,4) The open range of 4. M2 The statement executed is to insert a secondary index value (2,1)

According to the clearance lock principle , We can push out M2 It will be blocked by the clearance lock , And that's exactly what happened .

ps.  The primary key stored in the secondary index will participate in the secondary index sorting , I think this is very important . I don't know why many reference books have passed by intentionally or unintentionally .

------------------------------------------------------------------------------------------------------------------------------

summary ：

Mysql MVCC It's solved RR The readability of transactions , Using clearance lock solves RR Level unreal reading problem

Mysql The gap lock is for RR Level is introduced to solve the problem of unreal reading , The clearance lock is gap lock , and mysql It uses clearance lock and gap Lock combination , That is to say next-key lock, On different indexes ,mysql The way of locking is also different ：

On the unique index ： If the condition is =5 , Gap lock degenerates into row lock , That is, only the row of objects in the condition will be locked , If it is >5, A... Will be added [5, ∞) One of the next-key lock ,5 This row lock and (5,∞) This clearance lock

On the general index ： If the condition is 5, that mysql Will pass 5 Query a gap on the left and right , That is to say, bi 5 Small first value and ratio 5 The big first value , Then add a clearance lock , For example, the index value of two pieces of data in the database is 3 and 7, that mysql One will be added (3-5)[5](5-7] Such a gap lock , Why do ordinary indexes need to be added like this , That's because ordinary indexes can be repeated , Here is a previous sentence

Primary key stored in secondary index , Will participate in secondary index sorting , such as age When the index is sorted , Actually, it's (age,uid) To sort . And the reason why I use uid I think most of the reasons for participating in the ranking should be B+ The same value is not allowed to be stored in the tree . Use age,uid After splicing, all secondary indexes can be guaranteed , stay B+ The value in the tree must be unique .

let me put it another way , We can't just lock age=4 This condition , Because there might be （age,uid）= (4,1)/(4,2)/(4,5) Wait for any index .

Secondary index in stitching , because age before uid After , therefore age To some extent, the value of represents the entire index value . That's why the clearance lock can be locked age=4 This condition .

        The clearance lock is an interval that opens on the left and closes on the right , Take the example above , Equivalent query where c = 5, Then one will be added (3,7] A left open right close clearance lock , If we insert c=3 A record of is not blocked , But if we insert one c=7 The record of , It will block

No index ： because mysql All locks are on the index , If there is no index, the table lock is used