MySQL deadlock analysis

1. What is a deadlock

Deadlock refers to that multiple transactions hold locks required by each other , And they are waiting for each other to release the lock they need , Thus, multiple transactions have been waiting .

2. Why is a deadlock formed

1. Row locks of different table records conflict ：
   Multiple in a transaction sql When operating the records in different tables separately, it is the same as the records in these tables of other transaction operations , Circular waiting lock condition occurs . for example ： Business A Delete first table_1 In the table id=1 The record of , Business B Update again table_2 In the table id=2 The record of , Business A Update again table_2 In the table id=2 The record of , Business B And then delete table_1 In the table id=1 The record of .
   
2. Row lock conflict of the same table record ：
   The same records in a table are processed simultaneously in different transactions . for example ： Business A Handle id by (1,2,3,4) The record of , Business B Handle id by (2,1,5,6) The record of .
   
3. Different index lock conflicts ：
   Different index records are processed simultaneously in different transactions . for example ： Business A When performing an ordinary index operation , The locking order of this general index corresponding to multiple clustered indexes is （1,2,3）, And the business B When performing cluster index operation , The corresponding locking sequence is （2,1,4）, This creates the possibility of deadlock .
   
4. gap Lock conflict ：
   innodb stay RR Below grade , If different transactions contain the same interval , Deadlock will be caused by the overlapping range of the generated gap lock . for example ： Business A The interval of cluster index execution is ：（1,5）, Business B The interval of cluster index execution is ：（3,8）. When a transaction A Execute to 4 Time belongs to transaction B Clearance lock range , Need to wait for a transaction B Release the clearance lock , When a transaction B Execute to 4 when , It also belongs to affairs A Clearance lock range , You also need to wait for transactions A Release the clearance lock , Then it causes these two transactions to wait for each other to release the lock , So it's a deadlock .
   

3. How to deal with deadlock

1. Set the wait time （innodb_lock_wait_timeout=50s）： When a waiting time exceeds a set threshold , Roll back one of the transactions , Another transaction can continue .（ shortcoming ： If the rolled back transaction updates many rows , It takes up a lot of undo log, Then it may take more time to rollback than another normally executed transaction , It's not suitable ）
2. Initiate deadlock detection ：innodb It also provides wait-for graph Algorithm to actively detect deadlock , The database is required to store lock information linked list and transaction waiting linked list , Through these two parts of information, a graph is constructed , When each transaction requests a lock and a wait occurs, it determines whether there is a loop , If a circuit is detected in the diagram , It indicates that there is a deadlock , Now InnoDB The storage engine will choose to rollback undo The smallest amount of transactions .

4. How to avoid deadlock

1. Access tables and rows in a fixed order , Avoid cross execution of the same records by different transactions .
2. Big business, small business . Big business tends to deadlock , If business permits , Break up big business into small ones .
3. In the same transaction , Try to lock all the resources you need at once , Reduce deadlock probability .
4. Reduce isolation level . If business permits , Lowering the isolation level is also a good choice , Let's take the isolation level from RR Adjusted for RC, A lot can be avoided because gap A deadlock caused by a lock .
5. Add a reasonable index to the table . You can see that if you do not index, you will add locks for each row of the table , The probability of deadlock is greatly increased .
6. Commit the transaction immediately after the operation , Especially on the interactive command line .
7. With a transaction , Don't use lock tables .
8. It's better not to use it (SELECT … FOR UPDATE or SELECT … LOCK IN SHARE MODE).