Do you still understand the deadlock handling methods in MySQL performance testing and tuning?

The following is from deadlock detection 、 Deadlock avoidance 、 Deadlock resolution 3 To explore how to MySQL Deadlock problem for performance tuning .

Deadlock detection

adopt SQL Statement to query lock table related information ：

（1） Query table opening

SHOW OPEN TABLES WHERE IN_USE> 0

（2） Query the lock status list

SELECT * FROM INFORMATION_SCHEMA.INNODB_LOCKS

（3） Query lock waiting information , among blocking_lock_id Is the transaction that the current transaction is waiting for

SELECT * FROM INFORMATION_SCHEMA.INNODB_LOCK_WAITS

（4） Query deadlock log

SHOW ENGINE INNODB STATUS

This statement can only display the latest deadlock , Unable to fully capture all deadlock information in the system .

If you want to log all deadlocks , Need to open innodb_print_all_deadlocks Parameters , Log all deadlocks to errorlog in .

（5） Query lock wait time

SHOW STATUS LIKE '%lock%'

Deadlock avoidance

When a deadlock occurs , It is usually due to lengthy transactions in the program of the project , Or because the isolation level is not set properly .

We need to pay attention to the following points in the use of transactions ：

（1） Try to keep things short , Commit the transaction immediately after making a series of associated update operations , To reduce the possibility of deadlock . Especially don't let the relevant MySQL Session suspends uncommitted transactions for a long time .

（2） A lower isolation level is recommended , Such as READ COMMITTED.

（3） Modify multiple tables in the same transaction , Or different rows in a table , Perform operations in the same order each time . In order to make the transaction form a clear lock operation queue and avoid deadlock .

Deadlock resolution

MySQL The database passed deadlock detection （innodb_deadlock_detect） And deadlock timeout （innodb_lock_wait_timeout） These two parameters are used for deadlock resolution .

Deadlock detection （innodb_deadlock_detect)： stay MySQL 8.0 in , Added a new dynamic variable innodb_deadlock_detect, Used to control InnoDB Whether to perform deadlock detection .

The default value for this parameter is ON, That is, open deadlock detection . After opening InnoDB When locking, it will detect whether it will cause deadlock after locking , If you lock it , Just roll back the transaction with the lowest cost .

Deadlock timeout （innodb_lock_wait_timeout）： This parameter can be used to handle undetectable deadlocks , Or avoid waiting for a long time for a long transaction .

For highly concurrent systems , When a large number of threads wait for the same lock , Deadlock detection may cause performance degradation .

here , If deadlock detection is disabled , Instead, it depends on parameters innodb_lock_wait_timeout It may be more efficient to release transactions that occupy lock resources for a long time .

in other words , When confirming that the deadlock detection function affects the performance of the system and disabling deadlock detection will not have a negative impact , You can try closing innodb_deadlock_detect Options .

in addition , If disabled InnoDB Deadlock detection , The parameters need to be adjusted in time innodb_lock_wait_timeout Value , To meet the actual needs . 

The author of this article ： Han Han becomes a god silently

Link to this article ：https://cloud.tencent.com/developer/column/92828