记一次死锁排查

背景

业务系统在进行压测的时候，反馈账户系统部分请求未成功，查询日志发现出现了 dead lock，查询当前测试环境数据库事务隔离级别为 SERIALIZABLE

01 排查命令

1.1 查询锁信息

查询当前锁

Mysql8.0 之前使用：select * from information_schema.innodb_locks;

Mysql8.0 使用：select * from performance_schema.data_locks;

查询等待锁

Mysql8.0 之前使用：select * from information_schema.innodb_lock_waits;

Mysql8.0 使用：select * from performance_schema.data_lock_waits;

1.2 事务隔离级别

修改全局事务隔离级别

SET GLOBAL transaction_isolation = 'READ-COMMITTED';

修改当前会话事务隔离级别

SET transaction_isolation = 'READ-COMMITTED';

查询全局事务隔离级别

select @@global.tx_isolation;

查询事务隔离级别

select @@tx_isolation;

1.3 查询死锁信息

命令可以查询到最近一次死锁的日志

show engine innodb status\G;

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

LATEST DETECTED DEADLOCK

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

2024-04-14 15:23:17 0x7fe5d82ae700

*** (1) TRANSACTION:

TRANSACTION 1854, ACTIVE 37 sec starting index read

mysql tables in use 1, locked 1

LOCK WAIT 2 lock struct(s), heap size 1136, 1 row lock(s)

MySQL thread id 11, OS thread handle 140625150658304, query id 239 localhost root updating

UPDATE tb_account SET balance= 101 WHERE id = 1

*** (1) WAITING FOR THIS LOCK TO BE GRANTED:

RECORD LOCKS space id 26 page no 3 n bits 72 index PRIMARY of table `test`.`tb_account` trx id 1854 lock_mode X locks rec but not gap waiting

Record lock, heap no 2 PHYSICAL RECORD: n_fields 6; compact format; info bits 0

0: len 8; hex 8000000000000001; asc ;;

1: len 6; hex 00000000073d; asc =;;

2: len 7; hex ac000001200110; asc ;;

3: len 8; hex 8000000000000309; asc ;;

4: len 4; hex 80000001; asc ;;

5: len 10; hex 80000000002b67000000; asc +g ;;

*** (2) TRANSACTION:

TRANSACTION 1855, ACTIVE 21 sec starting index read

mysql tables in use 1, locked 1

4 lock struct(s), heap size 1136, 2 row lock(s)

MySQL thread id 13, OS thread handle 140625150928640, query id 240 localhost root updating

UPDATE tb_account SET balance= 101 WHERE id = 1

*** (2) HOLDS THE LOCK(S):

RECORD LOCKS space id 26 page no 3 n bits 72 index PRIMARY of table `test`.`tb_account` trx id 1855 lock mode S locks rec but not gap

Record lock, heap no 2 PHYSICAL RECORD: n_fields 6; compact format; info bits 0

0: len 8; hex 8000000000000001; asc ;;

1: len 6; hex 00000000073d; asc =;;

2: len 7; hex ac000001200110; asc ;;

3: len 8; hex 8000000000000309; asc ;;

4: len 4; hex 80000001; asc ;;

5: len 10; hex 80000000002b67000000; asc +g ;;

*** (2) WAITING FOR THIS LOCK TO BE GRANTED:

RECORD LOCKS space id 26 page no 3 n bits 72 index PRIMARY of table `test`.`tb_account` trx id 1855 lock_mode X locks rec but not gap waiting

Record lock, heap no 2 PHYSICAL RECORD: n_fields 6; compact format; info bits 0

0: len 8; hex 8000000000000001; asc ;;

1: len 6; hex 00000000073d; asc =;;

2: len 7; hex ac000001200110; asc ;;

3: len 8; hex 8000000000000309; asc ;;

4: len 4; hex 80000001; asc ;;

5: len 10; hex 80000000002b67000000; asc +g ;;

*** WE ROLL BACK TRANSACTION (1)

lock_mode X locks rec but not gap waiting Record lock 表示这是一个 X 记录锁
HOLDS THE LOCK(S): 表示当前持有的锁
WAITING FOR THIS LOCK TO BE GRANTED: 表示当前需要获取的锁

03 复现

3.1 修改当前事务隔离级别

SET transaction_isolation = 'SERIALIZABLE';

3.2 初始化数据

创建表

CREATE TABLE `tb_account` (

`id` bigint(20) NOT NULL,

`user_id` bigint(20) NOT NULL,

`account_type` int(11) NOT NULL,

`balance` decimal(20,6) NOT NULL,

PRIMARY KEY (`id`),

UNIQUE KEY `idx_user_id_account_type` (`user_id`,`account_type`)

) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_unicode_ci;

插入数据

INSERT INTO `tb_coin_account95` (`id`, `user_id`, `account_type`, `balance`, `create_time`, `last_update_time`, `version`, `expire_at`, `freeze_amount`)

VALUES

(743576, 1239095, 32, 11111330.000000, '2022-09-22 02:46:45', '2024-04-12 16:30:21', 182, 0, 313.000000);

3.3 执行语句

1. 开始两个事务并且同时执行查询语句

事务1

begin;

select * from tb_account where id = 1;

事务2

begin;

select * from tb_account where id = 1;

2. 在两个事务上分别执行

UPDATE tb_account SET balance= 101 WHERE id = 1;

3. 执行结果如下

事务1:

事务2:

可以看到出现了死锁

4. 执行 show engine innodb status\G; 命令查询死锁信息

04 思考

1. 对于一个事务想要获取到某条记录的 X 锁，需要其它事务没有持有这条记录的 S 锁或等待其它事务释放 S 锁。

2. SERIALIZABLE 这种隔离级别普通的查询也会获取记录的 S 锁。

3. S 锁可以重复获取，但事务 A 等待事务 B 释放 S 锁，事务 B 等待事务 A 释放 S 锁，由于循环等待造成了死锁，而正常的读已提交读的是视图，是不会获取 S 锁的