最近项目中某个模块稳定复现MySQL死锁问题,本文记录死锁的发生原因以及解决办法。
1. 预备知识
1.1 表锁和行锁
- 表锁
表锁是MySQL中最基本的锁策略,并且是开销最小的策略。表锁会锁定整张数据表,用户的写操作(插入/删除/更新)前,都需要获取写锁(写锁会相互阻塞);没有写锁时,读取用户才能获取读锁(读锁不会相互阻塞)。
- 行锁(仅限定于InnoDB)
行级锁可以最大程度的支持并发处理(同时也带来了最大的锁开销)。行级锁只在存储引擎实现,而MySQL服务器层没有实现。服务器层完全不了解存储引擎中的具体实现。
1.2 行锁简介
行锁的模式有:读/写意向锁(IS/IX锁),读锁(S锁),写锁(X锁)以及自增锁(AI)。
行锁根据场景的不同又可以进一步细分,依次为Next-Key Lock,Gap Lock间隙锁,Record Lock记录锁和插入意向GAP锁。不同的锁锁定的位置是不同的,比如说记录锁只锁住对应的记录,而间隙锁锁住记录和记录之间的间隔,Next-Key Lock锁住记录和记录之前的间隙。不同类型锁的锁定范围大致如下图所示。
此外,锁对应的死锁日志信息标记如下所示:
- 记录锁(LOCK_REC_NOT_GAP): lock_mode X locks rec but not gap
- 间隙锁(LOCK_GAP): lock_mode X locks gap before rec
- Next-key锁(LOCK_ORNIDARY): lock_mode X
- 插入意向锁(LOCK_INSERT_INTENTION): lock_mode X locks gap before rec insert intention
1.3 行锁加锁示例
InnoDB是聚簇索引,也就是B+树的叶子节点存储了主键索引以及数据行;InnoDB的二级索引的叶子节点存储的则是主键值,所以通过二级索引查询数据时,需要根据查询到的主键去聚簇索引中再次进行查询。
update user set age = 10 where id = 49;
update user set age = 10 where name = "Tom";
(1)第一条SQL使用主键进行查询,则只需要在id=49
主键上加上写锁(X锁);
(2)第二条SQL使用二级索引查询,首先在name="Tom"
上加写锁,然后根据获取的主键索引查询,在id=49
主键上添加写锁。
具体如下图所示:
以上是基于单条数据讨论,针对多条数据:
update user set age = 10 where id > 49;
执行步骤:
(1)MySQL Server根据where条件读取满足条件的第一条记录,InnoDB引擎返回行记录并加锁;
(2)MySQL Server发起更新行记录的update请求,更新此记录;
(3)反复循环(1)(2)步骤,直到所有满足条件的记录均被修改。
具体如下图所示:
2.准备工作
2.1 创建数据表并初始化
create table dead_lock_test
(
id int auto_increment
primary key,
v1 int not null,
v2 int not null
);
insert into dead_lock_test (v1,v2) value (1,1);
insert into dead_lock_test (v1,v2) value (2,2);
insert into dead_lock_test (v1,v2) value (3,3);
需要注意,数据表中仅存在主键索引。此外,默认数据库引擎为InnoDB,事务隔离级别为RR(可重复读,相对于RC解决了幻读)。
2.2 开启锁监控
使用如下语句,开启MySQL锁监控:
# 开启
set GLOBAL innodb_status_output=ON;
set GLOBAL innodb_status_output_locks=ON;
# 关闭
set GLOBAL innodb_status_output_locks=OFF;
3.场景复现
开启两个数据库连接,分别执行如下SQL语句:
# session1
start transaction ;
insert into dead_lock_test (v1,v2) value (4,4);
delete from dead_lock_test where v1 = 4 and v2 = 4;
commit;
# session2
start transaction;
insert into dead_lock_test (v1,v2) value (5,5);
delete from dead_lock_test where v1 = 5 and v2 = 5;
commit;
不要问事务里就两条SQL,插入后删除走回滚就可以了之类的问题(我也不知道为什么这么写的)。
事务执行步骤如下表所示:
session1 | session2 | stage | 备注 |
---|---|---|---|
start transaction; |
start transaction; |
||
insert into dead_lock_test (v1,v2) value (4,4); |
do nothing | 执行成功 | |
do nothing | insert into dead_lock_test (v1,v2) value (5,5); |
stage1 | 执行成功 |
delete from dead_lock_test where v1 = 4 and v2 = 4; |
do nothing | stage2 | session1执行结果阻塞 |
do nothing | delete from dead_lock_test where v1 = 5 and v2 = 5; |
stage3 | session2执行结果报deadlock |
3.1 stage1
执行show engine innodb status;
节选事务信息如下所示:
------------
TRANSACTIONS
------------
Trx id counter 91328
Purge done for trx"s n:o < 91327 undo n:o < 0 state: running but idle
History list length 19
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 91327, ACTIVE 37 sec
1 lock struct(s), heap size 1136, 0 row lock(s), undo log entries 1
MySQL thread id 24, OS thread handle 15668, query id 3147 localhost 127.0.0.1 root
TABLE LOCK table `igw_proxy_rule_management`.`dead_lock_test` trx id 91327 lock mode IX
---TRANSACTION 91322, ACTIVE 44 sec
1 lock struct(s), heap size 1136, 0 row lock(s), undo log entries 1
MySQL thread id 23, OS thread handle 22788, query id 3103 localhost 127.0.0.1 root
TABLE LOCK table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock mode IX
输出内容中节选当前事务信息,当前存在两个运行中事务,trx id分别为91322以及91327。
TABLE LOCK table
igw_proxy_rule_management.
dead_lock_testtrx id 91322 lock mode IX
:dead_lock_test
表上添加IX锁。
91322事务对应session1,91327事务对应session2.
3.2 stage2
执行delete from dead_lock_test where v1 = 4 and v2 = 4;
后可发现,当前事务被阻塞。
执行show engine innodb status;
节选事务信息如下所示:
因为输出内容较多,直接在输出内容中添加注释进行解析。
------------
TRANSACTIONS
------------
Trx id counter 91332
Purge done for trx"s n:o < 91332 undo n:o < 0 state: running but idle
History list length 21
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 91327, ACTIVE 58 sec
* 2 lock strcut(s): 事务91327中锁链表长度为2(每个链表节点表示该事务持有的一个锁结构,包括表锁/记录锁等),当前事务包含表锁(IX)以及一个行锁(记录锁);
* 1 row lock(s):当前事务持有的行锁个数;
* undo log entries 1:当前事务的undo log个数
2 lock struct(s), heap size 1136, 1 row lock(s), undo log entries 1
MySQL thread id 24, OS thread handle 15668, query id 3147 localhost 127.0.0.1 root
* TABLE LOCK:当前事务持有的表锁(IX)
TABLE LOCK table `igw_proxy_rule_management`.`dead_lock_test` trx id 91327 lock mode IX
* RECORD LOCKS:当前事务持有的行锁(lock_mode X locks rec but not gap)
* space id 92: dead_lock_test所在空间编号
* page no 4: 当前记录所在页码
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91327 lock_mode X locks rec but not gap
* 行锁信息: heap no=6
Record lock, heap no 6 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000005; asc ;; * hex 80000005:当前加锁的记录id=5
1: len 6; hex 0000000164bf; asc d ;; * hex 0000000164bf: 事务ID;
2: len 7; hex 81000000b20110; asc ;; * hex 81000000b20110: 回滚指针;
3: len 4; hex 80000005; asc ;; * hex 80000005: v1字段对应数值;
4: len 4; hex 80000005; asc ;; * hex 80000005:v2字段对应数值;
---TRANSACTION 91322, ACTIVE 65 sec fetching rows
* tables in use 1: 有1个表正在被使用;
* locked 1: 有一个表锁
mysql tables in use 1, locked 1
* LOCK WAIT:事务91322处于锁等待状态;其他字段解释详见上问
LOCK WAIT 5 lock struct(s), heap size 1136, 6 row lock(s), undo log entries 2
MySQL thread id 23, OS thread handle 22788, query id 3199 localhost 127.0.0.1 root updating
* 事务91322当前执行SQL语句
delete from dead_lock_test where v1 = 4 and v2 = 4
* 事务91322等待的锁信息
------- TRX HAS BEEN WAITING 8 SEC FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X waiting
* 事务91322等待的记录锁(锁对应记录主键为5,被事务91327持有)
Record lock, heap no 6 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000005; asc ;;
1: len 6; hex 0000000164bf; asc d ;;
2: len 7; hex 81000000b20110; asc ;;
3: len 4; hex 80000005; asc ;;
4: len 4; hex 80000005; asc ;;
------------------
* 以下展示事务91322所持有的锁以及尝试获取的锁,首先是表意向锁(IX锁)
TABLE LOCK table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock mode IX
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X
* 记录锁(锁对应记录主键为1)
Record lock, heap no 2 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000001; asc ;;
1: len 6; hex 0000000164b3; asc d ;;
2: len 7; hex 81000000ad0110; asc ;;
3: len 4; hex 80000001; asc ;;
4: len 4; hex 80000001; asc ;;
* 记录锁(锁对应记录主键为2)
Record lock, heap no 3 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000002; asc ;;
1: len 6; hex 0000000164b4; asc d ;;
2: len 7; hex 82000000ad0110; asc ;;
3: len 4; hex 80000002; asc ;;
4: len 4; hex 80000002; asc ;;
* 记录锁(锁对应记录主键为3)
Record lock, heap no 4 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000003; asc ;;
1: len 6; hex 0000000164b9; asc d ;;
2: len 7; hex 81000000b00110; asc ;;
3: len 4; hex 80000003; asc ;;
4: len 4; hex 80000003; asc ;;
* 记录锁:锁定记录(添加记录时创建的锁)
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X locks rec but not gap
Record lock, heap no 5 PHYSICAL RECORD: n_fields 5; compact format; info bits 32
0: len 4; hex 80000004; asc ;;
1: len 6; hex 0000000164ba; asc d ;;
2: len 7; hex 020000011a03cb; asc ;;
3: len 4; hex 80000004; asc ;;
4: len 4; hex 80000004; asc ;;
* 间隙锁:锁定记录(删除记录时创建的锁,在RR模式下生效,主要解决幻读)
* 需要注意,InnoDB的删除记录不是物理删除,而是标记删除(等待后续记录覆盖),因此可理解删除类似于更新操作
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X locks gap before rec
Record lock, heap no 5 PHYSICAL RECORD: n_fields 5; compact format; info bits 32
0: len 4; hex 80000004; asc ;;
1: len 6; hex 0000000164ba; asc d ;;
2: len 7; hex 020000011a03cb; asc ;;
3: len 4; hex 80000004; asc ;;
4: len 4; hex 80000004; asc ;;
* 事务91322尝试获取的锁(被事务91327持有)
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X waiting
Record lock, heap no 6 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000005; asc ;;
1: len 6; hex 0000000164bf; asc d ;;
2: len 7; hex 81000000b20110; asc ;;
3: len 4; hex 80000005; asc ;;
4: len 4; hex 80000005; asc ;;
(1)由以上注释可知,事务91322在尝试删除时,会对表中所有记录添加记录锁。
这是因为当前删除记录条件为v1 = 4 and v2 = 4
,在v1与v2字段上,并未建立相应的索引。
因为无法通过索引确定主键,导致MySQL Server会先尝试锁定当前dead_lock_test
表中所有记录添加记录锁(可以设置参数进行优化,根据where条件逐渐解除不满足条件记录上的记录锁)。
(2)事务91322尝试对dead_lock_test表中所有记录添加锁,发现记录(id=5)已经被事务91327添加记录锁,导致事务91322只能等待事务91327放弃记录锁。
3.3 stage3
执行delete from dead_lock_test where v1 = 5 and v2 = 5;
后即可发现终端输出:
[2021-05-13 15:33:29] [40001][1213] Deadlock found when trying to get lock; try restarting transaction
执行show engine innodb status;
节选死锁信息如下所示:
因为内容较多,不再列出解释,详见输出信息中文注释部分
------------------------
LATEST DETECTED DEADLOCK
------------------------
2021-05-13 17:27:09 0xca4
*** (1) TRANSACTION:
* 事务91322持有锁情况,在stage2已经详细解释,此处不再赘述
TRANSACTION 91322, ACTIVE 78 sec fetching rows
mysql tables in use 1, locked 1
LOCK WAIT 5 lock struct(s), heap size 1136, 6 row lock(s), undo log entries 2
MySQL thread id 23, OS thread handle 22788, query id 3199 localhost 127.0.0.1 root updating
delete from dead_lock_test where v1 = 4 and v2 = 4
*** (1) HOLDS THE LOCK(S):
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X
Record lock, heap no 2 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000001; asc ;;
1: len 6; hex 0000000164b3; asc d ;;
2: len 7; hex 81000000ad0110; asc ;;
3: len 4; hex 80000001; asc ;;
4: len 4; hex 80000001; asc ;;
Record lock, heap no 3 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000002; asc ;;
1: len 6; hex 0000000164b4; asc d ;;
2: len 7; hex 82000000ad0110; asc ;;
3: len 4; hex 80000002; asc ;;
4: len 4; hex 80000002; asc ;;
Record lock, heap no 4 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000003; asc ;;
1: len 6; hex 0000000164b9; asc d ;;
2: len 7; hex 81000000b00110; asc ;;
3: len 4; hex 80000003; asc ;;
4: len 4; hex 80000003; asc ;;
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91322 lock_mode X waiting
Record lock, heap no 6 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000005; asc ;;
1: len 6; hex 0000000164bf; asc d ;;
2: len 7; hex 81000000b20110; asc ;;
3: len 4; hex 80000005; asc ;;
4: len 4; hex 80000005; asc ;;
*** (2) TRANSACTION:
TRANSACTION 91327, ACTIVE 71 sec starting index read
mysql tables in use 1, locked 1
LOCK WAIT 3 lock struct(s), heap size 1136, 2 row lock(s), undo log entries 1
MySQL thread id 24, OS thread handle 15668, query id 3237 localhost 127.0.0.1 root updating
delete from dead_lock_test where v1 = 5 and v2 = 5
*** (2) HOLDS THE LOCK(S):
* 事务91327持有记录(id=5)的记录锁,此锁正在被事务91322等待持有
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91327 lock_mode X locks rec but not gap
Record lock, heap no 6 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000005; asc ;;
1: len 6; hex 0000000164bf; asc d ;;
2: len 7; hex 81000000b20110; asc ;;
3: len 4; hex 80000005; asc ;;
4: len 4; hex 80000005; asc ;;
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
* 事务91327等待持有记录锁,锁信息见后续解释
RECORD LOCKS space id 92 page no 4 n bits 72 index PRIMARY of table `igw_proxy_rule_management`.`dead_lock_test` trx id 91327 lock_mode X waiting
* 事务91327等待持有记录(id=1)的记录锁(delete无法走索引查询,因此会尝试对所有表记录进行加锁,但是事务91322持有id=1/2/3/4的记录锁,死锁条件构成)
Record lock, heap no 2 PHYSICAL RECORD: n_fields 5; compact format; info bits 0
0: len 4; hex 80000001; asc ;;
1: len 6; hex 0000000164b3; asc d ;;
2: len 7; hex 81000000ad0110; asc ;;
3: len 4; hex 80000001; asc ;;
4: len 4; hex 80000001; asc ;;
*** WE ROLL BACK TRANSACTION (2)
由上可知:
(1)事务91322执行删除操作时,尝试获取表中所有记录的记录锁,其中记录(id=5)的锁被事务91327持有;
(2)事务91327执行删除操作时,尝试获取表中所有记录的记录锁,发现记录(id=1/2/3/4)的锁被事务91322持有;
(3)至此,事务91322与事务91327构成互相等待,死锁形成。
解决方案
4.1 添加索引
由上述的分析可知,删除时因为where条件无法利用索引,导致MySQL会尝试对表中所有记录加记录锁,产生死锁。
我们仅需在v1以及v2字段上建立联合索引,缩小记录冲突范围。
create index dead_lock_test_v2_v1_index on dead_lock_test (v1, v2);
此处没有设置唯一索引,如果多个事务根据索引查询,锁定的记录存在重叠,也容易复现死锁现象。
不过当前业务侧的数据插入,可保证在短暂时间范围内,不存在重叠记录,且表中存在一些重复数据,因此不使用唯一索引。
4.2 最终
表中添加索引。事务中添加后再删除,通过回滚实现。
参考文章
把MySQL中的各种锁及其原理都画出来
MySQL锁系列(二)之 锁解读