转载

【MySQL】InnoDB锁机制之一

一背景
MySQL锁机制是一个极其复杂的实现，为数据库并发访问和数据一致提供保障。这里仅仅针对MySQL访问数据的三种锁做介绍，加深自己对锁方面的掌握。
二常见的锁机制
我们知道对于InnoDB存储引擎而言,MySQL 的行锁机制是通过在索引上加锁来锁定要目标数据行的。常见的有如下三种锁类型，本文未声明情况下都是在RR 事务隔离级别下的描述。
2.1 Record Locks
记录锁实际上是索引上的锁，锁定具体的一行或者多行记录。当表上没有创建索引时，InnoDB会创建一个隐含的聚族索引，并且使用该索引锁定数据。通常我们可以使用 show innodb status 看到行锁相关的信息。
2.2 Gap Locks
间隙锁是锁定具体的范围，但是不包含行锁本身。比如

select * from tab where id>10 and id<20；

RR事务隔离级别下会锁定10-20之间的记录，不允许类似15这样的值插入到表里，以便消除“幻读”带来的影响。间隙锁的跨度可以是1条记录(Record low就可以认为是一个特殊的间隙锁，多行，或者为空。当访问的字段是唯一键/主键时，间隙锁会降级为Record lock。RR事务隔离级别下访问一个空行，也会有间隙锁，后续会举例子说明。
我们可以通过将事务隔离级别调整为RC 模式或者设置innodb_locks_unsafe_for_binlog=1 (该参数已经废弃)来禁用Gap锁。

2.3 Next-Key Locks
是Record Lock+Gap Locks,锁定一个范围并且包含索引本身。例如索引值包含 2，4，9，14 四个值，其gap锁的区间如下：
（-∞,2]，(2,4]，(4,9]，(9,14]，(14,+∞）
本文着重从主键,唯一键、非唯一索引,不存在值访问四个方面来阐述RR模式下锁的表现。
三测试案例
3.1 主键/唯一键

CREATE TABLE `lck_primarkey` (
`id` int(11) NOT NULL,
val int(11) not null default 0,
primary key (`id`),
key idx_val(val)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;
insert into lck_primarkey values(2,3),(4,5),(9,8),(14,13)

会话1

[session1] >select * from lck_primarkey;
+----+-----+
| id | val |
+----+-----+
| 2 | 3 |
| 4 | 5 |
| 9 | 8 |
| 14 | 13 |
+----+-----+
4 rows in set (0.00 sec)
[session1] >begin;
Query OK, 0 rows affected (0.00 sec)
[session1] >select * from lck_primarkey where id=9 for update;
+----+-----+
| id | val |
+----+-----+
| 9 | 8 |
+----+-----+
1 row in set (0.00 sec)

会话2

[session2] >begin;
Query OK, 0 rows affected (0.00 sec)
[session2] >insert into lck_primarkey values(7,6);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_primarkey values(5,5);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_primarkey values(13,13);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_primarkey values(10,9);
Query OK, 1 row affected (0.00 sec)

分析
从例子看，当访问表的where字段是主键或者唯一键的时候，session2中的插入操作并未被 session1 中的id=8 影响。官方表述

“Gap locking is not needed for statements that lock rows using a unique index to search for a unique row. (This does not include the case that the search condition includes only some columns of a multiple-column unique index; in that case, gap locking does occur.) For example, if the id column has a unique index, the following statement uses only an index-record lock for the row having id value 100 and it does not matter whether other sessions insert rows in the preceding gap:
select * from tab where id=100 for update”
就是说当语句通过主键或者唯一键访问数据的时候，Innodb会使用Record lock锁住记录本身，而不是使用间隙锁锁定范围。

需要注意以下两种情况:
1 通过主键或则唯一索引访问不存在的值，也会产生GAP锁。

[session1] >begin;
Query OK, 0 rows affected (0.00 sec)
[session1] >select * from lck_primarkey where id=7 for update;
Empty set (0.00 sec)
[session2] >insert into lck_primarkey values(8,13);
^CCtrl-C -- sending "KILL QUERY 303042481" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_primarkey values(5,13);
^CCtrl-C -- sending "KILL QUERY 303042481" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_primarkey values(3,13);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_primarkey values(10,13);
Query OK, 1 row affected (0.00 sec)

2 通过唯一索引中的一部分字段来访问数据，比如unique key(a,b,c) ，select * from tab where a=x and b=y; 读者朋友可以自己做这个例子。

3.2 非唯一键

CREATE TABLE `lck_secondkey` (
`id` int(11) NOT NULL,
KEY `idx_id` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=utf8;
insert into lck_secondkey values(2),(4),(9),(14)

会话1

[session1] >begin ;
Query OK, 0 rows affected (0.00 sec)
[session1] >select * from lck_secondkey;
+----+
| id |
+----+
| 2 |
| 3 |
| 4 |
| 9 |
| 14 |
+----+
5 rows in set (0.00 sec)
[session1] >select * from lck_secondkey where id=9 for update;
+----+
| id |
+----+
| 9 |
+----+
1 row in set (0.00 sec)

会话2

[session2] >begin;
Query OK, 0 rows affected (0.00 sec)
[session2] >insert into lck_secondkey values(3);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_secondkey values(4);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(5);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(6);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(7);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(8);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(9);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(10);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(11);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(12);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(13);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_secondkey values(14);
Query OK, 1 row affected (0.00 sec)

分析
事务1 对id=9进行for update 访问,session2 插入[4,13]的值都是失败的。根据MySQL的锁原理，Innodb 范围索引或者表是通过Next-key locks 算法，RR事务隔离级别下，通过非唯一索引访问数据行并不是锁定唯一的行，而是一个范围。从例子上可以看出来MySQL对 [4,9] 和(9,14]之间的记录加上了锁，防止其他事务对4-14范围中的值进行修改。可能有读者对其中 id=4 不能修改，但是id=14的值去可以插入有疑问？可以看接下来的例子

[session1] >select * from lck_primarkey;
+----+-----+
| id | val |
+----+-----+
| 2 | 3 |
| 4 | 5 |
| 9 | 8 |
| 14 | 13 |
+----+-----+
4 rows in set (0.00 sec)
[session1] >begin;
Query OK, 0 rows affected (0.00 sec)
[session1] >select * from lck_primarkey where val=8 for update;
+----+-----+
| id | val |
+----+-----+
| 9 | 8 |
+----+-----+
1 row in set (0.00 sec)

会话2

[session2] >begin;
Query OK, 0 rows affected (0.00 sec)
[session2] >insert into lck_primarkey values(3,5);
Query OK, 1 row affected (0.00 sec)
[session2] >insert into lck_primarkey values(15,13);
Query OK, 1 row affected (0.00 sec)
[session2] >select * from lck_primarkey;
+----+-----+
| id | val |
+----+-----+
| 2 | 3 |
| 3 | 5 |
| 4 | 5 |
| 9 | 8 |
| 14 | 13 |
| 15 | 13 |
+----+-----+
6 rows in set (0.00 sec)
[session2] >insert into lck_primarkey values(16,12);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_primarkey values(16,6);
^CCtrl-C -- sending "KILL QUERY 303040567" to server ...
Ctrl-C -- query aborted.
ERROR 1317 (70100): Query execution was interrupted
[session2] >insert into lck_primarkey values(16,5);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
[session2] >
[session2] >insert into lck_primarkey values(1,5);
Query OK, 1 row affected (0.00 sec)

分析
因为session1 对非唯一键val=8 加上了gap锁 [4,5] -[14,13],非此区间的记录都可以插入表中。记录(1,5),(15,13)不在此gap锁区间，记录(16,12),(16,6),(16,5)中的val值在被锁的范围内，故不能插入。
四总结
写本文的目的主要是在于温故而知新，侧重于温故。本文着重介绍了三种锁，其实还有两种锁Insert Intention Locks和AUTO-INC Locks 留作后面继续分析。
五推荐资料
1 官方资料
2 Innodb锁机制：Next-Key Lock 浅谈
3 MySQL 加锁处理分析

正文到此结束