【MySQL】gap lock 浅析
【定义】
innodb 行级锁 record-level lock大致有三种：record lock, gap lock and Next-KeyLocks。
record lock 锁住某一行记录
gap lock 锁住某一段范围中的记录
next key lock 是前两者效果的叠加。
下面是MYSQL官方文档中相关内容的链接
http://dev.mysql.com/doc/refman/5.1/en/innodb-record-level-locks.html
【实验环境】
session 1 20:39:29> show create table gap G
***************** 1. row *****************
Table: gap
Create Table: CREATE TABLE gap (
id int(11) DEFAULT NULL,
KEY ind_gap_id (id)
) ENGINE=InnoDB DEFAULT CHARSET=utf8
1 row in set (0.00 sec)
insert into gap values(17);
insert into gap values(20);
insert into gap values(33);
insert into gap values(39);
insert into gap values(42);
insert into gap values(43);
session 1 20:39:32> select * from gap;
+——+
| id |
+——+
| 17 |
| 20 |
| 33 |
| 39 |
| 42 |
| 43 |
+——+
6 rows in set (0.00 sec)
【实验】
两个会话都在REPEATABLE-READ 事务隔离级别。且都要在事务中进行。
session 1 20:39:37> start transaction;
Query OK, 0 rows affected (0.00 sec)
session 1 20:39:41> delete from gap where id=33;
Query OK, 1 row affected (0.00 sec)
session 20:40:07>
在会话2中 插入id <20 和 >=39的值 可以执行成功,而当要插入的id [20,39)的值时 会遇到gap lock 。
session 2 20:40:15> start transaction;
Query OK, 0 rows affected (0.00 sec)
session 2 20:40:30> insert into gap values(14);
Query OK, 1 row affected (0.00 sec)
session 2 20:40:59> insert into gap values(18);
Query OK, 1 row affected (0.00 sec)
session 2 20:41:06> insert into gap values(20);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
session 2 20:41:12> insert into gap values(24);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
session 2 20:42:17>
session 2 20:42:53> insert into gap values(35);
ERROR 1205 (HY000): Lock wait timeout exceeded; try restarting transaction
session 2 20:44:09>
session 2 20:44:56> insert into gap values(39);
Query OK, 1 row affected (0.00 sec)
session 2 20:45:13> insert into gap values(40);
Query OK, 1 row affected (0.00 sec)
从上面的实验中可以看出会话1 执行删除语句之后,不仅仅锁住 id=33的记录,同时也锁住区间为[20,39)的记录。具体的原因是执行delete from gap where id=33语句,mysql 会执行索引扫描并在该表上施加一个next-key
lock ,向左扫描到20,向右扫描到39 ,锁定区间左闭右开,所以lock的范围是 [20,39)。
【gap 锁带来的问题】
生产环境中有这样的一个情况：
程序会对一个表message 进行update 和insert
session 1
UPDATE message SET gmt_modified = now(),deal_times = deal_times +1 , status = ‘sending’ , gmt_retry = ‘2012-11-17 23:54:10’
WHERE message_id=18;
insert into message (body ,user_id,status,message_type,version,deal_times,gmt_create,gmt_modified,gmt_retry)
values (‘hello ！’,-1,’sending’,’instance_status_sync’,2,127,now(),now(),now());
session 2
UPDATE message SET gmt_modified = now(),deal_times = deal_times +1 , status = ‘sending’ , gmt_retry = ‘2012-11-17 23:54:10’
WHERE message_id=19;
insert into message (body ,user_id,status,message_type,version,deal_times,gmt_create,gmt_modified,gmt_retry)
values (‘hello world！’,-2,’sending’,’instance_status_sync’,1,17,now(),now(),now());
对于上述程序在无并发情况下,运行正常,但是并发量大的情况下,执行顺序可能就会变成下面的：
UPDATE message SET gmt_modified = now(),deal_times = deal_times +1 , status = ‘sending’ , gmt_retry = ‘2012-11-17 23:54:10’
WHERE message_id= 61;
UPDATE message SET gmt_modified = now(),deal_times = deal_times +1 , status = ‘sending’ , gmt_retry = ‘2012-11-17 23:54:10’
WHERE message_id= 73;
insert into message (body ,user_id,status,message_type,version,deal_times,gmt_create,gmt_modified,gmt_retry)
values (‘hello world！’,-2,’sending’,’instance_status_sync’,1,17,now(),now(),now());
insert into message (body ,user_id,status,message_type,version,deal_times,gmt_create,gmt_modified,gmt_retry)
values (‘hello ！’,-1,’sending’,’instance_status_sync’,2,127,now(),now(),now());
此时 往往会报错
[ERROR] Could not execute Write_rows event on table db.message; Deadlock found when trying toget lock; ; try restarting transaction, Error_code: 1213;
前两条update 类型的语句都已经获得了[59,75 )区间内记录的S锁，然后两个事务又分别对该区间段内的message_id=10这个位置请求X锁，这时就发生死锁，谁都请求不到X锁，因为互相都持有S锁。
【解决方案有两种】
1、改变程序中数据库操作的逻辑
2、取消gap lock机制
Gap locking can be disabled explicitly.This occurs if you change the transaction isolation level to READ COMMITTED orenable the innodb_locks_unsafe_for_binlog system variable.
Gap locking 可以被显示的关掉,如果你改变事务隔离级别为READ COMMITTED 或者 启动innodb_locks_unsafe_for_binlog system variable.