关于MySQL insert into ... select 的锁情况

摘要：

      一直以为"insert into tb select * from tbx" 这样的导入操作是会把tbx表给锁住的，在锁期间是不允许任何操作（保证一致性）。看完这篇写的之后，发现其实我错了一半。tbx表是会被锁住，但这个锁有2种情况，现在逐一进行分析：

分析

环境：

root@127.0.0.1 : test 02:10:40>select @@global.tx_isolation,@@session.tx_isolation;
+-----------------------+------------------------+
| @@global.tx_isolation | @@session.tx_isolation |
+-----------------------+------------------------+
| REPEATABLE-READ       | REPEATABLE-READ        |
+-----------------------+------------------------+
1 row in set (0.00 sec)

root@127.0.0.1 : test 02:10:50>select @@version;
+------------+
| @@version  |
+------------+
| 5.6.10-log |
+------------+
1 row in set (0.00 sec)

1：按照主键排序插入的情况

直接插入，不加排序字段（默认）：

session1：执行操作,表只有5W条记录
root@127.0.0.1 : test 02:10:51>insert into uu select * from user;

session2：查看操作锁的情况（锁的行数）
root@127.0.0.1 : (none) 02:13:30>pager grep "lock(s)"
PAGER set to 'grep "lock(s)"'root@127.0.0.1 : (none) 02:18:08>show engine innodb status; #被锁的行数逐步增加
274 lock struct(s), heap size 31160, 17746 row lock(s), undo log entries 17474

root@127.0.0.1 : (none) 02:18:16>show engine innodb status;
500 lock struct(s), heap size 63928, 32572 row lock(s), undo log entries 32074

root@127.0.0.1 : (none) 02:18:17>show engine innodb status;
676 lock struct(s), heap size 80312, 44308 row lock(s), undo log entries 43635

用主键升序插入：

  情况和1一样。即默认的"select * from tb" 和 "select * from tb order id(PK) ASC " 是一样的情况。

用主键倒序插入：

  情况和1一样。即默认的"select * from tb" 和 "select * from tb order id(PK) DESC" 是一样的情况，这里说的一样是锁方式一样（都是逐步，只是顺序不一样）。

从上面可知：通过主键排序或则不加排序字段的导入操作"insert into tb select * from tbx"，是会锁tbx表，但他的锁是逐步地锁定已经扫描过的记录。

2：按照非主键排序插入的情况

session1：执行操作
root@127.0.0.1 : test 02:33:00>insert into uu select * from user order by createTime ;

session2：查看操作锁的情况（行数）
root@127.0.0.1 : (none) 02:27:29>pager grep "lock(s)"
root@127.0.0.1 : (none) 02:27:54>show engine innodb status;  #被锁的行数一样，不变（整张表）
773 lock struct(s), heap size 80312, 50771 row lock(s), undo log entries 1843

root@127.0.0.1 : (none) 02:33:19>show engine innodb status;
773 lock struct(s), heap size 80312, 50771 row lock(s), undo log entries 17680

root@127.0.0.1 : (none) 02:33:20>show engine innodb status;
773 lock struct(s), heap size 80312, 50771 row lock(s), undo log entries 22260

root@127.0.0.1 : (none) 02:33:21>show engine innodb status;
773 lock struct(s), heap size 80312, 50771 row lock(s), undo log entries 28960

从上面可知：通过非主键排序的导入操作"insert into tb select * from tbx"，是会锁tbx表，但他的锁是一开始就会锁定整张表。

总之，"insert into tb select * from tbx" 的导入操作是会锁定原表，但是锁是有2种情况：“逐步锁”，“全锁”。



验证：

针对1的情况：逐步锁定扫描过的记录，那操作未扫描的数据会怎么样？

session1：执行操作
root@127.0.0.1 : test 02:55:27>insert into uu select * from user;
Query OK, 49998 rows affected (9.06 sec)

session2：测试操作锁的情况
root@127.0.0.1 : test 02:54:49>delete from user where id = 33333;update user set username='TEST' where id = 44444;insert into user(id,username,company) values(1000,'ASD','ABCASDA');
Query OK, 0 rows affected (0.00 sec) #可以删除未扫描（锁）的数据(id=33333)

Query OK, 0 rows affected (0.00 sec)
Rows matched: 1  Changed: 0  Warnings: 0 #可以更新为扫描（锁）的数据(id=44444)

Query OK, 1 row affected (8.09 sec)#插入（更新，删除）操作被锁了，因为该记录已经被扫描到(id=1000)

session3：查看操作的锁情况：
root@127.0.0.1 : (none) 02:55:33>show engine innodb status;
LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s)
272 lock struct(s), heap size 31160, 17574 row lock(s), undo log entries 17305
1 row in set (0.09 sec)

root@127.0.0.1 : (none) 02:55:35>show engine innodb status;
LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s)
448 lock struct(s), heap size 47544, 29109 row lock(s), undo log entries 28664
1 row in set (0.01 sec)

root@127.0.0.1 : (none) 02:55:37>show engine innodb status;
LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s)
612 lock struct(s), heap size 63928, 40034 row lock(s), undo log entries 39425
1 row in set (0.00 sec)

root@127.0.0.1 : (none) 02:55:39>show engine innodb status;
1 row in set (0.01 sec)

从上面看出，刚好说明了1的情况：逐步的锁定已经扫描过的记录。

默认、主键升序的select ：从第一行开始扫描到最后，即第一行开始锁直到最后。

主键倒序select             ：从最后一行开始扫描到最前，即最后一行开始锁直到第一行。

针对2的情况：锁定整张表，那就是表锁；不能进行任何操作，直到锁释放了？

session1：执行操作
root@127.0.0.1 : test 03:23:06>insert into uu select * from user order by company;
Query OK, 49994 rows affected (13.70 sec)

session2：测试操作锁的情况
root@127.0.0.1 : test 03:22:44>delete from user where id = 33337;update user set username='TESAAST' where id = 44443;insert into user(id,username,company) values(1000,'ASD','ABCASDA');
Query OK, 1 row affected (9.58 sec)  #直接被锁住了，等待session1释放了。

Query OK, 0 rows affected (0.00 sec)
Rows matched: 1  Changed: 0  Warnings: 0 #同上

Query OK, 1 row affected (0.00 sec) #同上

session3：查看操作的锁情况：
root@127.0.0.1 : (none) 03:22:45>pager grep "lock(s)"
PAGER set to 'grep "lock(s)"'
root@127.0.0.1 : (none) 03:23:20>show engine innodb status;
773 lock struct(s), heap size 80312, 50765 row lock(s), undo log entries 4433
1 row in set (0.02 sec)

root@127.0.0.1 : (none) 03:23:28>show engine innodb status;
LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s)
773 lock struct(s), heap size 80312, 50765 row lock(s), undo log entries 25383
1 row in set (0.06 sec)

root@127.0.0.1 : (none) 03:23:32>show engine innodb status;
LOCK WAIT 2 lock struct(s), heap size 376, 1 row lock(s)
773 lock struct(s), heap size 80312, 50765 row lock(s), undo log entries 42464
1 row in set (0.01 sec)

从上面看出，刚好说明了2的情况：一开始就会锁定整张表的记录，不能进行任何操作，直到锁释放了。



总结：

类似"insert into tb select * from tbx" 的操作，最好确保tbx表不被做dml操作，不然很可能出现锁等待的情况。另：通过设置隔离级别：read committed & ROW(binlog_format)可以让dml和该语句并发操作。

session1：执行操作
root@127.0.0.1 : test 04:05:08>insert into uu select * from user order by company;
Query OK, 49990 rows affected (14.09 sec)

session2：测试操作锁的情况
root@127.0.0.1 : test 04:04:57>delete from user where id = 33318;update user set username='TESAAeST' where id = 44423;insert into user(id,username,company) values(1000,'ASD','ABCASDA');
Query OK, 0 rows affected (0.00 sec)

Query OK, 0 rows affected (0.00 sec)
Rows matched: 1  Changed: 0  Warnings: 0

Query OK, 1 row affected (0.00 sec)

session3：查看操作的锁情况：
root@127.0.0.1 : (none) 03:22:45>pager grep "lock(s)"
PAGER set to 'grep "lock(s)"'
root@127.0.0.1 : test 04:05:23>show engine innodb status;
1 lock struct(s), heap size 376, 0 row lock(s), undo log entries 6256
1 row in set (0.05 sec)

root@127.0.0.1 : test 04:05:28>show engine innodb status;
1 lock struct(s), heap size 376, 0 row lock(s), undo log entries 32958
1 row in set (0.01 sec)

root@127.0.0.1 : test 04:05:35>show engine innodb status;
1 lock struct(s), heap size 376, 0 row lock(s), undo log entries 33784
1 row in set (0.00 sec)

root@127.0.0.1 : test 04:05:36>show engine innodb status;
1 lock struct(s), heap size 376, 0 row lock(s), undo log entries 34789
1 row in set (0.00 sec)

参考：

http://www.cnblogs.com/zhoujinyi/archive/2013/04/28/3049382.html

http://blog.csdn.net/muzizhuben/article/details/46533799

http://blog.chinaunix.net/uid-9950859-id-181376.html