InnoDB相比较MyISAM一是支持事务,二是支持了行级锁,提到InnoDB锁问题就不得不提到事务,所以在这之前先了解下事务的一些知识
事务及其ACID属性
事务是由一组SQL语句组成的逻辑处理单元,事务具有以下4个属性,通常简称为事务的ACID属性。
原子性(Atomicity):事务开始后所有操作,要么全部做完,要么全部不做,不可能停滞在中间环节。事务执行过程中出错,会回滚到事务开始前的状态,所有的操作就像没有发生一样。也就是说事务是一个不可分割的整体,就像化学中学过的原子,是物质构成的基本单位。
一致性(Consistency):事务开始前和结束后,数据库的完整性约束没有被破坏 。比如A向B转账,不可能A扣了钱,B却没收到。
隔离性(Isolation):同一时间,只允许一个事务请求同一数据,不同的事务之间彼此没有任何干扰。比如A正在从一张银行卡中取钱,在A取钱的过程结束前,B不能向这张卡转账。
持久性(Durability):事务完成后,事务对数据库的所有更新将被保存到数据库,不能回滚。
并发事务处理带来的问题
相对于串行处理来说,并发事务处理能大大增加数据库资源的利用率,提高数据库系统的事务吞吐量,从而可以支持更多的用户。但是并发事务处理也会带来以下问题。
小结:不可重复读的和幻读很容易混淆,不可重复读侧重于修改,幻读侧重于新增或删除。解决不可重复读的问题只需锁住满足条件的行,解决幻读需要锁表
事务隔离级别
MySQL数据库实现事务隔离的方式,基本上可分为两种。
数据多版本并发控制
简称MVCC。四种事务隔离级别:
可以通过检查InnoDB_row_lock状态变量来分析系统伤的行锁的争夺情况:
mysql> show status like 'innodb_row_lock%';
+-------------------------------+-------+
| Variable_name | Value |
+-------------------------------+-------+
| Innodb_row_lock_current_waits | 0 |
| Innodb_row_lock_time | 0 |
| Innodb_row_lock_time_avg | 0 |
| Innodb_row_lock_time_max | 0 |
| Innodb_row_lock_waits | 0 |
+-------------------------------+-------+
5 rows in set (0.04 sec)
如果锁争用比较严重,Innodb_row_lock_waits和Innodb_row_lock_time_avg的值比较高,可以通过查询information_schema数据库中相关的表来查看锁情况,或者通过设置InnodDB Monitors来进一步观察发生锁冲突的表、数据行等,并分析其原因。
(1)通过查询information_schema数据库中的innodb_locks表了解锁的等待情况:
mysql> use information_schema;
Reading table information for completion of table and column names
You can turn off this feature to get a quicker startup with -A
Database changed
mysql> select * from innodb_locks;
Empty set, 1 warning (0.01 sec)
mysql>
(2)通过设置InnoDB Monitors观察锁冲突情况:
mysql> create table innodb_monitor(a INT) ENGINE=INNODB;
Query OK, 0 rows affected (0.05 sec)
然后通过下面语句来进行查看:
mysql> show engine innodb status;
| Type | Name | Status
| InnoDB | |
...
------------
TRANSACTIONS
------------
Trx id counter 6076
Purge done for trx's n:o < 6071 undo n:o < 0 state: running but idle
History list length 0
LIST OF TRANSACTIONS FOR EACH SESSION:
---TRANSACTION 421657844005624, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
---TRANSACTION 421657844004704, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
---TRANSACTION 421657844006544, not started
0 lock struct(s), heap size 1136, 0 row lock(s)
--------
FILE I/O
--------
...
--------------
ROW OPERATIONS
--------------
0 queries inside InnoDB, 0 queries in queue
0 read views open inside InnoDB
Process ID=1, Main thread ID=140182422546176, state: sleeping
Number of rows inserted 55250, updated 1240, deleted 376, read 22512
0.00 inserts/s, 0.00 updates/s, 0.00 deletes/s, 0.00 reads/s
----------------------------
END OF INNODB MONITOR OUTPUT
============================
1 row in set (0.01 sec)
监视器可以通过下列语句来停止:
mysql> drop table innodb_monitor;
Query OK, 0 rows affected (0.02 sec)
设置监视器后,在show innodb status的显示内容中,会有详细的当前锁等待的信息,包括表名、锁类型、锁定记录的情况等,便于进一步分析和定位问题。
InnoDB实现了两种类型的行锁
另外,为了允许行锁和表锁共存,实现多粒度锁机制,InnoDB还有两种内部使用的意向锁。
InnoDB行锁模式兼容性列表:
X | IX | S | IS | |
---|---|---|---|---|
X | 冲突 | 冲突 | 冲突 | 冲突 |
IX | 冲突 | 兼容 | 冲突 | 兼容 |
S | 冲突 | 冲突 | 兼容 | 兼容 |
IS | 冲突 | 兼容 | 兼容 | 兼容 |
如果一个事务请求的锁模式与当前的锁兼容,InnoDB就将请求的锁收于授予该事务,否则该事务就要等待锁的释放。
意向锁是InnoDB自动加的,对于update,delete和insert语句,InnoDB是会自动给涉及数据集加排他锁(X);对于普通SELECT语句,InnoDB不会加任何锁;事务可以通过以下语句显示给记录集加共享锁或排他锁。
用lock in share mode获得共享锁,主要用在需要数据依存关系时来确认某行记录是否存在,并确保没有人对这个记录进行update或delete。但是如果当前事务也需要对该记录进行更新操作,则很有可能造成死锁。
注:session1和session2是两个连接到MySQL的客户端,使用的数据库是从mysql官网下载的,下载地址:http://downloads.mysql.com/docs/sakila-db.zip
session1:
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select * from actor where actor_id=178;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | MONROE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
session2:
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select * from actor where actor_id=178;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | MONROE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
session1对actor_id=178的记录加share mode的共享锁,session2也对actor_id=178加share mode的共享锁,此时session1和session2能够加共享锁,如下:
session1
mysql> select * from actor where actor_id=178 lock in share mode;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | MONROE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
session2
mysql> select * from actor where actor_id=178 lock in share mode;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | MONROE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
紧接着session1对178记录进行update,此时session1会等待锁,与此同时session2也对178记录更新,此时session2发生死锁,退出;
session1
mysql> update actor set last_name='monore t' where actor_id=178;
...等待
session2
mysql> mysql> update actor set last_name='monore t' where actor_id=178;
ERROR 1064 (42000): You have an error in your SQL syntax; check the manual that corresponds to your MySQL server version for the right syntax to use near 'mysql> update actor set last_name='monore t' where actor_id=178' at line 1
session2提交事务,session1获取到锁,update成功。
session2
mysql> commit;
Query OK, 0 rows affected (0.00 sec)
session1
Query OK, 1 row affected (49.29 sec)
Rows matched: 1 Changed: 1 Warnings: 0
mysql> select * from actor where actor_id=178;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | monore t | 2021-09-02 12:47:50 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
session1对actor_id=178的行记录使用for update加排他锁,此时session2再次对178加排他锁是不会获取到锁的,会等待。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.01 sec)
mysql> select * from actor where actor_id=178 for update;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | monore t | 2021-09-02 12:47:50 |
+----------+------------+-----------+---------------------+
1 row in set (0.00 sec)
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select * from actor where actor_id=178 for update;
...等待
session1提交事务,session2获取到锁。
session1
mysql> commit;
Query OK, 0 rows affected (0.00 sec)
session2
mysql> select * from actor where actor_id=178 for update;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 178 | LISA | monore t | 2021-09-02 12:47:50 |
+----------+------------+-----------+---------------------+
1 row in set (4.84 sec)
InnoDB行锁是通过给索引上的索引项加锁来实现的,如果没有索引,InnoDB将通过隐藏的聚集索引Row_id来对记录加锁,InnoDB行锁分为3种情形。
InnoDB这种行锁实现特点意味着:如果不通过索引条件检索数据,那么InnoDB将对表中的所有记录加锁,实际效果跟锁表一样!在实际应用中,要特别注意InnoDB行锁这一特性,否则可能导致大量的锁冲突,从而影响并发性能。
session1加锁查询amount=8.99的数据,然后session2在加锁查询amount=3.99的数据,此时session2就会等待锁,session1 commit后session2获取到锁查询到数据。看起来session1只给amount=8.99的行加了锁,但是却出现了锁等待,原因就是在没有索引的情况下InnoDB对所有的记录加锁。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where amount=8.99 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 62 | 3 | 1 | 8.99 |
| 81 | 3 | 1 | 8.99 |
| 83 | 3 | 2 | 8.99 |
...
+------------+-------------+----------+--------+
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where amount=3.99 for update;
...等待
下面我们看一下通过索引条件加锁时的情况,例如session1加锁查询payment_id=62,session2加锁查询payment_id=81;此时使用了索引,加锁就只加在符合索引条件的记录上了,并没有出现等待锁情况。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where payment_id=62 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 62 | 3 | 1 | 8.99 |
+------------+-------------+----------+--------+
1 row in set (0.00 sec)
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where payment_id=81 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 81 | 3 | 1 | 8.99 |
+------------+-------------+----------+--------+
1 row in set (0.00 sec)
比如payment表staff_id有索引,amount没有索引,session1加锁查询staff_id=1 and amount=8.99的记录,session2加锁查询staff_id=1 and amount=3.99的记录,session2就会等待获取锁,虽然访问的是不同的行,因为锁是加在索引上,所以会产生锁冲突。session1 commit后session2获取锁成功。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.01 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where staff_id=1 and amount=8.99 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 62 | 3 | 1 | 8.99 |
| 81 | 3 | 1 | 8.99 |
| 122 | 5 | 1 | 8.99 |
| 188 | 7 | 1 | 8.99 |
...
+------------+-------------+----------+--------+
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where staff_id=1 and amount=3.99 for update;
...等待
payment表的customer_id和staff_id是索引,session1加锁查询customer_id=3,session2加锁查询staff_id=1的行,customer_id=3的数据中包含staff_id=1的数据,session2会等待锁,因为session1锁了所有customer_id=3的行,包含staff_id=1的,所以session2或等待获取锁。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where customer_id=3 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 60 | 3 | 1 | 1.99 |
| 61 | 3 | 1 | 2.99 |
| 62 | 3 | 1 | 8.99 |
| 63 | 3 | 1 | 6.99 |
| 64 | 3 | 2 | 6.99 |
| 65 | 3 | 1 | 2.99 |
| 66 | 3 | 1 | 4.99 |
| 67 | 3 | 1 | 4.99 |
| 68 | 3 | 1 | 5.99 |
| 69 | 3 | 2 | 10.99 |
| 70 | 3 | 2 | 7.99 |
| 71 | 3 | 2 | 6.99 |
| 72 | 3 | 1 | 4.99 |
| 73 | 3 | 2 | 4.99 |
| 74 | 3 | 1 | 2.99 |
| 75 | 3 | 1 | 1.99 |
| 76 | 3 | 2 | 3.99 |
| 77 | 3 | 1 | 2.99 |
| 78 | 3 | 2 | 4.99 |
| 79 | 3 | 2 | 5.99 |
| 80 | 3 | 2 | 4.99 |
| 81 | 3 | 1 | 8.99 |
| 82 | 3 | 2 | 2.99 |
| 83 | 3 | 2 | 8.99 |
| 84 | 3 | 2 | 0.99 |
| 85 | 3 | 1 | 2.99 |
+------------+-------------+----------+--------+
26 rows in set (0.00 sec)
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where staff_id=1 for update;
...等待
当用范围条件而不是相等条件检索数据,并请求共享锁或排他锁时,InnoDB会给符合条件的数据行的索引项加锁,对于键值在条件范围内但并不存在的记录,叫做“间隙(GAP)”,InnoDB会对这个“间隙”加锁,这种锁机制就是所谓的Next-Key锁。
举个例子,加锁查询payment_id>16048的数据,16049的记录会加锁,大于16049的记录(不存在)的“间隙”也会加锁。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where payment_id>16048 for update;
+------------+-------------+----------+--------+
| payment_id | customer_id | staff_id | amount |
+------------+-------------+----------+--------+
| 16049 | 599 | 2 | 2.99 |
+------------+-------------+----------+--------+
1 row in set (0.00 sec)
session2
mysql> insert into payment (payment_id, customer_id, staff_id, amount) value (16050, 3, 2, 1.99);
... 等待
还要特别说的是,InnoDB除了通过范围条件加锁时使用Next-Key锁外,如果使用相等条件请求一个不存在的记录加锁,InnoDB也会使用Next-Key。例如session1加锁请求payment_id=16051的记录(该记录不存在),session2 插入payment_id=16051的记录就会等待锁。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select payment_id, customer_id, staff_id, amount from payment where payment_id=16051 for update;
Empty set (0.01 sec)
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> insert into payment (payment_id, customer_id, staff_id, amount) value (16051, 3, 2, 1.99);
...等待
当然这两种情况事务不能太多,否则,就应该考虑使用MyISAM表了。
死锁示例,session1加锁查询payment表payment_id=15866记录,session2加锁查询actor表actor_id=200记录,之后session1加锁查询actor表actor_id=200,此时因为session2持有锁所以session1等待锁,然后session2加锁查询payment表payment_id=15866记录,这时InnoDB检测到了死锁,session2退出,session1查询到actor_id=200的记录。
session1
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select * from payment where payment_id=15866 for update;
+------------+-------------+----------+-----------+--------+---------------------+---------------------+
| payment_id | customer_id | staff_id | rental_id | amount | payment_date | last_update |
+------------+-------------+----------+-----------+--------+---------------------+---------------------+
| 15866 | 592 | 1 | 11410 | 8.99 | 2005-08-02 19:29:01 | 2006-02-15 22:23:29 |
+------------+-------------+----------+-----------+--------+---------------------+---------------------+
1 row in set (0.00 sec)
session2
mysql> start transaction;
Query OK, 0 rows affected (0.00 sec)
mysql> select * from actor where actor_id=200 for update;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 200 | THORA | TEMPLE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (0.01 sec)
session1
mysql> select * from actor where actor_id=200 for update;
... 等待锁
session2,InnoDB检测到死锁,退出事务
mysql> select * from payment where payment_id=15866 for update;
ERROR 1213 (40001): Deadlock found when trying to get lock; try restarting transaction
session2退出事务,session1获取到锁
mysql> select * from actor where actor_id=200 for update;
+----------+------------+-----------+---------------------+
| actor_id | first_name | last_name | last_update |
+----------+------------+-----------+---------------------+
| 200 | THORA | TEMPLE | 2006-02-15 04:34:33 |
+----------+------------+-----------+---------------------+
1 row in set (6.53 sec)
通过过例子可以看出,InnoDB一般能自动检测到死锁,并且使一个事务释放锁并回退,另一个事务获取到锁后可以继续正常执行,但是涉及外部锁以及表锁的情况下,InnoDB并不能完全自动检测到死锁,这就需要设置锁等待的超时时间innodb_lock_wait_timeout来解决。这个参数并不是只用来结局死锁问题,在并发比较高的情况下如果大量事务因无法立即获得所需的锁而挂起,会占用大量计算机资源,造成严重的性能问题,甚至拖垮数据库。通过设置合适的锁等待超市阈值可以避免这种情况发生。
避免死锁的常用方法