MySQL锁机制
SELECT ········FOR UPDATE
的语法开启排他锁。select … lock in share
可加共享锁
innodb
加行锁的前提是:必须是通过索引条件来检索数据,否则会切换为表锁。
因此,悲观锁在未通过索引条件检索数据时,会锁定整张表。导致其他程序不允许“加锁的查询操作”,影响吞吐。故如果在查询居多的情况下,推荐使用乐观锁。
共享锁与排他锁
说明:数据库的增删改操作默认都会加排他锁,而查询不会加任何锁。
共享锁
对某一资源加共享锁,自身可以读该资源,其他人也可以读该资源(也可以再继续加共享锁,即共享锁可多个共存),但无法修改。要想修改就必须等所有共享锁都释放完之后。
语法为:
select * from table lock in share mode
排他锁
对某一资源加排他锁,自身可以进行增删改查,其他人无法进行任何操作。
语法为:
select * from table for update
使用悲观锁(其实说白了也就是排他锁)(推荐)
|-- 程序A在查询库存数时使用排他锁 (select * from table where id=10 for update)
|-- 然后进行后续的操作,包括更新库存数,最后提交事务。
|-- 程序B在查询库存数时,如果A还未释放排他锁,它将等待……
|-- 程序C同B……
注意: 排他锁的真实含义不是排他锁锁住一行数据后,其他事务就不能读取和修改该行数据,而是一个事务在一行数据加上排他锁后,其他事务不能再在其上加其他的锁。
使用乐观锁(靠表设计和代码来实现)
|-- 一般是在该商品表添加version版本字段或者timestamp时间戳字段
|-- 程序A查询后,执行更新变成了:
update table set num=num-1 where id=10 and version=23
悲观锁解决商品超卖问题的实现
使用场景举例
以MySQL InnoDB为例
商品good
表,假设商品的id为1,购买数量为1,status
为1表示上架中,2表示下架。现在用户购买此商品,在不是高并发的情况下处理逻辑是:
- 查找此商品的信息;
(返回 '商品不存在')
- 检查商品库存是否大于购买数量;
('库存不足')
- 修改商品库存和销量;
[danger] 上面这种场景在高并发访问的情况下很可能会出现问题。如果商品库存是100个,高并发的情况下可能会有1000个同时访问,在到达第2步的时候,都会检测通过。这样会出现商品库存是-900个的情况
数据表
商品表 good
CREATE TABLE
good
(
id
bigint(20) unsigned NOT NULL AUTO_INCREMENT,
name
varchar(100) NOT NULL DEFAULT '' COMMENT '商品名称',
status
tinyint(4) unsigned NOT NULL DEFAULT '1' COMMENT '是否上架(1:上架中,2:表示下架)',
total
int(11) unsigned NOT NULL DEFAULT '0' COMMENT '库存',
sell
int(11) unsigned NOT NULL DEFAULT '0' COMMENT '销售数',
price
decimal(10,2) unsigned NOT NULL COMMENT '价格',
PRIMARY KEY (id
)
) ENGINE=InnoDB AUTO_INCREMENT=2 DEFAULT CHARSET=utf8mb4;
INSERT INTO
good
(id
,name
,status
,total
,sell
,price
) VALUES (1, '商品', 1, 100, 0, 15.00);
订单表:order
CREATE TABLE order
(
id
bigint(20) unsigned NOT NULL AUTO_INCREMENT,
uid
int(11) NOT NULL DEFAULT '0' COMMENT '用户id',
status
tinyint(1) NOT NULL DEFAULT '1',
goods_id
int(11) NOT NULL DEFAULT '0',
order_no
varchar(200) COLLATE utf8_unicode_ci NOT NULL DEFAULT '',
create_time
datetime NOT NULL,
PRIMARY KEY (id
)
) ENGINE=InnoDB AUTO_INCREMENT=31 DEFAULT CHARSET=utf8 COLLATE=utf8_unicode_ci;
使用悲观锁处理
实现逻辑: 当我们在查询出
good
信息后就把当前的数据锁定,直到我们修改完毕后再解锁。那么在这个过程中,因为good
被锁定了,就不会出现有第三者来对其进行修改了。注:要使用悲观锁,我们必须关闭mysql数据库的自动提交属性,因为MySQL默认使用autocommit模式,也就是说,当你执行一个更新操作后,MySQL会立刻将结果进行提交。
这里
ThinkORM
中使用事务,手动进行提交回滚。
并发测试
使用Apache JMeter
工具进行压力并发测试
JMeter是一个开源的Java应用程序,由Apache软件基金会开发和维护,可用于性能测试、压力测试、接口测试等。
测试代码
public function noLock()
{
$num = 1;
$goods_id = 1;
Db::startTrans();
try {
$where = [];
where['id'] = goods_id;
$where['status'] = 1;
// goodInfo = Db::table('good')->lock(true)->where(where)->find(); // 加悲观锁$goodInfo = Db::table('good')->where($where)->find(); // 不加悲观锁 if (empty($goodInfo)) { return '商品不存在'; } if ($goodInfo['total'] < $num) { return '库存不足'; } $orderUpdate['total'] = $goodInfo['total'] - $num; $orderUpdate['sell'] = $goodInfo['sell'] + $num; Db::table('good')->where(['id' => $goods_id])->update($orderUpdate); $orderData = []; $orderData['uid'] = rand(10000, 99999); $orderData['status'] = 1; $orderData['create_time'] = date('Y-m-d H:i:s'); $orderData['goods_id'] = $goods_id; $orderData['order_no'] = date('YmdHis') . rand(1000, 10000); Db::table('order')->insert($orderData); Db::commit(); } catch (\Exception $exception) { Db::rollback(); return '异常回滚' . $exception->getMessage(); } return '恭喜!抢购成功!';
}
1、创建测试
(1)创建线程测试组 模拟1s内1000个用户同时访问
(2)创建http请求
(3)添加察看结果树
2、测试开始
(1)100个商品不加锁的结果
100个商品库存,生成订单1000个,超卖944个商品,还剩下44个商品
(2)100个商品,加锁结果
可以看到 ,通过加锁可以解决该问题哈
其他问题
General error: 1205 Lock wait timeout exceeded; try restarting transaction
从报错看是因为事务锁超时了,那么超时时间设置问题?
事务锁超时时间 innodb_lock_wait_timeout
mysql> show variables like 'innodb_lock_wait_timeout';
+--------------------------+-------+
| Variable_name | Value |
+--------------------------+-------+
| innodb_lock_wait_timeout | 50 |
+--------------------------+-------+
1 row in set (0.06 sec)
超时时间是50秒
sudo vim /etc/mysql/mysql.conf.d/mysqld.cnf
添加一下,设置为120秒超时时间
innodb_lock_wait_timeout=120
重启MySQL
查询进程 查询数据库线程情况,看是否有执行很慢的SQL
mysql> show processlist;
+----+------+---------------------+-----------------------+---------+------+------------+-----------------------------------------------------------------------------+
| Id | User | Host | db | Command | Time | State | Info |
+----+------+---------------------+-----------------------+---------+------+------------+-----------------------------------------------------------------------------+
| 27 | www | 47.98.155.92:60996 | webman.tinywan.com | Sleep | 17 | | NULL |
| 28 | www | 39.170.63.237:60955 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 29 | www | 39.170.63.237:60952 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 30 | www | 39.170.63.237:60949 | cms.webman.tinywan.cn | Execute | 87 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 31 | www | 39.170.63.237:60951 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 32 | www | 39.170.63.237:60956 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 33 | www | 39.170.63.237:60950 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 34 | www | 39.170.63.237:60948 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 35 | www | 39.170.63.237:60954 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 36 | www | 39.170.63.237:60953 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 37 | www | 39.170.63.237:60957 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 38 | www | 39.170.63.237:60958 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 39 | www | 39.170.63.237:60959 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 40 | www | 39.170.63.237:60961 | cms.webman.tinywan.cn | Execute | 87 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 41 | www | 39.170.63.237:60960 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 42 | www | 39.170.63.237:60962 | cms.webman.tinywan.cn | Execute | 87 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 43 | www | 39.170.63.237:60963 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 44 | www | 39.170.63.237:60964 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 45 | www | 39.170.63.237:60967 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 46 | www | 39.170.63.237:60966 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 47 | www | 39.170.63.237:60965 | cms.webman.tinywan.cn | Sleep | 7 | | NULL |
| 48 | www | 39.170.63.237:60968 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 49 | www | 39.170.63.237:60969 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 50 | www | 39.170.63.237:60970 | cms.webman.tinywan.cn | Sleep | 7 | | NULL |
| 51 | www | 39.170.92.205:60971 | cms.webman.tinywan.cn | Execute | 86 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 52 | www | 39.170.92.205:60973 | cms.webman.tinywan.cn | Execute | 87 | statistics | SELECT * FROM good
WHERE id
= ? AND status
= ? LIMIT 1 FOR UPDATE |
| 53 | www | 39.170.63.237:60999 | cms.webman.tinywan.cn | Query | 0 | starting | show processlist |
| 54 | www | 39.170.92.205:61083 | cms.webman.tinywan.cn | Sleep | 84 | | NULL |
| 55 | www | 39.170.92.205:61084 | cms.webman.tinywan.cn | Sleep | 130 | | NULL |
+----+------+---------------------+-----------------------+---------+------+------------+-----------------------------------------------------------------------------+
29 rows in set (0.11 sec)
在执行结果中可以看到是否有表锁等待或者死锁,如果有死锁发生,可以通过下面的命令来杀掉当前运行的事务
查询到相对应的进程
再去查看innodb的事务表INNODB_TRX
,看下里面是否有正在锁定的事务线程,看看ID是否在show full processlist里面的sleep线程中,如果是,就证明这个sleep的线程事务一直没有commit或者rollback而是卡住了,我们需要手动kill掉
kill id