网站首页 > 技术文章 正文
前言
在日常开发中,对于 LEFT JOIN 和 JOIN 的用法大部分应该都是一样的,如果有两个表 A,B,如果两个表的数据都想要,就使用 JOIN,如果只想要一个表的全部数据,另一个表数据可有可无,就使用 LEFT JOIN。(当然这么描述是不太准确的,但是很符合我的日常业务开发)。
在 MYSQL LEFT JOIN 详解 这篇文章中我们已经知道了,LEFT JOIN 是自己选择驱动表的,而 JOIN 是 MYSQL 优化器选择驱动标的。
那么,当我们写了一条 LEFT JOIN 语句,MYSQL 会将这条语句优化成 JOIN 语句吗?
如果会优化的话,那么什么时候会优化呢?
事实上,这正是我遇到的一个线上问题。我们一起来看一下。
问题描述
在我们线上有这么一条慢 SQL(已处理),执行时间超过 0.5 秒。
select
count(distinct order.order_id)
from order force index(shop_id)
left join `order_extend`
on `order`.`order_id` = `order_extend`.`order_id`
where `order`.`create_time` >= "2020-08-01 00:00:00"
and `order`.`create_time` <= "2020-08-01 23:59:59"
and `order`.`shop_id` = 328449726569069326
and `order`.`status` = 1
and `order_extend`.`shop_id` = 328449726569069326
and `order_extend`.`status` = 1
复制代码
explain 结果如下:
+----+-------------+--------------+------------+--------+------------------+----------+---------+------------------------+------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+--------------+------------+--------+------------------+----------+---------+------------------------+------+-------------+
| 1 | SIMPLE | order_extend | NULL | ref | order_id,shop_id | shop_id | 8 | const | 3892 | Using where |
| 1 | SIMPLE | order | NULL | eq_ref | shop_id | shop_id | 16 | example.order.order_id | 1 | Using where |
+----+-------------+--------------+------------+--------+------------------+----------+---------+------------------------+------+-------------+
2 rows in set, 1 warning (0.00 sec)
复制代码
问题分析
通过 explain,再结合我们之前讲的 MYSQL 连接查询算法,驱动表为 order_extend,循环 3892 次,说多也不多,说少也不少,被驱动表数据查询类型为 eq_ref,所以应该不会太慢,那么问题就出现在 3892 次上面了,想办法将这个数字降下来即可。
等等!为什么驱动表是 order_extend?我明明使用的是 LEFT JOIN 啊,按理说驱动表应该是 order 表,为什么会变成了 order_extend 了。难道是 MYSQL 内部优化了?
顺着这个思路,既然驱动表变了,说明这条 SQL 变为 JOIN 语句了。
我们顺着分析 JOIN 语句的方式来分析一下这条语句。(ps:需要对 MYSQL JOIN 内部执行过程有一定的理解,如果不太熟悉,请先移步看这篇文章 → MYSQL 连接查询算法 )
MYSQL 选择 order_extend 当做驱动表,说明在 where 条件下 order_extend 查询的数据更少,MYSQL 会选择一个小的表当做驱动表。
我们来分别适用上述的 where 条件单独执行 select count(*) 语句,查看一下大致每个表都涉及到多少条 SQL 记录。
为了不影响我们的分析,我们使用 explain 语句,这样整个过程就都是估算的结果,模拟一下 MYSQL 分析的过程。
mysql> explain select
count(distinct order.order_id)
from order force index(shop_id)
where `order`.`create_time` >= "2020-08-01 00:00:00"
and `order`.`create_time` <= "2020-08-01 23:59:59"
and `order`.`shop_id` = 328449726569069326
and `order`.`status` = 1;
+----+-------------+-------+------------+------+--------------------------------+---------+---------+-------+--------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | Extra |
+----+-------------+-------+------------+------+--------------------------------+---------+---------+-------+--------+-------------+
| 1 | SIMPLE | order | NULL | ref | PRIMARY,shop_id,create_time... | shop_id | 8 | const | 320372 | Using where |
+----+-------------+-------+------------+------+--------------------------------+---------+---------+-------+--------+-------------+
1 row in set, 1 warning (0.00 sec)
复制代码
select
count(distinct order_extend.order_id)
and `order_extend`.`shop_id` = 328449726569069326
and `order_extend`.`status` = 1
+----+-------------+--------------+------------+------+------------------+---------+---------+-------+------+----------+-------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+--------------+------------+------+------------------+---------+---------+-------+------+----------+-------------+
| 1 | SIMPLE | order_extend | NULL | ref | order_id,shop_id | shop_id | 8 | const | 3892 | 10.00 | Using where |
+----+-------------+--------------+------------+------+------------------+---------+---------+-------+------+----------+-------------+
1 row in set, 1 warning (0.00 sec)
复制代码
可以看到,在上述 where 条件下,order_extend 表只会查询 3892 条数据,而 order 表会查询 320372 条数据,所以 order_extend 表当驱动表是完全没有问题的。
那么我们再来看看为什么 order 表会扫描这么多数据呢?在 2020-08-01 这一天可能也没有这么多数据啊。那么这个时候我们应该会很容易的想到,是强制走索引的问题,因为在上述查询语句中,我们强制走了 shop_id 索引,这个索引可能不是最优索引,我们把 force index(shop_id) 去掉再试试看
mysql> explain select
count(distinct order.order_id)
where `order`.`create_time` >= "2020-08-01 00:00:00"
and `order`.`create_time` <= "2020-08-01 23:59:59"
and `order`.`shop_id` = 328449726569069326
and `order`.`status` = 1;
+----+-------------+-------+------------+------+---------------+-------------+---------+-------+-------+----------+--------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+-------------+---------+-------+-------+----------+--------------------------+
| 1 | SIMPLE | order | NULL | ref | create_time | create_time | 8 | const | <3892 | 10.00 | Using where; Using index |
+----+-------------+-------+------------+------+---------------+-------------+---------+-------+-------+----------+--------------------------+
1 row in set, 1 warning (0.00 sec)
复制代码
可以看到,如果不强制走 shop_id 索引的话,走 create_time 索引的话,扫描的行数会更少,假设说 100 行,只会循环 100 次,扫描 100 x 3892 行数据,而之前的总共要循环 3892 次,扫描 3892 x 300000 行数据。
问题结论
所以最终的这条慢 SQL 的原因确定了,是因为我们强制走 shop_id 索引,导致 MYSQL 扫描的行数更多了,我们只需要去掉强制走索引即可,大多数时间 MYSQL 都会选择正确的索引,所以强制使用索引的时候一定要小心谨慎。
问题延伸
SQL 慢的问题我们已经解决了,我们再来回顾一下文章开头的问题:LEFT JOIN 会被优化为 JOIN 吗?
答案是会的。那么什么时候会出现这种情况呢?
我们再来回顾一下 MYSQL LEFT JOIN 详解 文章中的内容。
为了方便阅读,我们将部分内容粘贴出来。
mysql> select * from goods left join goods_category on goods.category_id = goods_category.category_id;
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 1 | 男鞋1 | 1 | 1 | 鞋 |
| 2 | 男鞋2 | 1 | 1 | 鞋 |
| 3 | 男鞋3 | 3 | 3 | 羽绒服 |
| 4 | T恤1 | 2 | 2 | T恤 |
| 5 | T恤2 | 2 | 2 | T恤 |
+----------+------------+-------------+-------------+---------------+
5 rows in set (0.00 sec)
mysql> select * from goods left join goods_category on goods.category_id = goods_category.category_id;
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 1 | 男鞋1 | 1 | 1 | 鞋 |
| 2 | 男鞋2 | 1 | 1 | 鞋 |
| 3 | 男鞋3 | 4 | NULL | NULL |
| 4 | T恤1 | 2 | 2 | T恤 |
| 5 | T恤2 | 2 | 2 | T恤 |
+----------+------------+-------------+-------------+---------------+
5 rows in set (0.00 sec)
mysql> select * from goods g left join goods_category c on (g.category_id = c.category_id and g.goods_name = 'T恤1');
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 1 | 男鞋1 | 1 | NULL | NULL |
| 2 | 男鞋2 | 1 | NULL | NULL |
| 3 | 男鞋3 | 4 | NULL | NULL |
| 4 | T恤1 | 2 | 2 | T恤 |
| 5 | T恤2 | 2 | NULL | NULL |
+----------+------------+-------------+-------------+---------------+
5 rows in set (0.00 sec)
mysql> select * from goods g left join goods_category c on (g.category_id = c.category_id and c.category_name = 'T恤');
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 1 | 男鞋1 | 1 | NULL | NULL |
| 2 | 男鞋2 | 1 | NULL | NULL |
| 3 | 男鞋3 | 4 | NULL | NULL |
| 4 | T恤1 | 2 | 2 | T恤 |
| 5 | T恤2 | 2 | 2 | T恤 |
+----------+------------+-------------+-------------+---------------+
5 rows in set (0.00 sec)
mysql> select * from goods g left join goods_category c on (g.category_id = c.category_id) where c.category_name = '鞋';
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 1 | 男鞋1 | 1 | 1 | 鞋 |
| 2 | 男鞋2 | 1 | 1 | 鞋 |
+----------+------------+-------------+-------------+---------------+
2 rows in set (0.00 sec)
mysql> select * from goods g left join goods_category c on (g.category_id = c.category_id) where g.goods_name = 'T恤1';
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 4 | T恤1 | 2 | 2 | T恤 |
+----------+------------+-------------+-------------+---------------+
1 row in set (0.00 sec)
mysql> select * from goods g left join goods_category c on (g.category_id = c.category_id and g.goods_name = 'T恤2') where g.goods_name = 'T恤1';
+----------+------------+-------------+-------------+---------------+
| goods_id | goods_name | category_id | category_id | category_name |
+----------+------------+-------------+-------------+---------------+
| 4 | T恤1 | 2 | NULL | NULL |
+----------+------------+-------------+-------------+---------------+
1 row in set (0.00 sec)
复制代码
我们可以看到,当 where 条件中有被驱动表的条件时,查询结果是和 JOIN 的结果是一致的,无 NULL 值的出现。
所以,我们可以想到,LEFT JOIN 优化为 JOIN 的条件为:where 条件中有被驱动表的非空条件时,LEFT JOIN 等价于 JOIN。
这不难理解,LEFT JOIN 会返回驱动表所有数据,当有被驱动表的 where 条件时,会过滤掉 NULL 的值,此时和 JOIN 的结果一致了,那么 MYSQL 会选择将 LEFT JOIN 优化为 JOIN,这样就可以自己选择驱动表了。
实例测试
我们再来编写一个测试用例来验证一下我们的结论。
CREATE TABLE `A` (
`id` int(11) auto_increment,
`a` int(11) DEFAULT NULL,
PRIMARY KEY (`id`),
KEY `a` (`a`)
) ENGINE=InnoDB;
delimiter ;;
create procedure idata()
begin
declare i int;
set i=1;
while(i<=100)do
insert into A (`a`) values(i);
set i=i+1;
end while;
end;;
delimiter ;
call idata();
CREATE TABLE `B` (
`id` int(11) auto_increment,
`b` int(11) DEFAULT NULL,
PRIMARY KEY (`id`),
KEY `b` (`b`)
) ENGINE=InnoDB;
delimiter ;;
create procedure idata()
begin
declare i int;
set i=1;
while(i<=100)do
insert into B (`b`) values(i);
set i=i+1;
end while;
end;;
delimiter ;
call idata();
复制代码
我们创建了两张一模一样的表,每个表中有 100 条数据,然后我们执行一下 LEFT JOIN 语句。
mysql> explain select * from A left join B on A.id = B.id where A.a <= 100;
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| 1 | SIMPLE | A | NULL | index | a | a | 5 | NULL | 100 | 100.00 | Using where; Using index |
| 1 | SIMPLE | B | NULL | eq_ref | PRIMARY | PRIMARY | 4 | example2.A.id | 1 | 100.00 | NULL |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
2 rows in set, 1 warning (0.00 sec)
复制代码
mysql> explain select * from A left join B on A.id = B.id where A.a <= 100 and B.b <= 50;
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| 1 | SIMPLE | B | NULL | range | PRIMARY,b | b | 5 | NULL | 50 | 100.00 | Using where; Using index |
| 1 | SIMPLE | A | NULL | eq_ref | PRIMARY,a | PRIMARY | 4 | example2.B.id | 1 | 100.00 | Using where |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
2 rows in set, 1 warning (0.00 sec)
复制代码
mysql> explain select * from A left join B on A.id = B.id where A.a <= 100 and B.b <= 100;
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
| 1 | SIMPLE | A | NULL | index | PRIMARY,a | a | 5 | NULL | 100 | 100.00 | Using where; Using index |
| 1 | SIMPLE | B | NULL | eq_ref | PRIMARY,b | PRIMARY | 4 | example2.A.id | 1 | 100.00 | Using where |
+----+-------------+-------+------------+--------+---------------+---------+---------+---------------+------+----------+--------------------------+
2 rows in set, 1 warning (0.00 sec)
复制代码
从上面看,给 B 表增加了 where 条件之后,如果 B 表扫描的行数更少,那么是有可能换驱动表的,这也说明了,LEFT JOIN 语句被优化成了 JOIN 语句。
总结
上面我们分析了一条慢 SQL 的问题,分析的过程涉及到了很多知识点,希望大家可以认真研究一下。
同时我们得出了一条结论:当有被驱动表的非空 where 条件时,MYSQL 会将 LEFT JOIN 语句优化为 JOIN 语句。
猜你喜欢
- 2024-11-16 从阿里手册引出的Join查询思考(阿里技术手册)
- 2024-11-16 Python中join()方法——字符串、路径拼接的常用方法
- 2024-11-16 oracle 多表连接查询 join(一)(oracle多表联查效率优化)
- 2024-11-16 让程序员头疼的微服务下数据聚合join(一)
- 2024-11-16 神奇的 SQL 之 联表细节 → MySQL JOIN 的执行过程你是否知道?
- 2024-11-16 MySQL的Join到底能不能用,一文搞懂它
- 2024-11-16 一文看懂Mycat跨分片Join实现--全局表、ER分片、HBT、sharejoin
- 2024-11-16 为什么大厂不建议使用多表join?(为什么要多表查询)
- 2024-11-16 SQL Server 2012 高级用法(三)之Join
- 2024-11-16 LINUX中JOIN用法(linux jour)
- 标签列表
-
- content-disposition (47)
- nth-child (56)
- math.pow (44)
- 原型和原型链 (63)
- canvas mdn (36)
- css @media (49)
- promise mdn (39)
- readasdataurl (52)
- if-modified-since (49)
- css ::after (50)
- border-image-slice (40)
- flex mdn (37)
- .join (41)
- function.apply (60)
- input type number (64)
- weakmap (62)
- js arguments (45)
- js delete方法 (61)
- blob type (44)
- math.max.apply (51)
- js (44)
- firefox 3 (47)
- cssbox-sizing (52)
- js删除 (49)
- js for continue (56)
- 最新留言
-