[MySQL优化案例]系列 — RAND()优化

众所周知,在MySQL中,如果直接 ORDER BY RAND() 的话,效率非常差,因为会多次执行。事实上,如果等值查询也是用 RAND() 的话也如此,我们先来看看下面这几个SQL的不同执行计划和执行耗时。
首先,看下建表DDL,这是一个没有显式自增主键的InnoDB表:

[yejr@imysql]> show create table t_innodb_random\G
*************************** 1. row ***************************
Table: t_innodb_random
Create Table: CREATE TABLE `t_innodb_random` (
`id` int(10) unsigned NOT NULL,
`user` varchar(64) NOT NULL DEFAULT '',
KEY `idx_id` (`id`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1

往这个表里灌入一些测试数据,至少10万以上, id 字段也是乱序的。

[yejr@imysql]> select count(*) from t_innodb_random\G
*************************** 1. row ***************************
count(*): 393216

1、常量等值检索:

[yejr@imysql]> explain select id from t_innodb_random where id = 13412\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: t_innodb_random
type: ref
possible_keys: idx_id
key: idx_id
key_len: 4
ref: const
rows: 1
Extra: Using index

[yejr@imysql]> select id from t_innodb_random where id = 13412;
1 row in set (0.00 sec)

可以看到执行计划很不错,是常量等值查询,速度非常快。

2、使用RAND()函数乘以常量,求得随机数后检索:

[yejr@imysql]> explain select id from t_innodb_random where id = round(rand()*13241324)\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using where; Using index

[yejr@imysql]> select id from t_innodb_random where id = round(rand()*13241324)\G
Empty set (0.26 sec)

可以看到执行计划很糟糕,虽然是只扫描索引,但是做了全索引扫描,效率非常差。因为WHERE条件中包含了RAND(),使得MySQL把它当做变量来处理,无法用常量等值的方式查询,效率很低。

我们把常量改成取t_innodb_random表的最大id值,再乘以RAND()求得随机数后检索看看什么情况:

[yejr@imysql]> explain select id from t_innodb_random where id = round(rand()*(select max(id) from t_innodb_random))\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using where; Using index
*************************** 2. row ***************************
id: 2
select_type: SUBQUERY
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: Select tables optimized away

[yejr@imysql]> select id from t_innodb_random where id = round(rand()*(select max(id) from t_innodb_random))\G
Empty set (0.27 sec)

可以看到,执行计划依然是全索引扫描,执行耗时也基本相当。

3、改造成普通子查询模式 ,这里有两次子查询

[yejr@imysql]> explain select id from t_innodb_random where id = (select round(rand()*(select max(id) from t_innodb_random)) as nid)\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using where; Using index
*************************** 2. row ***************************
id: 3
select_type: SUBQUERY
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: Select tables optimized away

[yejr@imysql]> select id from t_innodb_random where id = (select round(rand()*(select max(id) from t_innodb_random)) as nid)\G
Empty set (0.27 sec)

可以看到,执行计划也不好,执行耗时较慢。

4、改造成JOIN关联查询,不过最大值还是用常量表示

[yejr@imysql]> explain select id from t_innodb_random t1 join (select round(rand()*13241324) as id2) as t2 where t1.id = t2.id2\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: <derived2>
type: system
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 1
Extra:
*************************** 2. row ***************************
id: 1
select_type: PRIMARY
table: t1
type: ref
possible_keys: idx_id
key: idx_id
key_len: 4
ref: const
rows: 1
Extra: Using where; Using index
*************************** 3. row ***************************
id: 2
select_type: DERIVED
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: No tables used

[yejr@imysql]> select id from t_innodb_random t1 join (select round(rand()*13241324) as id2) as t2 where t1.id = t2.id2\G
Empty set (0.00 sec)

这时候执行计划就非常完美了,和最开始的常量等值查询是一样的了,执行耗时也非常之快。
这种方法虽然很好,但是有可能查询不到记录,改造范围查找,但结果LIMIT 1就可以了:

[yejr@imysql]> explain select id from t_innodb_random where id > (select round(rand()*(select max(id) from t_innodb_random)) as nid) limit 1\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using where; Using index
*************************** 2. row ***************************
id: 3
select_type: SUBQUERY
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: Select tables optimized away

[yejr@imysql]> select id from t_innodb_random where id > (select round(rand()*(select max(id) from t_innodb_random)) as nid) limit 1\G
*************************** 1. row ***************************
id: 1301
1 row in set (0.00 sec)

可以看到,虽然执行计划也是全索引扫描,但是因为有了LIMIT 1,只需要找到一条记录,即可终止扫描,所以效率还是很快的。

小结:
从数据库中随机取一条记录时,可以把RAND()生成随机数放在JOIN子查询中以提高效率。

5、再来看看用ORDRR BY RAND()方式一次取得多个随机值的方式:

[yejr@imysql]> explain select id from t_innodb_random order by rand() limit 1000\G
*************************** 1. row ***************************
id: 1
select_type: SIMPLE
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using index; Using temporary; Using filesort

[yejr@imysql]> select id from t_innodb_random order by rand() limit 1000;
1000 rows in set (0.41 sec)

全索引扫描,生成排序临时表,太差太慢了。

6、把随机数放在子查询里看看:

[yejr@imysql]> explain select id from t_innodb_random where id > (select rand() * (select max(id) from t_innodb_random) as nid) limit 1000\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: t_innodb_random
type: index
possible_keys: NULL
key: idx_id
key_len: 4
ref: NULL
rows: 393345
Extra: Using where; Using index
*************************** 2. row ***************************
id: 3
select_type: SUBQUERY
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: Select tables optimized away

[yejr@imysql]> select id from t_innodb_random where id > (select rand() * (select max(id) from t_innodb_random) as nid) limit 1000\G
1000 rows in set (0.04 sec)

嗯,提速了不少,这个看起来还不赖:)

7、仿照上面的方法,改成JOIN和随机数子查询关联

[yejr@imysql]> explain select id from t_innodb_random t1 join (select rand() * (select max(id) from t_innodb_random) as nid) t2 on t1.id > t2.nid limit 1000\G
*************************** 1. row ***************************
id: 1
select_type: PRIMARY
table: <derived2>
type: system
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: 1
Extra:
*************************** 2. row ***************************
id: 1
select_type: PRIMARY
table: t1
type: range
possible_keys: idx_id
key: idx_id
key_len: 4
ref: NULL
rows: 196672
Extra: Using where; Using index
*************************** 3. row ***************************
id: 2
select_type: DERIVED
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: No tables used
*************************** 4. row ***************************
id: 3
select_type: SUBQUERY
table: NULL
type: NULL
possible_keys: NULL
key: NULL
key_len: NULL
ref: NULL
rows: NULL
Extra: Select tables optimized away

[yejr@imysql]> select id from t_innodb_random t1 join (select rand() * (select max(id) from t_innodb_random) as nid) t2 on t1.id > t2.nid limit 1000\G
1000 rows in set (0.00 sec)

可以看到,全索引检索,发现符合记录的条件后,直接取得1000行,这个方法是最快的。

综上,想从MySQL数据库中随机取一条或者N条记录时,最好把RAND()生成随机数放在JOIN子查询中以提高效率。
上面说了那么多的废话,最后简单说下,就是把下面这个SQL:

SELECT id FROM table ORDER BY RAND() LIMIT n;

改造成下面这个:

SELECT id FROM table t1 JOIN (SELECT RAND() * (SELECT MAX(id) FROM table) AS nid) t2 ON t1.id > t2.nid LIMIT n;

如果想要达到完全随机,还可以改成下面这种写法:

SELECT id FROM table t1 JOIN (SELECT round(RAND() * (SELECT MAX(id) FROM table)) AS nid FROM table LIMIT n) t2 ON t1.id = t2.nid;

就可以享受在SQL中直接取得随机数了,不用再在程序中构造一串随机数去检索了。

This post has already been read 20549 times!

叶金荣

Oracle MySQL ACE Director,腾讯云TVP成员

11 thoughts to “[MySQL优化案例]系列 — RAND()优化”

  1. 貌似有问题:

    SELECT id FROM t ORDER BY RAND() LIMIT 10;

    +——–+
    | id |
    +——–+
    | 807507 |
    | 268104 |
    | 279004 |
    | 523691 |
    | 625104 |
    | 135054 |
    | 449935 |
    | 444616 |
    | 630964 |
    | 315123 |
    +——–+

    结果是随机的

    SELECT id FROM t t1 JOIN (SELECT RAND() * (SELECT MAX(id) FROM t) AS nid) t2 ON t1.id > t2.nid LIMIT 10;

    +——–+
    | id |
    +——–+
    | 385771 |
    | 385772 |
    | 385774 |
    | 385774 |
    | 385775 |
    | 385779 |
    | 385779 |
    | 385779 |
    | 385781 |
    | 385782 |
    +——–+

    结果不随机

  2. 哈哈,你没注意到我最开始交代的测试场景:
    往这个表里灌入一些测试数据,至少10万以上, id 字段也是乱序的。

    不过,你的结果看起来像是顺序的,但实际应用中,是先随机取得最小值,因此对终端用户来说,其实还是随机的 :)

  3. 试试看这个
    select a.* from t as a join (select round(rand()*(select max(id) from t)) as id from t limit 100) as b on a.id = b.id limit 10;
    只要id从1开始,并基本保证自增,应该可以满足需求。但是id不连续的话,这个办法不能排除取不满10个的可能。

  4. 我的测试场景:

    CREATE TABLE test.t (
    id int(10) UNSIGNED NOT NULL,
    name varchar(64) NOT NULL DEFAULT ”,
    INDEX id (id)
    )
    ENGINE = INNODB;

    PROCEDURE test.p1(n int)
    BEGIN
    SET @x = 0;
    WHILE @x < n DO
    INSERT INTO t (id)
    SELECT
    FLOOR(1 + RAND() * (n – 1));
    SET @x = @x + 1;
    END WHILE;
    COMMIT;
    END

    CALL p1(500000);

    版本5.6.10

    你的意思是说如果取n条随机,可以执行n次取一条的随机,是么?

  5. 随机取N条记录,可以这样:
    1、从数据库中随机顺序取得N条记录;
    2、从随机顺序X开始,往后取N条记录;
    后面这种看起来不太像真正的随机,不过也是可以接受的:)

  6. 在程序里生成随机数再去数据库查是比较好的方法,但凡是要拿数据库做一些奇淫技巧的事情的时候,我就会反省这条路是否有问题。

发表回复

您的电子邮箱地址不会被公开。 必填项已用*标注

To create code blocks or other preformatted text, indent by four spaces:

    This will be displayed in a monospaced font. The first four 
    spaces will be stripped off, but all other whitespace
    will be preserved.
    
    Markdown is turned off in code blocks:
     [This is not a link](http://example.com)

To create not a block, but an inline code span, use backticks:

Here is some inline `code`.

For more help see http://daringfireball.net/projects/markdown/syntax

Time limit is exhausted. Please reload CAPTCHA.