MySQL中，使用RAND()的查询优化

随机找出2万条纪录，

方案一：

最直接的语句是：

SELECT *
FROM moni_role
ORDER BY RAND()
LIMIT 20000;

mysql> EXPLAIN SELECT *  FROM moni_role ORDER BY RAND() LIMIT 20000;
+----+-------------+-----------+------+---------------+------+---------+------+---------+---------------------------------+
| id | select_type | table     | type | possible_keys | key  | key_len | ref  | rows    | Extra                           |
+----+-------------+-----------+------+---------------+------+---------+------+---------+---------------------------------+
|  1 | SIMPLE      | moni_role | ALL  | NULL          | NULL | NULL    | NULL | 1344687 | Using temporary; Using filesort |
+----+-------------+-----------+------+---------------+------+---------+------+---------+---------------------------------+

mysql> SELECT * FROM moni_role ORDER BY RAND() LIMIT 20000;
...此处省去结果
20000 rows in set (57.16 sec)

这条语句花了将近一分钟，尝试优化后的语句。

方案二：

借助子查询，在子查询里覆盖索引
SELECT *
FROM moni_role
WHERE id IN (
    SELECT id FROM moni_role
    ORDER BY RAND()
)
LIMIT 20000;

mysql> EXPLAIN SELECT *  FROM moni_role  WHERE id IN ( SELECT id FROM moni_role  ORDER BY RAND()  ) LIMIT 20000;
+----+--------------------+-----------+-----------------+---------------+---------+---------+------+---------+-------------+
| id | select_type        | table     | type            | possible_keys | key     | key_len | ref  | rows    | Extra       |
+----+--------------------+-----------+-----------------+---------------+---------+---------+------+---------+-------------+
|  1 | PRIMARY            | moni_role | ALL             | NULL          | NULL    | NULL    | NULL | 1344687 | Using where |
|  2 | DEPENDENT SUBQUERY | moni_role | unique_subquery | PRIMARY       | PRIMARY | 4       | func |       1 | Using index |
+----+--------------------+-----------+-----------------+---------------+---------+---------+------+---------+-------------+


mysql> SELECT *  FROM moni_role  WHERE id IN ( SELECT id FROM moni_role  ORDER BY RAND()  ) LIMIT 20000;
...此处省去结果
20000 rows in set (0.33 sec)

可以看到优化后的语句只花了330毫秒，性能提高了200倍左右。