查询在适用时不使用覆盖索引。

Question

我下载了雇员数据库并为基准测试目的执行了一些查询。

然后，我注意到有一个查询没有使用覆盖索引，尽管我前面创建了一个相应的索引。只有当我向查询中添加FORCE INDEX子句时，才使用覆盖索引。

我上传了两个文件，一个是执行的SQL查询，另一个是结果。

您能知道为什么只在添加FORCE INDEX子句时才使用覆盖索引吗？解释显示，在这两种情况下，索引dept_no_from_date_idx反正都在使用。

为了适应SO的标准，我还在这里编写这两个文件的内容：

SQL查询：

USE employees;

/* Creating an index for an index-covered query */
    CREATE INDEX dept_no_from_date_idx ON dept_emp (dept_no, from_date);

/* Show `dept_emp` table structure, indexes and generic data */
    SHOW TABLE STATUS LIKE "dept_emp";
    DESCRIBE dept_emp;
    SHOW KEYS IN dept_emp;

/* The EXPLAIN shows that the subquery doesn't use a covering-index */
EXPLAIN SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery should use a covering index, but isn't */
        SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`);

/* The EXPLAIN shows that the subquery DOES use a covering-index,
        thanks to the FORCE INDEX clause */
EXPLAIN SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery use a covering index */
        SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp FORCE INDEX(dept_no_from_date_idx) WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`);

结果：

--------------
/* Creating an index for an index-covered query */
    CREATE INDEX dept_no_from_date_idx ON dept_emp (dept_no, from_date)
--------------

Query OK, 331603 rows affected (33.95 sec)
Records: 331603  Duplicates: 0  Warnings: 0

--------------
/* Show `dept_emp` table structure, indexes and generic data */
    SHOW TABLE STATUS LIKE "dept_emp"
--------------

+----------+--------+---------+------------+--------+----------------+-------------+-----------------+--------------+-----------+----------------+---------------------+-------------+------------+-----------------+----------+----------------+---------+
| Name     | Engine | Version | Row_format | Rows   | Avg_row_length | Data_length | Max_data_length | Index_length | Data_free | Auto_increment | Create_time         | Update_time | Check_time | Collation       | Checksum | Create_options | Comment |
+----------+--------+---------+------------+--------+----------------+-------------+-----------------+--------------+-----------+----------------+---------------------+-------------+------------+-----------------+----------+----------------+---------+
| dept_emp | InnoDB |      10 | Compact    | 331883 |             36 |    12075008 |               0 |     21544960 |  29360128 |           NULL | 2010-05-04 13:07:49 | NULL        | NULL       | utf8_general_ci |     NULL |                |         |
+----------+--------+---------+------------+--------+----------------+-------------+-----------------+--------------+-----------+----------------+---------------------+-------------+------------+-----------------+----------+----------------+---------+
1 row in set (0.47 sec)

--------------
    DESCRIBE dept_emp
--------------

+-----------+---------+------+-----+---------+-------+
| Field     | Type    | Null | Key | Default | Extra |
+-----------+---------+------+-----+---------+-------+
| emp_no    | int(11) | NO   | PRI | NULL    |       |
| dept_no   | char(4) | NO   | PRI | NULL    |       |
| from_date | date    | NO   |     | NULL    |       |
| to_date   | date    | NO   |     | NULL    |       |
+-----------+---------+------+-----+---------+-------+
4 rows in set (0.05 sec)

--------------
    SHOW KEYS IN dept_emp
--------------

+----------+------------+-----------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
| Table    | Non_unique | Key_name              | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment |
+----------+------------+-----------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
| dept_emp |          0 | PRIMARY               |            1 | emp_no      | A         |      331883 |     NULL | NULL   |      | BTREE      |         |
| dept_emp |          0 | PRIMARY               |            2 | dept_no     | A         |      331883 |     NULL | NULL   |      | BTREE      |         |
| dept_emp |          1 | emp_no                |            1 | emp_no      | A         |      331883 |     NULL | NULL   |      | BTREE      |         |
| dept_emp |          1 | dept_no               |            1 | dept_no     | A         |           7 |     NULL | NULL   |      | BTREE      |         |
| dept_emp |          1 | dept_no_from_date_idx |            1 | dept_no     | A         |          13 |     NULL | NULL   |      | BTREE      |         |
| dept_emp |          1 | dept_no_from_date_idx |            2 | from_date   | A         |      165941 |     NULL | NULL   |      | BTREE      |         |
+----------+------------+-----------------------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+
6 rows in set (0.23 sec)

--------------
/* The EXPLAIN shows that the subquery doesn't use a covering-index */
EXPLAIN SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery should use a covering index, but isn't */
        SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`)
--------------

+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+-------------+
| id | select_type | table      | type   | possible_keys                                | key                   | key_len | ref                    | rows  | Extra       |
+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+-------------+
|  1 | PRIMARY     | <derived2> | ALL    | NULL                                         | NULL                  | NULL    | NULL                   |    50 |             |
|  1 | PRIMARY     | dept_emp   | eq_ref | PRIMARY,emp_no,dept_no,dept_no_from_date_idx | PRIMARY               | 16      | der.emp_no,der.dept_no |     1 |             |
|  2 | DERIVED     | dept_emp   | ref    | dept_no,dept_no_from_date_idx                | dept_no_from_date_idx | 12      |                        | 21402 | Using where |
+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+-------------+
3 rows in set (0.09 sec)

--------------
/* The EXPLAIN shows that the subquery DOES use a covering-index,
        thanks to the FORCE INDEX clause */
EXPLAIN SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery use a covering index */
        SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp FORCE INDEX(dept_no_from_date_idx) WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`)
--------------

+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+--------------------------+
| id | select_type | table      | type   | possible_keys                                | key                   | key_len | ref                    | rows  | Extra                    |
+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+--------------------------+
|  1 | PRIMARY     | <derived2> | ALL    | NULL                                         | NULL                  | NULL    | NULL                   |    50 |                          |
|  1 | PRIMARY     | dept_emp   | eq_ref | PRIMARY,emp_no,dept_no,dept_no_from_date_idx | PRIMARY               | 16      | der.emp_no,der.dept_no |     1 |                          |
|  2 | DERIVED     | dept_emp   | ref    | dept_no_from_date_idx                        | dept_no_from_date_idx | 12      |                        | 37468 | Using where; Using index |
+----+-------------+------------+--------+----------------------------------------------+-----------------------+---------+------------------------+-------+--------------------------+
3 rows in set (0.05 sec)

Bye

编辑：

我注意到，在最后两个查询之间有相当大的执行速度差异，结果摆在您面前：

SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
    SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
) AS `der` USING (`emp_no`, `dept_no`)
--------------

+--------+---------+------------+------------+
| emp_no | dept_no | from_date  | to_date    |
+--------+---------+------------+------------+
|  38552 | d001    | 1985-04-16 | 2000-10-20 |
            ... omitted ...
|  98045 | d001    | 1985-03-28 | 9999-01-01 |
+--------+---------+------------+------------+
50 rows in set (0.31 sec)

--------------
SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
    SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp FORCE INDEX(dept_no_from_date_idx) WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
) AS `der` USING (`emp_no`, `dept_no`)
--------------

+--------+---------+------------+------------+
| emp_no | dept_no | from_date  | to_date    |
+--------+---------+------------+------------+
|  38552 | d001    | 1985-04-16 | 2000-10-20 |
            ... omitted ...
|  98045 | d001    | 1985-03-28 | 9999-01-01 |
+--------+---------+------------+------------+
50 rows in set (0.06 sec)

但是，如果我更改了执行顺序(首先执行最后一个查询，最后执行第一个查询)，那么执行速度是相同的：

--------------
SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
    SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp FORCE INDEX(dept_no_from_date_idx) WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
) AS `der` USING (`emp_no`, `dept_no`)
--------------

+--------+---------+------------+------------+
| emp_no | dept_no | from_date  | to_date    |
+--------+---------+------------+------------+
|  38552 | d001    | 1985-04-16 | 2000-10-20 |
            ... omitted ...
|  98045 | d001    | 1985-03-28 | 9999-01-01 |
+--------+---------+------------+------------+
50 rows in set (0.08 sec)

--------------
SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
    SELECT SQL_NO_CACHE emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
) AS `der` USING (`emp_no`, `dept_no`)
--------------

+--------+---------+------------+------------+
| emp_no | dept_no | from_date  | to_date    |
+--------+---------+------------+------------+
|  38552 | d001    | 1985-04-16 | 2000-10-20 |
            ... omitted ...
|  98045 | d001    | 1985-03-28 | 9999-01-01 |
+--------+---------+------------+------------+
50 rows in set (0.08 sec)

不可能是从缓存中获取第二个查询，因为SQL_NO_CACHE是在这两个查询中编写的。那么，为什么在第一个示例中，第一个查询耗时0.31秒，第二个查询耗时0.06秒，而在第二个示例中，这两个查询花费0.08秒？

Edit2:

我认为执行速度的差异源于OS缓存，也许还有其他因素。当重复执行上述2个查询时，执行时间差异可以忽略不计。我重复了3次执行上述2次查询，得到了以下结果：

#1: 0.08 sec
#2: 0.03 sec
#1: 0.05 sec
#2: 0.05 sec
#1: 0.03 sec
#2: 0.05 sec

Answer 1

实际上，您的两个查询都使用了覆盖索引。

索引定义不包括emp_no，因此在MyISAM中，即使使用FORCE INDEX子句，Using index也是不可能的。

但是，InnoDB表是群集的，每个索引都隐式地包含PRIMARY KEY作为记录指针。

这意味着您的索引实际上是(dept_no, from_date, emp_no, dept_no)上的索引，因此包含了所需的所有字段。

EXPLAIN PLAN并不总是正确地反映这一点，但是InnoDB引擎确实能够处理这个问题。

您可以通过比较这两个查询的性能来检查它：

SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery use a covering index */
        SELECT SQL_NO_CACHE from_date, emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`);

和

SELECT SQL_NO_CACHE * FROM dept_emp INNER JOIN (
        /* The subquery use a covering index */
        SELECT SQL_NO_CACHE to_date, emp_no, dept_no FROM dept_emp WHERE dept_no="d001" ORDER BY from_date DESC LIMIT 20000,50
    ) AS `der` USING (`emp_no`, `dept_no`);

您将看到，尽管计划将显示为相同，但第二个查询将花费更多的时间(确切地说是因为没有涵盖to_date )。

这是EXPLAIN PLAN中的一个bug，而不是InnoDB引擎中的错误。