最左前缀原理与相关优化
- 2019 年 11 月 8 日
- 筆記
<a1, a2, …, an>，其中各个元素均为数据表的一列。另外，单列索引可以看成联合索引元素数为 1 的特例。
我们在 Employees Sample Database 中实验，MySQL 版本 5.7。
以 employees.titles 为例，查看其索引：
SHOW INDEX FROM employees.titles; +--------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ | Table | Non_unique | Key_name | Seq_in_index | Column_name | Collation | Cardinality | Sub_part | Packed | Null | Index_type | Comment | Index_comment | +--------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+ | titles | 0 | PRIMARY | 1 | emp_no | A | 301292 | NULL | NULL | | BTREE | | | | titles | 0 | PRIMARY | 2 | title | A | 442605 | NULL | NULL | | BTREE | | | | titles | 0 | PRIMARY | 3 | from_date | A | 442605 | NULL | NULL | | BTREE | | | +--------+------------+----------+--------------+-------------+-----------+-------------+----------+--------+------+------------+---------+---------------+
全列匹配
EXPLAIN SELECT * FROM employees.titles WHERE emp_no = '10009' AND title = 'Senior Engineer' AND from_date = '1995-02-18'; +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+ | 1 | SIMPLE | titles | NULL | const | PRIMARY | PRIMARY | 159 | const,const,const | 1 | 100.00 | NULL | +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+
当按照索引中所有列进行精确匹配（这里精确匹配指 = 或 IN 匹配）时，索引可以被用到。
这里有一点需要注意，理论上索引对顺序是敏感的，但是由于 MySQL 的 查询优化器会自动调整 where 子句的条件顺序 以使用适合的索引，例如我们将 where 中的条件顺序颠倒：
EXPLAIN SELECT * FROM employees.titles WHERE from_date = '1995-02-18' AND emp_no IN ( '10009' ) AND title = 'Senior Engineer'; +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+ | 1 | SIMPLE | titles | NULL | const | PRIMARY | PRIMARY | 159 | const,const,const | 1 | 100.00 | NULL | +----+-------------+--------+------------+-------+---------------+---------+---------+-------------------+------+----------+-------+
和上面是一样的。
最左前缀匹配
EXPLAIN SELECT * FROM employees.titles WHERE emp_no = '10009'; +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------+ | 1 | SIMPLE | titles | NULL | ref | PRIMARY | PRIMARY | 4 | const | 3 | 100.00 | NULL | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------+
当查询条件精确匹配索引的 左边连续一个或几个列 时，如 <emp_no> 或 <emp_no, title>，所以可以被用到，但是只能用到一部分，即条件所组成的最左前缀。
上面的查询从分析结果看用到了 PRIMARY 索引，但是 key_len 为 4，说明只用到了索引的第一列前缀。
中间某个条件未提供
查询条件用到了索引中列的精确匹配，但是中间某个条件未提供。
EXPLAIN SELECT * FROM employees.titles WHERE emp_no='10009' AND from_date='1995-02-18'; +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | ref | PRIMARY | PRIMARY | 4 | const | 1 | 10.00 | Using where | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+
此时索引使用情况和情况二相同，因为 title 未提供，所以查询只用到了索引的第一列，而后面的 from_date 虽然也在索引中，但是由于 title 不存在而无法和左前缀连接，因此需要对结果进行扫描过滤 from_date（这里由于 emp_no 唯一，所以不存在扫描）。
如果想让 from_date 也使用索引而不是 where 过滤，可以增加一个 辅助索引 <emp_no, from_date>，此时上面的查询会使用这个索引。
除此之外，还可以使用一种称之为 隔离列 的优化方法，将 emp_no 与 from_date 之间的 _坑_ 填上。
首先我们看下 title 有几种不同的值：
SELECT DISTINCT(title) FROM employees.titles;
只有 7 种。在这种成为 _坑_ 的列值比较少的情况下，可以考虑用 IN 来填补这个 _坑_ 从而形成最左前缀：
EXPLAIN SELECT * FROM employees.titles WHERE emp_no='10009' AND title IN ('Senior Engineer', 'Staff', 'Engineer', 'Senior Staff', 'Assistant Engineer', 'Technique Leader', 'Manager') AND from_date='1995-02-18'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 159 | NULL | 7 | 100.00 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
这次 key_len 为 159，说明索引被用全了，但是从 type 和 rows 看出 IN 实际上执行了一个 range 查询，这里检查了 7 个 key。看下两种查询的性能比较：
SET profiling = 1; SELECT * FROM... -- 1 SELECT * FROM... -- 2 SHOW PROFILES; | Query_ID | Duration | Query | +----------+------------+---------+ | 1 | 0.00083950 | SELECT * ..| | 2 | 0.00063700 | SELECT * ..|
“填坑” 后性能提升了一点。如果经过 emp_no 筛选后余下很多数据，则后者性能优势会更加明显。当然，如果 title 的值很多，用填坑就不合适了，必须建立辅助索引。（笔者：多次测试后发现是有快有慢，可能是数据的原因，效果并不明显）
查询条件没有指定索引第一列
EXPLAIN SELECT * FROM employees.titles WHERE from_date='1995-02-18'; +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+ | 1 | SIMPLE | titles | NULL | ALL | NULL | NULL | NULL | NULL | 442605 | 10.00 | Using where | +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+
由于不是最左前缀，索引这样的查询显然用不到索引。
匹配某列的前缀字符串
EXPLAIN SELECT * FROM employees.titles WHERE emp_no=10009 AND title LIKE 'Senior%'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 156 | NULL | 1 | 100.00 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
此时可以用到索引，如果通配符 % 不出现在开头，则可以用到索引，但根据具体情况不同可能只会用其中一个前缀。
范围查询
EXPLAIN SELECT * FROM employees.titles WHERE emp_no < '10010' and title = 'Senior Engineer'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 4 | NULL | 14 | 10.00 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
范围列可以用到索引（必须是最左前缀），但是范围列后面的列无法用到索引。同时，索引最多用于一个范围列，因此如果查询条件中有两个范围列则无法全用到索引。
EXPLAIN SELECT * FROM employees.titles WHERE emp_no < '10010' AND title = 'Senior Engineer' AND from_date BETWEEN '1986-01-01' AND '1986-12-31'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 4 | NULL | 14 | 1.11 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
可以看到索引对第二个范围索引无能为力。这里特别要说明 MySQL 一个有意思的地方，那就是仅用 explain 可能无法区分 范围索引 和 多值匹配，因为在 type 中这两者都显示为 range。
同时，用了 between 并不意味着就是范围查询，例如下面的查询：
EXPLAIN SELECT * FROM employees.titles WHERE emp_no BETWEEN '10001' AND '10010' AND title='Senior Engineer' AND from_date BETWEEN '1986-01-01' AND '1986-12-31'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 159 | NULL | 15 | 1.11 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
看起来是用了两个范围查询，但作用于 emp_no 上的 BETWEEN 实际上相当于 IN，也就是说 emp_no 实际是多值精确匹配。可以看到这个查询用到了索引全部三个列。因此在 MySQL 中要谨慎地区分多值匹配和范围匹配，否则会对 MySQL 的行为产生困惑。
还有个值得注意的事情：
EXPLAIN SELECT * FROM employees.titles where emp_no > 10000 AND emp_no < 10011 AND title='Senior Engineer' AND from_date BETWEEN '1986-01-01' AND '1986-12-31'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 4 | NULL | 15 | 1.11 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ EXPLAIN SELECT * FROM employees.titles where emp_no >= 10001 and emp_no <= 10010 AND title='Senior Engineer' AND from_date BETWEEN '1986-01-01' AND '1986-12-31'; +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | range | PRIMARY | PRIMARY | 159 | NULL | 15 | 1.11 | Using where | +----+-------------+--------+------------+-------+---------------+---------+---------+------+------+----------+-------------+
疑问：
=影响 范围索引 还是 多值匹配？
查询条件中含有函数或表达式
很不幸，如果查询条件中含有函数或表达式，则 MySQL 不会为这列使用索引（虽然某些在数学意义上可以使用）。例如：
EXPLAIN SELECT * FROM employees.titles WHERE emp_no='10009' AND left(title, 6)='Senior'; +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+ | 1 | SIMPLE | titles | NULL | ref | PRIMARY | PRIMARY | 4 | const | 3 | 100.00 | Using where | +----+-------------+--------+------------+------+---------------+---------+---------+-------+------+----------+-------------+
虽然这个查询和情况五中功能相同，但是由于使用了函数 left，则无法为 title 列应用索引，而情况五中用 LIKE 则可以。
EXPLAIN SELECT * FROM employees.titles WHERE emp_no - 1=10000; +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+ | id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra | +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+ | 1 | SIMPLE | titles | NULL | ALL | NULL | NULL | NULL | NULL | 442605 | 100.00 | Using where | +----+-------------+--------+------------+------+---------------+------+---------+------+--------+----------+-------------+
显然这个查询等价于查询 emp_no 为 10001 的函数，但是由于查询条件是一个表达式，MySQL 无法为其使用索引。看来 MySQL 还没有智能到自动优化常量表达式的程度，因此在写查询语句时尽量避免表达式出现在查询中，而是先手工私下代数运算，转换为无表达式的查询语句。
References
– EOF –
