项目中一般使用的都是单表查询，但是在一些业务场景下，偶尔会选择联表查询，一直对联表查询时如何使用索引一直感到很好奇。正好近期项目中遇到一个问题，联表查询时，没有建立索引，耗时居然达到了可耻的10分钟，所以趁机了解了一下。
表数据
一共3张表knowledge, knowledge_question, knowledge_answer，数据在6000~10000之间。
knowledge： 6126
knowledge_question：9647
knowledge_answer：8267
执行的语句：
SELECT DISTINCT(k.base_id) FROM knowledge AS k
LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id
LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id
WHERE k.update_time>'2019-01-01 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00'
没有索引（只有主键）
mysql > SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id WHERE k.update_time>'2019-01-01 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00';
+---------+
| base_id |
+---------+
| 159 |
...
| 413 |
| 414 |
+---------+
145 rows in set, 3 warnings (9 min 26.57 sec)
执行时间约10分钟，查看执行计划如下：
mysql > explain SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id WHERE k.update_time>'2019-01-01 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00';
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+--------------------------------------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+--------------------------------------------------------------+
| 1 | SIMPLE | k | NULL | ALL | NULL | NULL | NULL | NULL | 6238 | 100.00 | Using temporary |
| 1 | SIMPLE | q | NULL | ALL | NULL | NULL | NULL | NULL | 9540 | 100.00 | Using where; Distinct; Using join buffer (Block Nested Loop) |
| 1 | SIMPLE | a | NULL | ALL | NULL | NULL | NULL | NULL | 8410 | 100.00 | Using where; Distinct; Using join buffer (Block Nested Loop) |
+----+-------------+-------+------------+------+---------------+------+---------+------+------+----------+--------------------------------------------------------------+
3 rows in set, 4 warnings (0.00 sec)
全部都是全表扫描，根据MySQL联表查询的算法Nested-Loop Join，MySQL查询的结果集是3张表的笛卡尔积，所以效率特别低。
JOIN字段建立索引
explain SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id WHERE k.update_time>'2019-01-01 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00';
+----+-------------+-------+------------+------+---------------+---------+---------+-------------------+------+----------+------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+---------------+---------+---------+-------------------+------+----------+------------------------------+
| 1 | SIMPLE | k | NULL | ALL | PRIMARY | NULL | NULL | NULL | 6444 | 33.33 | Using where; Using temporary |
| 1 | SIMPLE | a | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
| 1 | SIMPLE | q | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
+----+-------------+-------+------------+------+---------------+---------+---------+-------------------+------+----------+------------------------------+
执行结果
+---------+
| base_id |
+---------+
| 159 |
...
| 413 |
| 414 |
+---------+
145 rows in set (0.02 sec)
耗时变成20毫秒
Where条件建立索引
给Where条件建立索引，并不一定会使用。
比如：在表knowledge的字段update上建立索引idx_time：
MySQL [knowledge_base]> alter table knowledge add index idx_time(update_time);
MySQL [knowledge_base]> explain SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id WHERE k.update_time>'2019-01-03 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00';
+----+-------------+-------+------------+------+------------------+---------+---------+-------------------+------+----------+------------------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+------------------+---------+---------+-------------------+------+----------+------------------------------+
| 1 | SIMPLE | k | NULL | ALL | PRIMARY,idx_time | NULL | NULL | NULL | 6444 | 19.01 | Using where; Using temporary |
| 1 | SIMPLE | a | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
| 1 | SIMPLE | q | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
+----+-------------+-------+------------+------+------------------+---------+---------+-------------------+------+----------+------------------------------+
结果执行上来看，并没有使用索引idx_time。
如果where条件从k.update_time>'2019-01-03 12:00:00'修改为k.update_time='2019-01-03 12:00:00'（从>变成=）
MySQL [knowledge_base]> explain SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q.knowledge_id LEFT JOIN knowledge_answer AS a ON k.id=a.knowledge_id WHERE k.update_time='2019-01-03 12:00:00' AND q.update_time>'2019-01-01 12:00:00' AND a.update_time>'2019-01-01 12:00:00';
+----+-------------+-------+------------+------+------------------+----------+---------+-------------------+------+----------+-----------------------+
| id | select_type | table | partitions | type | possible_keys | key | key_len | ref | rows | filtered | Extra |
+----+-------------+-------+------------+------+------------------+----------+---------+-------------------+------+----------+-----------------------+
| 1 | SIMPLE | k | NULL | ref | PRIMARY,idx_time | idx_time | 4 | const | 1 | 100.00 | Using temporary |
| 1 | SIMPLE | a | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
| 1 | SIMPLE | q | NULL | ref | idx_kid | idx_kid | 4 | knowledge_base.k.id | 1 | 33.33 | Using where; Distinct |
+----+-------------+-------+------------+------+------------------+----------+---------+-------------------+------+----------+-----------------------+
则会使用索引idx_time
继续试验发现，如果在
knowledge_question和knowledge_answer表上的字段update_time上建立索引，有时候会较大幅度的改变执行计划。 所以说，检查SQL语句是否用到索引，一定要用explain查看执行计划，MySQL优化器做了太多的工作了。
其他知识点
在建立索引的时候，会遇到Table Metadata Lock的问题，可以先show processlist，找到占用表锁的连接，然后kill。
MySQL [(none)]> show processlist;
+---------+-----------+----------------------+--------------+---------+------+--------------+------------------------------------------------------------------------------------------------------+
| Id | User | Host | db | Command | Time | State | Info |
+---------+-----------+----------------------+--------------+---------+------+--------------+------------------------------------------------------------------------------------------------------+
| 3468722 | Aics_user | 10.219.153.217:46574 | knowledge_base | Query | 94 | Sending data | SELECT DISTINCT(k.base_id) FROM knowledge AS k LEFT JOIN knowledge_question AS q ON k.id=q |
MySQL [(none)]> kill 3468722
结论
- 关联字段一定要添加索引
- where条件的索引建立，一定要查看explain，mysql的工作方式经常跟我们想的不一样
- 增加慢查询日志（dba呢？）
参考
- 关于 MySQL LEFT JOIN 你可能需要了解的三点
- MySQL JOIN原理
- MySQL查询优化——连接以及连接原理
- MySQL 性能优化神器 Explain 使用分析
- What is the meaning of filtered in MySQL explain?