本文整理自:袋鼠云技术荟 | SQL优化案例(2):OR条件优化
数栈是云原生—站式数据中台PaaS,我们在github上有一个有趣的开源项目:https://github.com/DTStack/flinkx
FlinkX是一个基于Flink的批流统一的数据同步工具,既可以采集静态的数据,比如MySQL,HDFS等,也可以采集实时变化的数据,比如MySQL binlog,Kafka等,是全域、异构、批流一体的数据同步引擎,大家如果有兴趣,欢迎来github社区找我们玩~
在MySQL中,同样的查询条件,如果变换OR在SQL语句中的位置,那么查询的结果也会有差异,在较为复杂的情况下,可能会带来索引选择不佳的性能隐患,为了避免执行效率大幅度下降的问题,我们可以适当考虑使用Union all 对查询逻辑较为复杂的SQL进行分离。
常见OR使用场景,请阅读以下案例:
案例一:不同列使用OR条件查询
1. 待优化场景
SELECT
..
..
FROM`t1` a
WHERE a.token= '16149684'
AND a.store_id= '242950'
AND(a.registrationId IS NOT NULL
AND a.registrationId<> '')
OR a.uid= 308475
AND a.registrationId IS NOT NULL
AND a.registrationId<> ''
执行计划
+--------------+-----------------------+-----------------+----------------+-------------------+-------------------+---------------+----------------+---------------------------------------------+
| id | select_type | table | type | key | key_len | ref | rows | Extra |
+--------------+-----------------------+-----------------+----------------+-------------------+-------------------+---------------+----------------+---------------------------------------------+
| 1 | SIMPLE | a | range |idx_registrationid | 99 | | 100445 | Using index condition; Using where |
+--------------+-----------------------+-----------------+----------------+-------------------+-------------------+---------------+----------------+---------------------------------------------+
共返回1 行记录,花费 5 ms。
2. 场景解析
从查询条件中可以看出 token 和 uid 过滤性都非常好,但是由于使用了 or, 需要采用 index merge 的方法才能获得比较好的性能。但在实际执行过程中MySQL优化器默认选择了使用registrationId 上的索引,导致 SQL 的性能很差。
3. 场景优化
我们将SQL改写成union all的形式。
SELECT
...
...
FROM`t1` a
WHERE a.token = '16054473'
AND a.store_id = '138343'
AND b.is_refund = 1
AND (a.registrationId IS NOT NULL
AND a.registrationId <> '')
union all
SELECT
...
...
FROM`t1` a
where a.uid = 181579
AND a.registrationId IS NOT NULL
AND a.registrationId <> ''
+--------------+-----------------------+-----------------+----------------+------------------------------+---------------+-------------------+------------------------------+----------------+------------------------------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+--------------+-----------------------+-----------------+----------------+------------------------------+---------------+-------------------+------------------------------+----------------+------------------------------------+
| 1 | PRIMARY | a | ref | IDX_TOKEN,IDX_STORE_ID_TOKEN | IDX_TOKEN | 63 | const | 1 | Using index condition; Using where |
| 1 | PRIMARY | b | eq_ref | PRIMARY | PRIMARY | 4 | youdian_life_sewsq.a.role_id | 1 | Using where |
| 2 | UNION | a | const | PRIMARY | PRIMARY | 4 | const | 1 | |
| 2 | UNION | b | const | PRIMARY | PRIMARY | 4 | const | 0 | unique row not found |
| | UNION RESULT | <union1,2> | ALL | | | | | | Using temporary |
+--------------+-----------------------+-----------------+----------------+------------------------------+---------------+-------------------+------------------------------+----------------+------------------------------------+
共返回5 行记录,花费 5 ms。
通过对比优化前后的执行计划,可以明显看出,将SQL拆分成两个子查询,再使用union对结果进行合并,稳定性和安全性更好,性能更高。
案例二:同一列使用OR查询条件
1. 待优化场景
select
....
....
from
t1 as mci
left join t1 as ccv2_1 on ccv2_1.unique_no = mci=category_no1
left join t1 as ccv2_2 on ccv2_2.unique_no = mci=category_no2
left join t1 as ccv2_3 on ccv2_3.unique_no = mci=category_no3
left join(
select product_id,
count(0) count
from t2 pprod
inner join t3 pinfo on pinfo.promotion_id = pprod.promotion_id
and pprod.is_enable =1
and ppinfo.is_enable=1
and pinfo.belong_t0 =1
and pinfo.end_time >=now()
and not (
pinfo.onshelv_time>'2019-06-30 00:00:00'
or pinfo.end_time>'2018-12-05 00:00:00'
)group by pprod.product_id
)as pc on pc.product_id = mci.product_id
where mci.is_enable =0
and mci.comodifty_type in ('1', '5', '6')
and (pc.count =0 or pc.count isnull ) limit 0,5;
执行计划
2. 场景解析
本例的SQL查询中有一个子查询,子查询被当成驱动表,产生了auto_key,通过SQL拆分进行测试,验证主要是(pc.count =0 , or pc.count is null )会影响到整个SQL的性能,需要进行比较改写。
3. 场景优化
首先我们可以单独思考(pc.count =0 , or pc.count is null ) 如何进行优化?先写一个类似的SQL
Select col from test where col =100 or col is null;
+--------+
| col |
+--------+
| 100 |
| NULL |
+--------+
2 rows in set (0.00 sec)
这个时候我们看到的其实是同一个列,但对应不同的值,这种情况可以利用case when进行转换。
Select col From test where case when col is null then 100 else col =100 end;
+--------+
| col |
+--------+
| 100 |
| NULL |
+--------+
2 rows in set (0.00 sec)
再回到原始SQL进行改写。
select
....
....
from
t1 as mci
left join t1 as ccv2_1 on ccv2_1.unique_no = mci=category_no1
left join t1 as ccv2_2 on ccv2_2.unique_no = mci=category_no2
left join t1 as ccv2_3 on ccv2_3.unique_no = mci=category_no3
left join(
select product_id,
count(0) count
from t2 pprod
inner join t3 pinfo on pinfo.promotion_id = pprod.promotion_id
and pprod.is_enable =1
and ppinfo.is_enable=1
and pinfo.belong_t0 =1
and pinfo.end_time >=now()
and not (
pinfo.onshelv_time>'2019-06-30 00:00:00'
or pinfo.end_time>'2018-12-05 00:00:00'
)group by pprod.product_id
)as pc on pc.product_id = mci.product_id
where mci.is_enable =0
and mci.comodifty_type in ('1', '5', '6')
and case when pc.count is null then 0 else pc.count end=0 limit 0,5;
可以看出优化后的SQL比原始SQL快了30秒,执行效率提升约50倍。
案例三:优化关联SQL OR条件
1. 待优化场景
SELECT user_msg.msg_id AS ‘msg_id’, user_msg.content AS ‘msg_content’, …
FROM user_msg
LEFT JOIN user ON user_msg.user_id = user.user_id
LEFT JOIN group ON user_msg.group_id = group.group_id
WHERE user_msg.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL30SECOND)
OR user.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL 30 SECOND)
OR group.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL 30 SECOND)
2.场景解析
我们仔细分析上述查询语句,发现虽然业务逻辑只需要查询半分钟内修改的数据,但执行过程却必须对所有的数据进行关联操作,带来不必要的性能损耗。
3.场景优化
我们对原始SQL进行拆分操作,第一部分sql-01如下:
SELECT user_msg.msg_id AS ‘msg_id’, user_msg.content AS ‘msg_content’, …
FROM user_msg
LEFT JOIN user ON user_msg.user_id = user.user_id
LEFT JOIN group ON user_msg.group_id = group.group_id
WHERE user_msg.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL 30 SECOND)
sql-01以user_msg 表为驱动,使用gmt_modified 索引过滤最新数据。
第二部分sql-02如下:
SELECT user_msg.msg_id AS ‘msg_id’, user_msg.content AS ‘msg_content’, …
FROM user_msg
LEFT JOIN user ON user_msg.user_id = user.user_id
LEFT JOIN group ON user_msg.group_id = group.group_id
WHERE user.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL 30 SECOND)
ql-02以user为驱动表,msg user_id 的索引过滤行很好。
第三部分sql-03如下:
SELECT user_msg.msg_id AS ‘msg_id’, user_msg.content AS ‘msg_content’, …
FROM user_msg
LEFT JOIN user ON user_msg.user_id = user.user_id
LEFT JOIN group ON user_msg.group_id = group.group_id
WHERE group.gmt_modified >= date_sub('2018-03-29 09:31:44', INTERVAL 30 SECOND)
sql-03以group为驱动表,使用gmt_modified 索引过滤最新数据。
总结
MySQL OR条件优化的常见场景主要有以下情况:
1、相同列可以使用IN进行代替
2、不同列及复杂的情况下,可以使用union all 进行分离
3、关联SQL OR条件
我们需要结合实际场景,分析优化。