MySQL 的SQL引擎很差吗? 由一个同学提出问题引出的实验

MySQL
24
0
0
2024-11-01
标签   MySQL优化

有了PostgreSQL的出现,MySQL的数据库在SQL的处理上的问题一直被人当做有意思的事情来去谈论,实际上每种数据库有自己不同的个性,我们掌握就好,无需特别的进行一些情感上的好恶。

MySQL 最近一个同学给我提了一个问题,关于为什么一个简单的语句,并且语句中提取的数据是唯一的一条,而却在下面循环了三次,因为我没有他的数据,也仅仅是看了执行计划和语句,又因为是截图,所以只是简单的看了看。

不过今天转念一想,到底MySQL 8 的数据处理部分,基于我之前在MySQL 5.6 5.7等积累的一些关于 子查询差的口碑,是否被改善了的事情又重新让我想起了,8.0 是不是更好,随即使用了8.031版本的MySQL 数据库。

这里我们使用了一个MYSQL 的emaple数据库,并编造了一段可以重新写成两种方式的SQL ,我们直接来看一下.

select o.*
from orders as o 
left join orderdetails as d on o.ordernumber = d.ordernumber and o.ordernumber = (select ordernumber  
from orders where requiredDate > '2003-01-12' and requiredDate < '2003-01-14')
where o.requiredDate > '2003-01-12' and o.requiredDate < '2003-01-14';

| -> Nested loop left join  (cost=1.36 rows=4) (actual time=0.070..0.075 rows=4 loops=1)
    -> Index range scan on o using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14'), with index condition: ((o.requiredDate > DATE'2003-01-12') and (o.requiredDate < DATE'2003-01-14'))  (cost=0.71 rows=1) (actual time=0.060..0.062 rows=1 loops=1)
    -> Filter: ((o.orderNumber = (select #2)) and (d.orderNumber = (select #2)))  (cost=0.65 rows=4) (actual time=0.009..0.011 rows=4 loops=1)
        -> Covering index lookup on d using PRIMARY (orderNumber=(select #2))  (cost=0.65 rows=4) (actual time=0.007..0.010 rows=4 loops=1)
        -> Select #2 (subquery in condition; run only once)
            -> Filter: ((orders.requiredDate > DATE'2003-01-12') and (orders.requiredDate < DATE'2003-01-14'))  (cost=0.46 rows=1) (actual time=0.028..0.032 rows=1 loops=1)
                -> Covering index range scan on orders using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14')  (cost=0.46 rows=1) (actual time=0.027..0.030 rows=1 loops=1)
        -> Select #2 (subquery in condition; run only once)
            -> Filter: ((orders.requiredDate > DATE'2003-01-12') and (orders.requiredDate < DATE'2003-01-14'))  (cost=0.46 rows=1) (actual time=0.028..0.032 rows=1 loops=1)
                -> Covering index range scan on orders using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14')  (cost=0.46 rows=1) (actual time=0.027..0.030 rows=1 loops=1)
 |

第二个写法

select o.*
from orders as o 
left join orderdetails as d on o.ordernumber = d.ordernumber 
left join orders as o2 on o2.ordernumber = d.ordernumber and o2.requiredDate > '2003-01-12' and o2.requiredDate < '2003-01-14' 
where o.requiredDate > '2003-01-12' and o.requiredDate < '2003-01-14';

| -> Nested loop left join  (cost=5.11 rows=9) (actual time=0.039..0.047 rows=4 loops=1)
    -> Nested loop left join  (cost=1.89 rows=9) (actual time=0.034..0.039 rows=4 loops=1)
        -> Index range scan on o using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14'), with index condition: ((o.requiredDate > DATE'2003-01-12') and (o.requiredDate < DATE'2003-01-14'))  (cost=0.71 rows=1) (actual time=0.023..0.025 rows=1 loops=1)
        -> Covering index lookup on d using PRIMARY (orderNumber=o.orderNumber)  (cost=1.18 rows=9) (actual time=0.009..0.012 rows=4 loops=1)
    -> Filter: ((o2.requiredDate > '2003-01-12') and (o2.requiredDate < '2003-01-14'))  (cost=0.26 rows=1) (actual time=0.001..0.002 rows=1 loops=4)
        -> Single-row index lookup on o2 using PRIMARY (orderNumber=d.orderNumber)  (cost=0.26 rows=1) (actual time=0.001..0.001 rows=1 loops=4)


从下面的截图语句,可以看到两种写法的语句均得到一样的逻辑数据结果,但是相关的语句的执行计划完全不同。

1 子查询类型的方式撰写的语句

select o.*

from orders as o

left join orderdetails as d on o.ordernumber = d.ordernumber and o.ordernumber = (select ordernumber

from orders where requiredDate > '2003-01-12' and requiredDate < '2003-01-14')

where o.requiredDate > '2003-01-12' and o.requiredDate < '2003-01-14';

从上面的语句中可以看到,将其中一个left join中的条件对接到子查询中,这里可以看到子查询得出的数据一定是一条数据,否则语句就会报错。

但从执行计划,我们明显可以看出,对于日期的部分,在执行计划中,出现了三次。

其中的两次是一致的,

-> Select #2 (subquery in condition; run only once)

-> Filter: ((orders.requiredDate > DATE'2003-01-12') and (orders.requiredDate < DATE'2003-01-14')) (cost=0.46 rows=1) (actual time=0.028..0.032 rows=1 loops=1)

-> Covering index range scan on orders using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14') (cost=0.46 rows=1) (actual time=0.027..0.030 rows=1 loops=1)

-> Select #2 (subquery in condition; run only once)

-> Filter: ((orders.requiredDate > DATE'2003-01-12') and (orders.requiredDate < DATE'2003-01-14')) (cost=0.46 rows=1) (actual time=0.028..0.032 rows=1 loops=1)

-> Covering index range scan on orders using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14') (cost=0.46 rows=1) (actual time=0.027..0.030 rows=1 loops=1)

|

出现这样的情况主要是因为,在MySQL 在通过子查询查询数据的时候,会在子查询内部重新评估条件,并产生相关的执行计划,导致重复的查询计划生成的操作,同时由于查询的复杂度,尤其子查询的嵌入会增加优化器生成和选择执行计划的难度,在这样的情况下优化器会多次进行过滤,还有在MySQL在进行执行计划产生的时候会尝试不同的执行计划并在其中做出优化策略的权衡时为保证查询结果的正确性,产生多次过滤方式保证最终结果的正确性。

而反观我们不在使用子查询撰写的语句在执行计划中并未有重复的日志数据的过滤,并且从执行时间上看,也要比子查询的方式查询的速度要快。

select o.*

from orders as o

left join orderdetails as d on o.ordernumber = d.ordernumber

left join orders as o2 on o2.ordernumber = d.ordernumber and o2.requiredDate > '2003-01-12' and o2.requiredDate < '2003-01-14'

where o.requiredDate > '2003-01-12' and o.requiredDate < '2003-01-14';

| -> Nested loop left join (cost=5.11 rows=9) (actual time=0.039..0.047 rows=4 loops=1)

-> Nested loop left join (cost=1.89 rows=9) (actual time=0.034..0.039 rows=4 loops=1)

-> Index range scan on o using idx_requiredDate over ('2003-01-12' < requiredDate < '2003-01-14'), with index condition: ((o.requiredDate > DATE'2003-01-12') and (o.requiredDate < DATE'2003-01-14')) (cost=0.71 rows=1) (actual time=0.023..0.025 rows=1 loops=1)

-> Covering index lookup on d using PRIMARY (orderNumber=o.orderNumber) (cost=1.18 rows=9) (actual time=0.009..0.012 rows=4 loops=1)

-> Filter: ((o2.requiredDate > '2003-01-12') and (o2.requiredDate < '2003-01-14')) (cost=0.26 rows=1) (actual time=0.001..0.002 rows=1 loops=4)

-> Single-row index lookup on o2 using PRIMARY (orderNumber=d.orderNumber) (cost=0.26 rows=1) (actual time=0.001..0.001 rows=1 loops=4)

通过此示例我们可以得出如下的一些结论

1 在MySQL中如果可以将子查询改为连接查询的,尽量改为连接查询 ,通常连接查询的被正确翻译并走更优的执行计划的可能性更高。

2 左连接通常比子查询更适合大数据量的情况,子查询会产生中间结果集,导致内存压力增大,和查询性能的下降,左连接可以更好的解决由于子查询在产生的子结果集较大时产生的问题,减少中间结果集的产生,提高执行效率。

这同时也体现了,MySQL SQL 处理引擎,在SQL的解析上应还有更多的进步的空间,在SQL转换为内部数据结果时,对于子查询上的语义的解读上还有改善的空间。