"一、 SQL 及分析查询SQL语句如下： select sum(l_extendedprice * (1 - l_discount)) as revenue from lineitem,pa .."

sjr 北京
乾学院 32 号会员
1 回帖 • 747 浏览 • 1 年前

用 TPCH 练习性能优化 Q19

计算＆AI

一、 SQL及分析

查询SQL语句如下：

select
    sum(l_extendedprice * (1 - l_discount)) as revenue
from
    lineitem,part
where
    (
        p_partkey = l_partkey
        and p_brand = 'Brand#32'
        and p_container in ('SM CASE', 'SM BOX', 'SM PACK', 'SM PKG')
        and l_quantity >= 7 and l_quantity <= 7 + 10
        and p_size between 1 and 5
        and l_shipmode in ('AIR', 'AIR REG')
        and l_shipinstruct = 'DELIVER IN PERSON'
    )
    or
    (
        p_partkey = l_partkey
        and p_brand = 'Brand#23'
        and p_container in ('MED BAG', 'MED BOX', 'MED PKG', 'MED PACK')
        and l_quantity >= 18 and l_quantity <= 18 + 10
        and p_size between 1 and 10
        and l_shipmode in ('AIR', 'AIR REG')
        and l_shipinstruct = 'DELIVER IN PERSON'
    )
    or
    (
        p_partkey = l_partkey
        and p_brand = 'Brand#45'
        and p_container in ('LG CASE', 'LG BOX', 'LG PACK', 'LG PKG')
        and l_quantity >= 22 and l_quantity <= 22 + 10
        and p_size between 1 and 15
        and l_shipmode in ('AIR', 'AIR REG')
        and l_shipinstruct = 'DELIVER IN PERSON'
    );

这是两表连接的过滤查询汇总，结构上比较简单，但是过滤条件比较复杂。

二、 SPL实现

条件表达式中有不少项只与较小的表part相关，而且这些项的计算复杂度还较高（有in运算），如果先针对part表把这些项计算出来，则计算量将比针对两表连接结果集再计算要少得多，因为连接结果集是和lineitem表规模相同，要大得多。

	A
1	=now()
2	>brand1="Brand#32"
3	>brand2="Brand#23"
4	>brand3="Brand#45"
5	>quantity1=7
6	>quantity2=18
7	>quantity3=22
8	=["SM CASE", "SM BOX", "SM PACK", "SM PKG"]
9	=["MED BAG", "MED BOX", "MED PKG", "MED PACK"]
10	=["LG CASE", "LG BOX", "LG PACK", "LG PKG"]
11	=["AIR","AIR REG"]
12	=file("part.ctx").open().cursor@m(P_PARTKEY,P_BRAND,P_SIZE,P_CONTAINER,0:FLAG;P_SIZE>=1)
13	=A12.run(FLAG=if(P_BRAND==brand1 && A8.contain(P_CONTAINER) && P_SIZE<=5:1, P_BRAND==brand2 && A9.contain(P_CONTAINER) && P_SIZE<=10:2, P_BRAND==brand3 && A10.contain(P_CONTAINER) && P_SIZE<=15:3; 0)).select(FLAG>0).fetch@o().keys@im(P_PARTKEY)
14	=file("lineitem.ctx").open().cursor@m(L_PARTKEY,L_QUANTITY,L_EXTENDEDPRICE,L_DISCOUNT;L_PARTKEY:A13,L_SHIPINSTRUCT=="DELIVER IN PERSON" && A11.contain(L_SHIPMODE) && L_QUANTITY>= quantity1 && L_QUANTITY<=quantity3+10)
15	=A14.groups(;sum(if(case(L_PARTKEY.FLAG,1:L_QUANTITY<=quantity1+10,2:L_QUANTITY>=quantity2 && L_QUANTITY<=quantity2+10;L_QUANTITY>=quantity3),L_EXTENDEDPRICE*(1-L_DISCOUNT),0)))
16	=interval@ms(A1,now())

A12在part表中新产生一个FLAG字段，用于将part相关的复杂条件项先计算出来。当记录满足SQL中的or条件1、2、3时分别赋值为1、2、3，都不满足则赋值0，然后用FLAG值是否大于0来过滤。在A15中再根据FLAG的值来选择用哪个quantity值对L_QUANTITY进行过滤。FLAG相关的复杂计算式只需要计算part表的行数次。

三、进一步优化

1. 优化方法

本例中要使用Q1中介绍的字符串整数化方法，part中的P_BRAND和P_CONTAINER已分别在Q16、Q17中转换过了，lineitem中的L_SHIPMODE已经在Q12中转换过了，本例中还需要转换L_SHIPINSTRUCT。还要使用Q2中介绍的维表主键序号化方法，part中的P_PARTKEY、lineitem中的L_PARTKEY都已在之前的例子中转换过了。

2. 数据转换代码

2.1 part转换

复制part_17.ctx重命名为part_19.ctx。

2.2 lineitem转换

	A
1	=file("lineitem.ctx").open().cursor(L_SHIPINSTRUCT)
2	=A1.id(L_SHIPINSTRUCT).sort()
3	=file("l_shipinstruct.txt").export(A2)
4	=file("lineitem_18.ctx").open().cursor()
5	=A4.run(L_SHIPINSTRUCT=A2.pos@b(L_SHIPINSTRUCT))
6	=file("lineitem_19.ctx").create@p(#L_ORDERKEY,#L_LINENUMBER,L_PARTKEY, L_SUPPKEY, L_QUANTITY, L_EXTENDEDPRICE,L_DISCOUNT, L_TAX, L_RETURNFLAG, L_LINESTATUS, L_SHIPDATE, L_COMMITDATE, L_RECEIPTDATE,L_SHIPINSTRUCT, L_SHIPMODE, L_COMMENT)
7	>A6.append(A5)

3. 数据转换后的计算代码

维表和字符串字段取值列表需要预加载，加载代码如下：

	A
1	>env(p_brand,file("p_brand.txt").import@si())
2	>env(p_container,file("p_container.txt").import@si())
3	>env(l_shipinstruct,file("l_shipinstruct.txt").import@si())
4	>env(l_shipmode,file("l_shipmode.txt").import@si())
5	>env(part, file("part_19.ctx").open().import())

在查询计算之前，需要先运行预加载代码，将小维表读入内存。

计算代码：

	A
1	=now()
2	>brand1=p_brand.pos@b("Brand#32")
3	>brand2=p_brand.pos@b("Brand#23")
4	>brand3=p_brand.pos@b("Brand#45")
5	>quantity1=7
6	>quantity2=18
7	>quantity3=22
8	=p_container.(["SM CASE", "SM BOX", "SM PACK", "SM PKG"].contain(~))
9	=p_container.(["MED BAG", "MED BOX", "MED PKG", "MED PACK"].contain(~))
10	=p_container.(["LG CASE", "LG BOX", "LG PACK", "LG PKG"].contain(~))
11	=l_shipmode.(["AIR","AIR REG"].contain(~))
12	=l_shipinstruct.pos@b("DELIVER IN PERSON")
13	=part.@m(if(P_SIZE<1:null, P_BRAND==brand1 && A8(P_CONTAINER) && P_SIZE<=5:1, P_BRAND==brand2 && A9(P_CONTAINER) && P_SIZE<=10:2, P_BRAND==brand3 && A10(P_CONTAINER) && P_SIZE<=15:3; null))
14	=file("lineitem_19.ctx").open().cursor@m(L_PARTKEY,L_QUANTITY,L_EXTENDEDPRICE,L_DISCOUNT;L_PARTKEY:A13:#,L_SHIPINSTRUCT==A12 && A11(L_SHIPMODE) && L_QUANTITY>= quantity1 && L_QUANTITY<=quantity3+10)
15	=A14.groups(;sum(if(case(L_PARTKEY,1:L_QUANTITY<=quantity1+10,2:L_QUANTITY>=quantity2 && L_QUANTITY<=quantity2+10,3:L_QUANTITY>=quantity3),L_EXTENDEDPRICE*(1-L_DISCOUNT),0)))
16	=interval@ms(A1,now())

四、使用企业版列式计算

1. 原始数据

	A
1	=now()
2	>brand1="Brand#32"
3	>brand2="Brand#23"
4	>brand3="Brand#45"
5	>quantity1=7
6	>quantity2=18
7	>quantity3=22
8	=["SM CASE", "SM BOX", "SM PACK", "SM PKG"].i()
9	=["MED BAG", "MED BOX", "MED PKG", "MED PACK"].i()
10	=["LG CASE", "LG BOX", "LG PACK", "LG PKG"].i()
11	=["AIR","AIR REG"].i()
12	=file("part.ctx").open().cursor@mv(P_PARTKEY,P_BRAND,P_SIZE,P_CONTAINER;P_SIZE>=1)
13	=A12.derive@o(if(P_BRAND==brand1 && A8.contain(P_CONTAINER) && P_SIZE<=5:1, P_BRAND==brand2 && A9.contain(P_CONTAINER) && P_SIZE<=10:2, P_BRAND==brand3 && A10.contain(P_CONTAINER) && P_SIZE<=15:3; 0):FLAG).select@v(FLAG>0).fetch@o().keys@im(P_PARTKEY)
14	=file("lineitem.ctx").open().cursor@mv(L_PARTKEY,L_QUANTITY,L_EXTENDEDPRICE,L_DISCOUNT;L_PARTKEY:A13,L_QUANTITY>= quantity1 && L_QUANTITY<=quantity3+10 && L_SHIPINSTRUCT=="DELIVER IN PERSON" && A11.contain(L_SHIPMODE))
15	=A14.groups(;sum(if(case(L_PARTKEY.FLAG,1:L_QUANTITY<=quantity1+10,2:L_QUANTITY>=quantity2 && L_QUANTITY<=quantity2+10;L_QUANTITY>=quantity3),L_EXTENDEDPRICE*(1-L_DISCOUNT),0)))
16	=interval@ms(A1,now())

2. 优化后数据