DM SQL关联列 like 优化案例

1.1、sql优化背景

达梦一哥们找我优化条SQL，反馈在DM8数据库执行时间很慢出不来结果，监控工具显示这条SQL的执行时间需要20多万毫秒，安排。

1.2、慢sql和执行时间

select a.col1 as d_id,

a.col2 as s_id,

a.col3 as bm,

a.col4,

a.col5,

(select b.col1 from table2 b where b.col_itname = 'zb1' and b.col1 = a.col20) as bb,

a.col6 as dzzlxr,

a.col7 as dzzlxdh,

(select b.col1 from table2 b where b.col_itname = 'zb2' and b.col1 = a.col21) as bc,

(select b.col1 from table2 b where b.col_itname = 'zb3' and b.col1 = a.col22) as cb,

a.col8,

date_format(a.col9, '%Y-%m-%d %H:%i:%s') as gx,

a.col10 as cid,

a.col11 as tp,

(select b.col5 from table1 b where b.col1 = a.col2) as sj,

(select count(*) from table3 dy left join table1 dzz on dy.col1 = dzz.col1 where dzz.col11 like concat(a.col11,'%')) as rc

from table1 a

where 1 = 1

and a.col1 in ( /* 这里 in 了 600 个 字符串条件 */ );

100条执行成功, 执行耗时1分 28秒 248毫秒. 执行号:1432757809

1.3、慢sql执行计划

1 #NSET2: [1330892675, 12345, 692]

2 #PIPE2: [1330892675, 12345, 692]

3 #PIPE2: [1330892669, 12345, 692]

4 #PIPE2: [1330892663, 12345, 692]

5 #PIPE2: [1330892657, 12345, 692]

6 #PIPE2: [1330892648, 12345, 692]

7 #PRJT2: [4, 12345, 692]; exp_num(17), is_atom(FALSE)

8 #NEST LOOP INDEX JOIN2: [4, 12345, 692]

9 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

10 #BLKUP2: [3, 1, 0]; INDEX33571964(A)

11 #SSEK2: [3, 1, 0]; scan_type(ASC), INDEX33571964(table1 as A), scan_range[DMTEMPVIEW_22201688.colname,DMTEMPVIEW_22201688.colname]

12 #SPL2: [1330892644, 1, 852]; key_num(2), spool_num(4), is_atom(FALSE), has_variable(0)

13 #PRJT2: [1330892644, 1, 852]; exp_num(3), is_atom(FALSE)

14 #HAGR2: [1330892644, 1, 852]; grp_num(1), sfun_num(3); slave_empty(0) keys(A.ROWID)

15 #NEST LOOP LEFT JOIN2: [1327131762, 71772595, 852]; join condition(DZZ.col11 LIKE exp11) partition_keys_num(0) ret_null(0)

16 #NEST LOOP INDEX JOIN2: [4, 12345, 692]

17 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

18 #BLKUP2: [3, 1, 0]; INDEX33571964(A)

19 #SSEK2: [3, 1, 0]; scan_type(ASC), INDEX33571964(table1 as A), scan_range[DMTEMPVIEW_22201689.colname,DMTEMPVIEW_22201689.colname]

20 #HASH2 INNER JOIN: [26, 116278, 160]; LKEY_UNIQUE KEY_NUM(1); KEY(DZZ.col1=DY.col1) KEY_NULL_EQU(0)

21 #CSCN2: [1, 12345, 104]; INDEX33571530(table1 as DZZ)

22 #SSCN: [13, 116278, 56]; IDX_DYJBXX_ORGID(table3 as DY)

23 #SPL2: [9, 9876, 740]; key_num(2), spool_num(3), is_atom(FALSE), has_variable(0)

24 #PRJT2: [9, 9876, 740]; exp_num(2), is_atom(FALSE)

25 #HASH RIGHT SEMI JOIN2: [9, 9876, 740]; n_keys(1) KEY(DMTEMPVIEW_22201694.colname=A.col1) KEY_NULL_EQU(0)

26 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

27 #HASH2 INNER JOIN: [9, 9876, 740]; LKEY_UNIQUE KEY_NUM(1); KEY(B.col1=A.col2) KEY_NULL_EQU(0)

28 #CSCN2: [1, 12345, 96]; INDEX33571530(table1 as B)

29 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

30 #SPL2: [5, 11618, 740]; key_num(2), spool_num(2), is_atom(FALSE), has_variable(0)

31 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

32 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201695.colname=A.col1) KEY_NULL_EQU(0)

33 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

34 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col22) KEY_NULL_EQU(0)

35 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb3',min),('zb3',max))

36 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

37 #SPL2: [5, 11618, 740]; key_num(2), spool_num(1), is_atom(FALSE), has_variable(0)

38 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

39 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201696.colname=A.col1) KEY_NULL_EQU(0)

40 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

41 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col21) KEY_NULL_EQU(0)

42 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb2',min),('zb2',max))

43 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

44 #SPL2: [5, 11618, 740]; key_num(2), spool_num(0), is_atom(FALSE), has_variable(0)

45 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

46 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201697.colname=A.col1) KEY_NULL_EQU(0)

47 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

48 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col20) KEY_NULL_EQU(0)

49 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb1',min),('zb1',max))

50 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

1.4、涉及表的数据量

select count(1) from table1

union all

select count(1) from table2

union all

select count(1) from table3;

1.5、分析过程

用瞪眼大法观察，目测是这几段标量子查询导致慢的（啥是瞪眼大法？问就是优化这么多案例的经验）

(select b.col1 from table2 b where b.col_itname = 'zb1' and b.col1 = a.col20) as bb,

(select b.col1 from table2 b where b.col_itname = 'zb2' and b.col1 = a.col21) as bc,

(select b.col1 from table2 b where b.col_itname = 'zb3' and b.col1 = a.col22) as cb,

(select count(*) from table3 dy left join table1 dzz on dy.col1 = dzz.col1 where dzz.col11 like concat(a.col11,'%')) as rc

每段标量子查询测试后，发现是最后一段标量子查询缓慢导致

-- (select b.col1 from table2 b where b.col_itname = 'zb1' and b.col1 = a.col20) as bb,

-- (select b.col1 from table2 b where b.col_itname = 'zb2' and b.col1 = a.col21) as bc,

-- (select b.col1 from table2 b where b.col_itname = 'zb3' and b.col1 = a.col22) as cb,

(select count(*) from table3 dy left join table1 dzz on dy.col1 = dzz.col1 where dzz.col11 like concat(a.col11,'%')) as rc

做了个测试，如果将 like 改成 = 的话，非常快出结果

(select count(*) from table3 dy left join table1 dzz on dy.col1 = dzz.col1 where dzz.col11 = a.col11 ) as rc

dzz.col11 字段是有索引，尝试过各种手段都用不上，只能改写SQL。

2.1、SQL等价改写

我想法就是将 like 关联这种模糊态查询改成 = 这种确定态的精准匹配逻辑，想了好几个小时都没什么头绪。

后面只能去翻翻落总博客，卧槽，还没想到真的给我看到类似的case ，瞬间有了灵感做了下面改写：

select a.col1 as d_id,

a.col2 as s_id,

a.col3 as bm,

a.col4,

a.col5,

(select b.col1 from table2 b where b.col_itname = 'zb1' and b.col1 = a.col20) as bb,

a.col6 as dzzlxr,

a.col7 as dzzlxdh,

(select b.col1 from table2 b where b.col_itname = 'zb2' and b.col1 = a.col21) as bc,

(select b.col1 from table2 b where b.col_itname = 'zb3' and b.col1 = a.col22) as cb,

a.col8,

date_format(a.col9, '%Y-%m-%d %H:%i:%s') as gx,

a.col10 as cid,

a.col11 as tp,

(select b.col5 from table1 b where b.col1 = a.col2) as sj,

b.cnt as rc

from table1 a

LEFT JOIN (

SELECT COUNT(*) cnt,

dzz.col11

FROM table3 dy

LEFT JOIN table1 dzz

ON dy.col1 = dzz.col1

GROUP BY dzz.col11

) b ON SUBSTR(b.col11, 1, LENGTH(a.col11)) = a.col11

where 1 = 1

and a.col1 in (

-- 这里 in 了 600 个 字符串条件

);

100条执行成功, 执行耗时5秒 326毫秒. 执行号:1435485506

改写完后5秒左右就能出结果了，差集比对后也是等价，呦西。

2.2、SQL改写后执行计划

1 #NSET2: [524737849, 358862, 740]

2 #PIPE2: [524737849, 358862, 740]

3 #PIPE2: [524737843, 358862, 740]

4 #PIPE2: [524737837, 358862, 740]

5 #PIPE2: [524737831, 358862, 740]

6 #PRJT2: [524737822, 358862, 740]; exp_num(16), is_atom(FALSE)

7 #NEST LOOP LEFT JOIN2: [524737822, 358862, 740]; join condition(A.col11 = exp11) partition_keys_num(0) ret_null(0)

8 #NEST LOOP INDEX JOIN2: [4, 12345, 692]

9 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

10 #BLKUP2: [3, 1, 0]; INDEX33571964(A)

11 #SSEK2: [3, 1, 0]; scan_type(ASC), INDEX33571964(table1 as A), scan_range[DMTEMPVIEW_22201592.colname,DMTEMPVIEW_22201592.colname]

12 #PRJT2: [33, 1162, 48]; exp_num(2), is_atom(FALSE)

13 #HAGR2: [33, 1162, 48]; grp_num(1), sfun_num(1); slave_empty(0) keys(DZZ.col11)

14 #HASH RIGHT JOIN2: [25, 116278, 48]; key_num(1), ret_null(0), KEY(DZZ.col1=DY.col1)

15 #CSCN2: [1, 12345, 96]; INDEX33571530(table1 as DZZ)

16 #SSCN: [13, 116278, 48]; IDX_DYJBXX_ORGID(table3 as DY)

17 #SPL2: [9, 9876, 740]; key_num(2), spool_num(3), is_atom(FALSE), has_variable(0)

18 #PRJT2: [9, 9876, 740]; exp_num(2), is_atom(FALSE)

19 #HASH RIGHT SEMI JOIN2: [9, 9876, 740]; n_keys(1) KEY(DMTEMPVIEW_22201597.colname=A.col1) KEY_NULL_EQU(0)

20 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

21 #HASH2 INNER JOIN: [9, 9876, 740]; LKEY_UNIQUE KEY_NUM(1); KEY(B.col1=A.col2) KEY_NULL_EQU(0)

22 #CSCN2: [1, 12345, 96]; INDEX33571530(table1 as B)

23 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

24 #SPL2: [5, 11618, 740]; key_num(2), spool_num(2), is_atom(FALSE), has_variable(0)

25 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

26 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201598.colname=A.col1) KEY_NULL_EQU(0)

27 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

28 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col22) KEY_NULL_EQU(0)

29 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb3',min),('zb3',max))

30 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

31 #SPL2: [5, 11618, 740]; key_num(2), spool_num(1), is_atom(FALSE), has_variable(0)

32 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

33 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201599.colname=A.col1) KEY_NULL_EQU(0)

34 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

35 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col21) KEY_NULL_EQU(0)

36 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb2',min),('zb2',max))

37 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

38 #SPL2: [5, 11618, 740]; key_num(2), spool_num(0), is_atom(FALSE), has_variable(0)

39 #PRJT2: [5, 11618, 740]; exp_num(2), is_atom(FALSE)

40 #HASH RIGHT SEMI JOIN2: [5, 11618, 740]; n_keys(1) KEY(DMTEMPVIEW_22201600.colname=A.col1) KEY_NULL_EQU(0)

41 #CONST VALUE LIST: [1, 600, 48]; row_num(600), col_num(1),

42 #HASH2 INNER JOIN: [5, 11618, 740]; KEY_NUM(1); KEY(B.col1=A.col20) KEY_NULL_EQU(0)

43 #SSEK2: [1, 120, 96]; scan_type(ASC), INDEX33572004(table2 as B), scan_range[('zb1',min),('zb1',max))

44 #CSCN2: [2, 12345, 644]; INDEX33571530(table1 as A)

2.3、总结

像这种用 like 做关联很明显是业务涉及不规范，不符合三范式要求。

在业务设计初期，尽量满足好三范式设计，后续才能少点用 like 这种模糊态的查询操作。

业务允许的情况下，尽量使用 = 精确匹配来代替like。

DM SQL关联列 like 优化案例的更多相关文章

SQL 优化案例 1
create or replace procedure SP_GET_NEWEST_CAPTCHA( v_ACCOUNT_ID in VARCHAR2, --接收短信的手机号 v_Tail_num i ...
SQL 优化案例
create or replace procedure SP_GET_NEWEST_CAPTCHA( v_ACCOUNT_ID in VARCHAR2, --接收短信的手机号 v_Tail_num i ...
SQL Server标量函数改写内联表值函数优化案例
问题SQL: SELECT TOP 1001 ha.HuntApplicationID , ha.PartyNumber , mht.Name AS MasterHuntTypeName , htly ...
mysql优化案例
MySQL优化案例 Mysql5.1大表分区效率测试 Mysql5.1大表分区效率测试MySQL | add at 2009-03-27 12:29:31 by PConline | view:60, ...
学习SQL关联查询
通过一个小问题来学习SQL关联查询原话题: 是关于一个left join的,没有技术难度,但不想清楚不一定能回答出正确答案来: TabA表有三个字段Id,Col1,Col2 且里面有一条数据1,1, ...
Hive优化案例
1.Hadoop计算框架的特点数据量大不是问题,数据倾斜是个问题. jobs数比较多的作业效率相对比较低,比如即使有几百万的表,如果多次关联多次汇总,产生十几个jobs,耗时很长.原因是map re ...
SQL业务审核与优化
审核什么是业务审核类似与code review 评审业务Schema和SQL设计偏重关注性能是业务优化的主要入口之一审核提前发现问题,进行优化上 ...
数据库优化案例——————某市中心医院HIS系统
记得在自己学习数据库知识的时候特别喜欢看案例,因为优化的手段是容易掌握的,但是整体的优化思想是很难学会的.这也是为什么自己特别喜欢看案例,今天也开始分享自己做的优化案例. 最近一直很忙,博客产出也少的 ...
SQL Server 列存储索引强化
SQL Server 列存储索引强化 SQL Server 列存储索引强化 1. 概述 2.背景 2.1 索引存储 2.2 缓存和I/O 2.3 Batch处理方式 3 聚集索引 3.1 提高索引创建 ...
mysql如何执行关联查询与优化
mysql如何执行关联查询与优化一.前言在数据库中执行查询(select)在我们工作中是非常常见的,工作中离不开CRUD,在执行查询(select)时,多表关联也非常常见,我们用的也比较多,那么m ...

随机推荐

在vue中使用html2canvas生成图片
首先,在vue中引入html2canvas,执行命令 npm install --save html2canvas 然后在需要生成图片的页面中引入 import html2canvas from 'h ...
F650A光猫的一些命令(一）
查看有 / # uname -a Linux F650A 4.1.25 #12 SMP Tue Aug 15 21:57:30 CST 2017 armv7l GNU/Linux / # cat /p ...
Educational Codeforces Round 102 (Rated for Div
Educational Codeforces Round 102 (Rated for Div. 2) No More Inversions 给定\(k\),序列\(a\)长度为\(n\):\(1,2 ...
.NET Core 异步(Async)底层原理浅谈
简介多线程与异步是两个完全不同的概念,常常有人混淆. 异步异步适用于"IO密集型"的场景,它可以避免因为线程等待IO形成的线程饥饿,从而造成程序吞吐量的降低. 其本质是:让线程 ...
S2P医药营销智能管理平台特点和优势
S2P医药营销智能管理平台是正也科技打造的一个专为医药行业设计的综合性营销解决方案,旨在通过智能化.数据驱动的方式提升医药企业的营销效率和效果.以下是关于S2P医药营销智能管理平台的一些主要特点和优势 ...
Flutter showModalBottomSheet改变高度
showModalBottomSheet改变高度将isScrollControlled设置为true,此时弹窗会全屏展示,再返回一个带高度的SizedBox,就可以指定弹窗的高度了 showModa ...
钉钉机器人发送信息shell
#钉钉机器人发送信息shell 可作为shell函数模块调用,用于监控警报.jenkins发版通知等微信API官方文档 https://ding-doc.dingtalk.com/doc#/serv ...
Spring Validation 校验
概述在 Web 应用中,客户端提交数据之前都会进行数据的校验,比如用户注册时填写的邮箱地址是否符合规范.用户名长度的限制等等,不过这并不意味着服务端的代码可以免去数据验证的工作,用户也可能使用 HT ...
fabric2.0开发部署fabric环境和fabric-samples的启动（2）
通过上一篇文章我们已经将fabric的基本环境搭建成功,接下来我们开始运行使用并初步认识fabric. 创建项目目录 mkdir -p ~/go/src/github.com/hyperledger ...
Qt音视频开发17-vlc内核回调拿图片进行绘制
一.前言在众多播放器中,支持的种类格式众多,并支持DVD影音光盘,VCD影音光盘及各类流式协议,提供了sdk进行开发,这点是至关重要的,尽管很多优秀的播放器很牛逼,由于没有提供sdk第三方开发,少了 ...