Oracle查询优化改写技巧与案例总结一

一、简介

本文将总结一些Oracle查询优化方法与改写技巧，通过一些案例说明用法，本文为观看《Oracle查询优化改写技巧与案例》一书观后的一些总结，大家有空的话可以去阅读一下，下面直接分点总结一些比较有用的优化技巧。

二、优化技巧

【1】. 判空处理使用is null / is not null,不能使用 = null进行判断

查询结果如上图，yxlx这个字段为空，如果使用 = null判断的话，查询结果不正确。

select t.* from zhxg_yx_yxxcgl_cssz t where t.yxlx = null;

必须使用is null / is not null进行判空处理：

select t.* from zhxg_yx_yxxcgl_cssz t where t.yxlx is null;

【2】. 巧用greatest()/least()/coalesce()函数

greatest(): 返回列表中最大的值，列表中的数据类型必须一致.

least(): 返回列表中最小的值，列表中的数据类型必须一致.

coalesce(): 返回列表中第一个非空的值，列表汇中的数据必须同一类型或者能转换成同一类型，效果类似nvl()/case when表达式.

--greatest 返回值列表中最大值
select greatest('A', 'B', 'C') from dual; -- 返回C
select greatest(null, 'B', 'C') from dual; -- 返回null--least 返回值列表中最小值
select least('A', 'B', 'C') from dual; -- 返回A
select least(null, 'B', 'C') from dual; -- 返回null--coalesce 返回该表达式列表的第一个非空value
select coalesce(1, null, 2) from dual; -- 返回1
select coalesce(null, 2, 1) from dual; -- 返回2select coalesce(t.empno, t.mgr) as col from scott.emp t;
--等价于
select nvl(t.empno, t.mgr) as col from scott.emp t;
--等价于
select case when t.empno is not null then t.empno when t.mgr is not null then t.mgr end as col from scott.emp t;

【3】. 查询语句尽量使用别名，易于读懂分析sql语句

select t.zhlx as 账号类型, t.xxjd as 经度, t.xxwd as 纬度from ZHXG_YX_YXXCGL_CSSZ t;--实际项目中，通常都是用简名取别名
select t.zhlx as lx, t.xxjd as jd, t.xxwd as wd from ZHXG_YX_YXXCGL_CSSZ t--在where子句中使用别名,注意要多嵌套一层
select r.*from (select t.zhlx as lx, t.xxjd as jd, t.xxwd as wdfrom ZHXG_YX_YXXCGL_CSSZ t) rwhere r.lx is not null

【4】. select子句中巧用条件判断逻辑

这里主要通过case when子句或者nvl(),nvl2(),decode()函数来实现：

--性别码
select t.xbm as xbmfrom ZHXG_YX_YXXCGL_XSJBXX twhere t.pkid = '07cc1b86-9df9-4b2e-9942-d14c7f238f89';--使用case..when
select t.xbm,casewhen t.xbm = '1' then'男'when t.xbm = '2' then'女'else'保密'end as sexfrom ZHXG_YX_YXXCGL_XSJBXX twhere t.pkid = '07cc1b86-9df9-4b2e-9942-d14c7f238f89';--使用decode
select t.xbm, decode(t.xbm, '1', '男', '2', '女', '保密') as sexfrom ZHXG_YX_YXXCGL_XSJBXX twhere t.pkid = '07cc1b86-9df9-4b2e-9942-d14c7f238f89';

【5】. 限制返回记录的行数

--rownum是一个伪列
select rownum, t.pkid, t.xh, t.lqhfrom zhxg_yx_yxxcgl_xsjbxx twhere rownum <= 10;

获取rownum = 5这条记录，直接使用如下sql是查询不出结果的，因为rownum是依次给记录加上序号的，需要先将记录查询出来才能确定所处的序号是多少。

select rownum, t.pkid, t.xh, t.lqhfrom zhxg_yx_yxxcgl_xsjbxx twhere rownum = 5;

下面进行改写：需要先查询出一些记录，然后进行编号，确定了序号之后才能使用=进行精确查询哪一条。

--获取rownum=5这条记录
select r.rn, r.pkid, r.xh, r.lqhfrom (select rownum rn, t.pkid, t.xh, t.lqhfrom zhxg_yx_yxxcgl_xsjbxx twhere rownum <= 10) rwhere r.rn = 5;

【6】. 从表中随机返回n条记录

这里要注意rownum的不变性，注意先取数据还是先排序，顺序反了，查询结果都会不一样。下面是错误做法：

--错误做法：先取数据后再随机排序(因为rownum已经不变,不管怎么随机,都是同样的数据之间进行排序)
--运行多次发现数据不变，只是顺序变化而已
select t.pkid, t.xh, t.lqh, t.xm, rownum as rm  --rownum排好后不变from zhxg_yx_yxxcgl_xsjbxx twhere rownum <= 3order by dbms_random.value();

运行多几次，我们发现，如果先取数据的话，对比两张图，发现三条记录对应的rownum是不会变化的，所以不管怎么随机排序都是在这三条记录之间进行随机排序。下面是正确做法：

--正确做法： 先排好序之后再取数据
select r.pkid, r.xh, r.xm, r.lqh, r.oldnum, rownum as newnum  --外部查询rownumfrom (select t.pkid, t.xh, t.lqh, t.xm, rownum as oldnum --子查询rownumfrom zhxg_yx_yxxcgl_xsjbxx torder by dbms_random.value()) rwhere rownum <= 3;

【7】. 模糊查询中转义字符的处理

构建结果集：

with temp as(select 'ABCEDF' as vnamefrom dualunion allselect '_BCEFG' as vnamefrom dualunion allselect '_BCEDF' as vnamefrom dualunion allselect '_\BCEDF' as vnamefrom dualunion allselect 'XYCEG' as vnamefrom dual)
SELECT * from temp

【a】查询vname中包含'_BCE'的记录：

with temp as(select 'ABCEDF' as vnamefrom dualunion allselect '_BCEFG' as vnamefrom dualunion allselect '_BCEDF' as vnamefrom dualunion allselect '_\BCEDF' as vnamefrom dualunion allselect 'XYCEG' as vnamefrom dual)
--查询vname中包含'_BCE'的记录
SELECT * from temp t where t.vname like '_BCE%'

上图中圈起来的记录其实并不符合条件，我们需要对‘_’进行转义处理：

with temp as(select 'ABCEDF' as vnamefrom dualunion allselect '_BCEFG' as vnamefrom dualunion allselect '_BCEDF' as vnamefrom dualunion allselect '_\BCEDF' as vnamefrom dualunion allselect 'XYCEG' as vnamefrom dual)
--查询vname中包含'_BCE'的记录
--SELECT * from temp t where t.vname like '_BCE%'
--转义处理
SELECT * from temp t where t.vname like '\_BCE%' escape '\'

【b】获取'_\BCEDF'这条记录

with temp as(select 'ABCEDF' as vnamefrom dualunion allselect '_BCEFG' as vnamefrom dualunion allselect '_BCEDF' as vnamefrom dualunion allselect '_\BCEDF' as vnamefrom dualunion allselect 'XYCEG' as vnamefrom dual)
--获取'_\BCEDF'这条记录
SELECT * from temp t where t.vname like '_\BCE%' escape '\';

转义之后，报错，如上图，这时候还需要使用 ‘\\’ 进行转义：

with temp as(select 'ABCEDF' as vnamefrom dualunion allselect '_BCEFG' as vnamefrom dualunion allselect '_BCEDF' as vnamefrom dualunion allselect '_\BCEDF' as vnamefrom dualunion allselect 'XYCEG' as vnamefrom dual)
--获取'_\BCEDF'这条记录
SELECT * from temp t where t.vname like '_\\BCE%' escape '\';

【8】. 巧用translate()函数：

针对字符进行替换,返回将出现在from中的每个字符替换为to中的相应字符以后的字符串;若from比to字符串长，那么在from中比to中多出的字符将会被删除; 三个参数中有一个是空，返回值也将是空值

--translate(char, from, to)替换函数
select translate('abc张三defabb', 'abcdef', '123456') as temp from dual;
--解析： a - 1   b - 2  c - 3  d - 4  e - 5  f - 6 依次替换字符

--三个参数中,只有有其中一个参数为null,结果就返回null
select translate('abc张三defabb', null, '123456') as temp from dual;

--若from比to字符串长，那么在from中比to中多出的字符将会被删除
select translate('abc张三defabb', 'abcdef', '123') as temp from dual;

【9】. 巧处理排序空值

使用nulls first/ nulls last进行空值排序处理：

--空值排前面
select t.pkid, t.xh, t.lqh, t.xm, t.mmfrom ZHXG_YX_YXXCGL_XSJBXX twhere rownum <= 5order by t.mm nulls first;--空值排后面
select t.pkid, t.xh, t.lqh, t.xm, t.mmfrom ZHXG_YX_YXXCGL_XSJBXX twhere rownum <= 5order by t.mm nulls last;

【10】. 根据条件取不同的列进行排序

针对这种需求，我们可以在查询语句中多查询出一列，然后再根据这列进行排序。

--按分数进行排序,并且70-75分数段要先排序
select t.pkid, t.gkfs, t.xh, t.xm, t.lqhfrom ZHXG_YX_YXXCGL_XSJBXX twhere rownum <= 10order by t.gkfs;--第一种方式
select *from (select t.pkid,t.gkfs,t.xh,t.xm,t.lqh,casewhen t.gkfs >= 70 and t.gkfs <= 75 then1else2end as pxhfrom ZHXG_YX_YXXCGL_XSJBXX twhere rownum <= 10) rorder by r.pxh, r.gkfs

当然，也可以直接在order by子句中判断：

--第二种方式
select t.pkid, t.gkfs, t.xh, t.xm, t.lqhfrom ZHXG_YX_YXXCGL_XSJBXX twhere rownum <= 10order by casewhen t.gkfs >= 70 and t.gkfs <= 75 then1else2end,t.gkfs

【11】. union all 与空字符串

在oracle中空字符串' '一般相当于null,但并不说明null 与空字符串' '是等价的，空字符串' '本质上还是varchar2()类型的，null可以被作为任何数据类型。看以下示例：

--''本质上还是varchar2()类型, null可以是任何类型
select 'a' from dual union all select '' from dual;
select 'a' from dual union all select null from dual;

但是下面的代码就会报错了，如果union all两边的数据类型没有对应上，就会报错:

--sysdate 是 时间类型  ''是varchar2类型, 报错
select sysdate from dual union all select '' from dual;

注意和下面代码的区别: null可以充当任何数据类型，所以不会报错。

--sysdate 是 时间类型  null可以是任何类型, 不会报错
select sysdate from dual union all select null from dual;

【12】. 使用union 代替 or可以提高效率

select *from EMP twhere t.empno = '7788'or t.ename = 'SCOTT';--会存在重复数据，必须去重
select *from EMP twhere t.empno = '7788'
union all
select *from EMP twhere t.ename = 'SCOTT';--使用到索引，效率相对较高
select *from EMP twhere t.empno = '7788'
union
select *from EMP twhere t.ename = 'SCOTT';

需要注意的是: 有重复数据的数据集使用union后得到的结果可能会不一致。看以下示例：

两个结果集有四条记录是重复的，使用or连接后查询得到五条数据:

select t.deptnofrom EMP twhere t.job = 'SALESMAN'or t.mgr = 7698;

接着，我们使用union来改写以下：

--去重之后,结果不正确
select t.deptnofrom EMP twhere t.job = 'SALESMAN'
union
select t.deptnofrom EMP twhere t.mgr = 7698;

如上图，union之后，结果集变成一条记录了，明显不正确。接着改写：使用唯一列标识/rowid/rownum来进行区别，可以避免union不必要的去重。

--解决方法一： 使用唯一列
select t.empno, t.deptnofrom EMP twhere t.job = 'SALESMAN'
union
select t.empno, t.deptnofrom EMP twhere t.mgr = 7698;--解决方法二：使用rowid(或者对查询一列rownum)
select rowid, t.deptnofrom EMP twhere t.job = 'SALESMAN'
union
select rowid, t.deptnofrom EMP twhere t.mgr = 7698;

【13】内连接尽量使用inner join，避免使用from 表1，表2这种方式

--推荐使用 inner join ,容易看清楚两张表之间的关联关系
select t.empno,t.ename,t.job,t.mgr,t.hiredate,t.sal,t.comm,t.deptno,d.dnamefrom emp tinner join dept don d.deptno = t.deptnowhere t.deptno = 10;--不推荐使用
select e.empno,e.ename,e.job,e.mgr,e.hiredate,e.sal,e.comm,e.deptno,d2.dnamefrom emp e, dept d2where e.deptno = 10and e.deptno = d2.deptno;

【14】 inner join / left join / right join / full join区别

inner join: 只返回左右表相匹配的记录。
left join: 左表返回全部数据，右表如果没有匹配的记录用null填充。
right join: 右表返回全部数据，左表如果没有匹配的记录用null填充。
full join: 左右两张表的数据都返回，只有相匹配的才显示在同一行。

看以下示例：

【inner join】：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
select L.str, L.val, R.strfrom Linner join Ron L.val = R. valorder by L.val;

另一种写法：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
select L.str, L.val, R.str from L, R where L.val = R. val order by L.val;

【left join】：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
select L.str, L.val, R.strfrom Lleft join Ron L.val = R. valorder by L.val;

【right join】：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
select L.str, L.val, R.strfrom Lright join Ron L.val = R. valorder by L.val;

【full join】：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
select L.str, L.val, R.strfrom Lfull join Ron L.val = R. valorder by L.val;

【15】. 自关联

自关联表示的是join表自身，只是连接条件的字段可能不一致。

--查询每个员工对应的主管
select t.empno, t.ename, t2.ename as mgrnamefrom emp tleft join emp t2on t.mgr = t2.empno;

【16】 not in / not exists / left join

使用这三个实现统计没有员工的部门信息：

--not in
select *from dept dwhere d.deptno not in--统计所有有员工的部门编号(select e.deptno from emp e where e.deptno is not null);--not exists
select *from dept d--不存在 e.deptno = d.deptno的部门信息where not exists (select * from emp e where e.deptno = d.deptno);--left join
select d.*from dept dleft join emp eon e.deptno = d.deptnowhere e.deptno is null

以上三条sql执行结果一致。

至于哪一种效率高一点，可以通过查看分析sql执行计划看出来：

explain plan for select *from dept d--不存在 e.deptno = d.deptno的部门信息where not exists (select * from emp e where e.deptno = d.deptno);select * from table(dbms_xplan.display());

【17】外连接条件需要注意放的位置

需要特别注意所加上的条件是要过滤外连接join之后的结果还是先过滤之后再进行外连接join操作。看以下示例：

假设我外连接只需要连接右表val= '3'这条记录，那么如果加条件的位置不对，结果截然不同。如下：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
--假设右表只需要查找val = 3这条记录--错误写法
select L.str, L.val, R.strfrom Lleft join Ron L.val = R. val
--加where等于对join之后的整个结果进行过滤where R.val = '3'order by L.val;

如上图，查询结果只有一条，并不是像预期那样：

下面提供两种方法，解决这种外连接加条件错误：

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
--假设右表只需要查找val = 3这条记录
--正确写法：
select L.str, L.val, R.strfrom Lleft join Ron L.val = R. valand R.val = '3'order by L.val;

with L as(select 'left_1' as str, '1' as valfrom dualunion allselect 'left_2' as str, '2' as valfrom dualunion allselect 'left_3' as str, '3' as valfrom dualunion allselect 'left_4' as str, '4' as valfrom dual),
R as(select 'right_3' as str, '3' as valfrom dualunion allselect 'right_4' as str, '4' as valfrom dualunion allselect 'right_5' as str, '5' as valfrom dualunion allselect 'right_6' as str, '6' as valfrom dual)
--假设右表只需要查找val = 3这条记录
--正确写法二：
select L.str, L.val, R.strfrom L
--先过滤val='3'之后再joinleft join (select * from R where R.val = '3') Ron L.val = R. valorder by L.val;

以上两种方式才是正确的。

【18】聚集与内连接

首先构造测试数据：

with emp_bonus as(select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select * from emp_bonus;

计算员工对应的奖金，以上是员工对应的奖金等级。

先连接：


with emp_bonus as(select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select e.deptno,e.sal,(e.sal * casewhen b.type = 1 then0.1when b.type = 2 then0.2when b.type = 3 then0.3end) as bonusfrom emp einner join emp_bonus bon b.empno = e.empnowhere e.deptno = 10

然后进行统计:

with emp_bonus as(select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select e.deptno,sum(e.sal),sum(e.sal * casewhen b.type = 1 then0.1when b.type = 2 then0.2when b.type = 3 then0.3end) as bonusfrom emp einner join emp_bonus bon b.empno = e.empnowhere e.deptno = 10group by e.deptno;--select sum(e.sal) as total from emp e where e.deptno = 10;

查看上图，发现总工资金额为10050，执行统计总金额sql:

select sum(e.sal) as total from emp e where e.deptno = 10;

我们发现总工资金额对不上，原因在于奖金等级那张表有个员工7934的有两条记录，正确的做法应该是先聚合员工的等级信息再进行连接join操作。

【1】先统计每个员工的奖金等级

with emp_bonus as(select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select b.empno,sum(casewhen b.type = 1 then0.1when b.type = 2 then0.2when b.type = 3 then0.3end) as ratefrom emp_bonus bgroup by b.empno;

【2】再进行聚合

select sum(e.sal) as total, sum(e.sal * r.rate) as total_bonus, e.deptnofrom emp einner join(with emp_bonus as (select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select b.empno,sum(casewhen b.type = 1 then0.1when b.type = 2 then0.2when b.type = 3 then0.3end) as ratefrom emp_bonus bgroup by b.empno) ron r.empno = e.empnowhere e.deptno = 10group by e.deptno

这样结果就正确了。

【19】聚集与外连接

如果需要返回各个部门的总工资以及总奖金，根据上面的示例，直接使用left join即可。

select sum(e.sal) as total, sum(e.sal * r.rate) as total_bonus, e.deptnofrom emp eleft join(with emp_bonus as (select 7934 as empno, 1 as typefrom dualunion allselect 7934 as empno, 2 as typefrom dualunion allselect 7839 as empno, 3 as typefrom dualunion allselect 7782 as empno, 1 as typefrom dual)
select b.empno,sum(casewhen b.type = 1 then0.1when b.type = 2 then0.2when b.type = 3 then0.3end) as ratefrom emp_bonus bgroup by b.empno) ron r.empno = e.empnogroup by e.deptno

【20】多表查询注意空值问题

示例：返回所有比‘ALLEN’提成低的员工

select e.comm, e.empno, e.enamefrom emp ewhere e.comm < (select e2.comm from emp e2 where e2.ename = 'ALLEN');

对比所有员工的提成结果

我们发现，上面统计结果明显不对，其中还有一部分员工没有提成，也算比‘ALLEN’提成低的范围，这些记录应该也要被查询出来：我们将提成为空的做一下转换，再进行比较。

select e.comm, e.empno, e.enamefrom emp ewhere nvl(e.comm, 0) <(select e2.comm from emp e2 where e2.ename = 'ALLEN');

这样结果就正确了。

另外，一些函数如in(null)会直接返回null，需要注意一下。

三、总结

这只是第一部分总结，下一篇继续总结。积少成多，keeping....