oracle分析函数--SQL*PLUS环境
一、总体介绍
12.1 分析函数如何工作
语法 FUNCTION_NAME(<参数>,…) OVER (<PARTITION BY 表达式,…> <ORDER BY 表达式 <ASC DESC> <NULLS FIRST NULLS LAST>> <WINDOWING子句>) PARTITION子句 ORDER BY子句 WINDOWING子句 缺省时相当于RANGE UNBOUNDED PRECEDING
1. 值域窗(RANGE WINDOW)
RANGE N PRECEDING 仅对数值或日期类型有效,选定窗为排序后当前行之前,某列(即排序列)值大于/小于(当前行该列值 –/+ N)的所有行,因此与ORDER BY子句有关系。
2. 行窗(ROW WINDOW)
ROWS N PRECEDING 选定窗为当前行及之前N行。
还可以加上BETWEEN AND 形式,例如RANGE BETWEEN m PRECEDING AND n FOLLOWING
函数 AVG(<distinct all> eXPr)
一组或选定窗中表达式的平均值 CORR(expr, expr) 即COVAR_POP(exp1,exp2) / (STDDEV_POP(expr1) * STDDEV_POP(expr2)),两个表达式的互相关,-1(反相关) ~ 1(正相关),0表示不相关
COUNT(<distinct> <*> <expr>) 计数
COVAR_POP(expr, expr) 总体协方差
COVAR_SAMP(expr, expr) 样本协方差
CUME_DIST 累积分布,即行在组中的相对位置,返回0 ~ 1
DENSE_RANK 行的相对排序(与ORDER BY搭配),相同的值具有一样的序数(NULL计为相同),并不留空序数
FIRST_VALUE 一个组的第一个值
LAG(expr, <offset>, <default>) 访问之前的行,OFFSET是缺省为1 的正数,表示相对行数,DEFAULT是当超出选定窗范围时的返回值(如第一行不存在之前行)
LAST_VALUE 一个组的最后一个值
LEAD(expr, <offset>, <default>) 访问之后的行,OFFSET是缺省为1 的正数,表示相对行数,DEFAULT是当超出选定窗范围时的返回值(如最后行不存在之前行)
MAX(expr) 最大值
MIN(expr) 最小值
NTILE(expr) 按表达式的值和行在组中的位置编号,如表达式为4,则组分4份,分别为1 ~ 4的值,而不能等分则多出的部分在值最小的那组
PERCENT_RANK 类似CUME_DIST,1/(行的序数 - 1)
RANK 相对序数,答应并列,并空出随后序号
RATIO_TO_REPORT(expr) 表达式值 / SUM(表达式值)
ROW_NUMBER 排序的组中行的偏移
STDDEV(expr) 标准差
STDDEV_POP(expr) 总体标准差
STDDEV_SAMP(expr) 样本标准差
SUM(expr) 合计
VAR_POP(expr) 总体方差
VAR_SAMP(expr) 样本方差
VARIANCE(expr) 方差
REGR_ xxxx(expr, expr) 线性回归函数
REGR_SLOPE:返回斜率,等于COVAR_POP(expr1, expr2) / VAR_POP(expr2)
REGR_INTERCEPT:返回回归线的y截距,等于
AVG(expr1) - REGR_SLOPE(expr1, expr2) * AVG(expr2)
REGR_COUNT:返回用于填充回归线的非空数字对的数目
REGR_R2:返回回归线的决定系数,计算式为:
If VAR_POP(expr2) = 0 then return NULL
If VAR_POP(expr1) = 0 and VAR_POP(expr2) != 0 then return 1
If VAR_POP(expr1) > 0 and VAR_POP(expr2 != 0 then
return POWER(CORR(expr1,expr),2)
REGR_AVGX:计算回归线的自变量(expr2)的平均值,去掉了空对(expr1, expr2)后,等于AVG(expr2)
REGR_AVGY:计算回归线的应变量(expr1)的平均值,去掉了空对(expr1, expr2)后,等于AVG(expr1)
REGR_SXX: 返回值等于REGR_COUNT(expr1, expr2) * VAR_POP(expr2)
REGR_SYY: 返回值等于REGR_COUNT(expr1, expr2) * VAR_POP(expr1)
REGR_SXY: 返回值等于REGR_COUNT(expr1, expr2) * COVAR_POP(expr1, expr2)
首先:创建表及接入测试数据
create table students
(id number(15,0),
area varchar2(10),
stu_type varchar2(2),
score number(20,2));
insert into students values(1, '111', 'g', 80 );
insert into students values(1, '111', 'j', 80 );
insert into students values(1, '222', 'g', 89 );
insert into students values(1, '222', 'g', 68 );
insert into students values(2, '111', 'g', 80 );
insert into students values(2, '111', 'j', 70 );
insert into students values(2, '222', 'g', 60 );
insert into students values(2, '222', 'j', 65 );
insert into students values(3, '111', 'g', 75 );
insert into students values(3, '111', 'j', 58 );
insert into students values(3, '222', 'g', 58 );
insert into students values(3, '222', 'j', 90 );
insert into students values(4, '111', 'g', 89 );
insert into students values(4, '111', 'j', 90 );
insert into students values(4, '222', 'g', 90 );
insert into students values(4, '222', 'j', 89 );
commit;
二、具体应用:
1、分组求和:
1)GROUP BY子句
--A、GROUPING SETS
select id,area,stu_type,sum(score) score
from students
group by grouping sets((id,area,stu_type),(id,area),id)
order by id,area,stu_type;
/*--------理解grouping sets
select a, b, c, sum( d ) from t
group by grouping sets ( a, b, c )
等效于
select * from (
select a, null, null, sum( d ) from t group by a
union all
select null, b, null, sum( d ) from t group by b
union all
select null, null, c, sum( d ) from t group by c
)
*/
--B、ROLLUP
select id,area,stu_type,sum(score) score
from students
group by rollup(id,area,stu_type)
order by id,area,stu_type;
/*--------理解rollup
select a, b, c, sum( d )
from t
group by rollup(a, b, c);
等效于
select * from (
select a, b, c, sum( d ) from t group by a, b, c
union all
select a, b, null, sum( d ) from t group by a, b
union all
select a, null, null, sum( d ) from t group by a
union all
select null, null, null, sum( d ) from t
)
*/
--C、CUBE
select id,area,stu_type,sum(score) score
from students
group by cube(id,area,stu_type)
order by id,area,stu_type;
/*--------理解cube
select a, b, c, sum( d ) from t
group by cube( a, b, c)
等效于
select a, b, c, sum( d ) from t
group by grouping sets(
( a, b, c ),
( a, b ), ( a ), ( b, c ),
( b ), ( a, c ), ( c ),
() )
*/
--D、GROUPING
/*从上面的结果中我们很容易发现,每个统计数据所对应的行都会出现null,
如何来区分到底是根据那个字段做的汇总呢,grouping函数判断是否合计列!*/
select decode(grouping(id),1,'all id',id) id,
decode(grouping(area),1,'all area',to_char(area)) area,
decode(grouping(stu_type),1,'all_stu_type',stu_type) stu_type,
sum(score) score
from students
group by cube(id,area,stu_type)
order by id,area,stu_type;
二、OVER()函数的使用
1、统计名次——DENSE_RANK(),ROW_NUMBER()
1)允许并列名次、名次不间断,DENSE_RANK(),结果如122344456……
将score按ID分组排名:dense_rank() over(partition by id order by score desc)
将score不分组排名:dense_rank() over(order by score desc)
select id,area,score,
dense_rank() over(partition by id order by score desc) 分组id排序,
dense_rank() over(order by score desc) 不分组排序
from students order by id,area;
2)不允许并列名次、相同值名次不重复,ROW_NUMBER(),结果如123456……
将score按ID分组排名:row_number() over(partition by id order by score desc)
将score不分组排名:row_number() over(order by score desc)
select id,area,score,
row_number() over(partition by id order by score desc) 分组id排序,
row_number() over(order by score desc) 不分组排序
from students order by id,area;
3)允许并列名次、复制名次自动空缺,rank(),结果如12245558……
将score按ID分组排名:rank() over(partition by id order by score desc)
将score不分组排名:rank() over(order by score desc)
select id,area,score,
rank() over(partition by id order by score desc) 分组id排序,
rank() over(order by score desc) 不分组排序
from students order by id,area;
4)名次分析,cume_dist()——-最大排名/总个数
函数:cume_dist() over(order by id)
select id,area,score,
cume_dist() over(order by id) a, --按ID最大排名/总个数
cume_dist() over(partition by id order by score desc) b, --ID分组中,scroe最大排名值/本组总个数
row_number() over (order by id) 记录号
from students order by id,area;
5)利用cume_dist(),允许并列名次、复制名次自动空缺,取并列后较大名次,结果如22355778……
将score按ID分组排名:cume_dist() over(partition by id order by score desc)*sum(1) over(partition by id)
将score不分组排名:cume_dist() over(order by score desc)*sum(1) over()
select id,area,score,
sum(1) over() as 总数,
sum(1) over(partition by id) as 分组个数,
(cume_dist() over(partition by id order by score desc))*(sum(1) over(partition by id)) 分组id排序,
(cume_dist() over(order by score desc))*(sum(1) over()) 不分组排序
from students order by id,area
2、分组统计--sum(),max(),avg(),RATIO_TO_REPORT()
select id,area,
sum(1) over() as 总记录数,
sum(1) over(partition by id) as 分组记录数,
sum(score) over() as 总计 ,
sum(score) over(partition by id) as 分组求和,
sum(score) over(order by id) as 分组连续求和,
sum(score) over(partition by id,area) as 分组ID和area求和,
sum(score) over(partition by id order by area) as 分组ID并连续按area求和,
max(score) over() as 最大值,
max(score) over(partition by id) as 分组最大值,
max(score) over(order by id) as 分组连续最大值,
max(score) over(partition by id,area) as 分组ID和area求最大值,
max(score) over(partition by id order by area) as 分组ID并连续按area求最大值,
avg(score) over() as 所有平均,
avg(score) over(partition by id) as 分组平均,
avg(score) over(order by id) as 分组连续平均,
avg(score) over(partition by id,area) as 分组ID和area平均,
avg(score) over(partition by id order by area) as 分组ID并连续按area平均,
RATIO_TO_REPORT(score) over() as "占所有%",
RATIO_TO_REPORT(score) over(partition by id) as "占分组%",
score from students;
3、LAG(COL,n,default)、LEAD(OL,n,default) --取前后边N条数据
取前面记录的值:lag(score,n,x) over(order by id)
取后面记录的值:lead(score,n,x) over(order by id)
参数:n表示移动N条记录,X表示不存在时填充值,iD表示排序列
select id,lag(score,1,0) over(order by id) lg,score from students;
select id,lead(score,1,0) over(order by id) lg,score from students;
4、FIRST_VALUE()、LAST_VALUE()
取第起始1行值:first_value(score,n) over(order by id)
取第最后1行值:LAST_value(score,n) over(order by id)
select id,first_value(score) over(order by id) fv,score from students;
select id,last_value(score) over(order by id) fv,score from students;
sum(...) over ...
【功能】连续求和分析函数
【参数】具体参示例
【说明】Oracle分析函数
NC示例:
select bdcode,sum(1) over(order by bdcode) aa from bd_bdinfo
【示例】
1.原表信息: SQL> break on deptno skip 1 -- 为效果更明显,把不同部门的数据隔段显示。
SQL> select deptno,ename,sal
2 from emp
3 order by deptno;
DEPTNO ENAME SAL
---------- ---------- ----------
10 CLARK 2450
KING 5000
MILLER 1300
20 SMITH 800
ADAMS 1100
FORD 3000
SCOTT 3000
JONES 2975
30 ALLEN 1600
BLAKE 2850
MARTIN 1250
JAMES 950
TURNER 1500
WARD 1250
2.先来一个简单的,注意over(...)条件的不同,
使用 sum(sal) over (order by ename)... 查询员工的薪水“连续”求和,
注意over (order by ename)如果没有order by 子句,求和就不是“连续”的,
放在一起,体会一下不同之处:
SQL> select deptno,ename,sal,
2 sum(sal) over (order by ename) 连续求和,
3 sum(sal) over () 总和, -- 此处sum(sal) over () 等同于sum(sal)
4 100*round(sal/sum(sal) over (),4) "份额(%)"
5 from emp
6 /
DEPTNO ENAME SAL 连续求和 总和 份额(%)
---------- ---------- ---------- ---------- ---------- ----------
20 ADAMS 1100 1100 29025 3.79
30 ALLEN 1600 2700 29025 5.51
30 BLAKE 2850 5550 29025 9.82
10 CLARK 2450 8000 29025 8.44
20 FORD 3000 11000 29025 10.34
30 JAMES 950 11950 29025 3.27
20 JONES 2975 14925 29025 10.25
10 KING 5000 19925 29025 17.23
30 MARTIN 1250 21175 29025 4.31
10 MILLER 1300 22475 29025 4.48
20 SCOTT 3000 25475 29025 10.34
20 SMITH 800 26275 29025 2.76
30 TURNER 1500 27775 29025 5.17
30 WARD 1250 29025 29025 4.31
3.使用子分区查出各部门薪水连续的总和。注意按部门分区。注意over(...)条件的不同,
sum(sal) over (partition by deptno order by ename) 按部门“连续”求总和
sum(sal) over (partition by deptno) 按部门求总和
sum(sal) over (order by deptno,ename) 不按部门“连续”求总和
sum(sal) over () 不按部门,求所有员工总和,效果等同于sum(sal)。
SQL> select deptno,ename,sal,
2 sum(sal) over (partition by deptno order by ename) 部门连续求和,--各部门的薪水"连续"求和
3 sum(sal) over (partition by deptno) 部门总和, -- 部门统计的总和,同一部门总和不变
4 100*round(sal/sum(sal) over (partition by deptno),4) "部门份额(%)",
5 sum(sal) over (order by deptno,ename) 连续求和, --所有部门的薪水"连续"求和
6 sum(sal) over () 总和, -- 此处sum(sal) over () 等同于sum(sal),所有员工的薪水总和
7 100*round(sal/sum(sal) over (),4) "总份额(%)"
8 from emp
9 /
DEPTNO ENAME SAL 部门连续求和 部门总和 部门份额(%) 连续求和 总和 总份额(%)
------ ------ ----- ------------ ---------- ----------- ---------- ------ ----------
10 CLARK 2450 2450 8750 28 2450 29025 8.44
KING 5000 7450 8750 57.14 7450 29025 17.23
MILLER 1300 8750 8750 14.86 8750 29025 4.48
20 ADAMS 1100 1100 10875 10.11 9850 29025 3.79
FORD 3000 4100 10875 27.59 12850 29025 10.34
JONES 2975 7075 10875 27.36 15825 29025 10.25
SCOTT 3000 10075 10875 27.59 18825 29025 10.34
SMITH 800 10875 10875 7.36 19625 29025 2.76
30 ALLEN 1600 1600 9400 17.02 21225 29025 5.51
BLAKE 2850 4450 9400 30.32 24075 29025 9.82
JAMES 950 5400 9400 10.11 25025 29025 3.27
MARTIN 1250 6650 9400 13.3 26275 29025 4.31
TURNER 1500 8150 9400 15.96 27775 29025 5.17
WARD 1250 9400 9400 13.3 29025 29025 4.31
4.来一个综合的例子,求和规则有按部门分区的,有不分区的例子
SQL> select deptno,ename,sal,sum(sal) over (partition by deptno order by sal) dept_sum,
2 sum(sal) over (order by deptno,sal) sum
3 from emp;
DEPTNO ENAME SAL DEPT_SUM SUM
---------- ---------- ---------- ---------- ----------
10 MILLER 1300 1300 1300
CLARK 2450 3750 3750
KING 5000 8750 8750
20 SMITH 800 800 9550
ADAMS 1100 1900 10650
JONES 2975 4875 13625
SCOTT 3000 10875 19625
FORD 3000 10875 19625
30 JAMES 950 950 20575
WARD 1250 3450 23075
MARTIN 1250 3450 23075
TURNER 1500 4950 24575
ALLEN 1600 6550 26175
BLAKE 2850 9400 29025
5.来一个逆序的,即部门从大到小排列,部门里各员工的薪水从高到低排列,累计和的规则不变。
SQL> select deptno,ename,sal,
2 sum(sal) over (partition by deptno order by deptno desc,sal desc) dept_sum,
3 sum(sal) over (order by deptno desc,sal desc) sum
4 from emp;
DEPTNO ENAME SAL DEPT_SUM SUM
---------- ---------- ---------- ---------- ----------
30 BLAKE 2850 2850 2850
ALLEN 1600 4450 4450
TURNER 1500 5950 5950
WARD 1250 8450 8450
MARTIN 1250 8450 8450
JAMES 950 9400 9400
20 SCOTT 3000 6000 15400
FORD 3000 6000 15400
JONES 2975 8975 18375
ADAMS 1100 10075 19475
SMITH 800 10875 20275
10 KING 5000 5000 25275
CLARK 2450 7450 27725
MILLER 1300 8750 29025
6.体会:在"... from emp;"后面不要加order by 子句,使用的分析函数的(partition by deptno order by sal)
里已经有排序的语句了,如果再在句尾添加排序子句,一致倒罢了,不一致,结果就令人费劲了。如:
SQL> select deptno,ename,sal,sum(sal) over (partition by deptno order by sal) dept_sum,
2 sum(sal) over (order by deptno,sal) sum
3 from emp
4 order by deptno desc;
DEPTNO ENAME SAL DEPT_SUM SUM
---------- ---------- ---------- ---------- ----------
30 JAMES 950 950 20575
WARD 1250 3450 23075
MARTIN 1250 3450 23075
TURNER 1500 4950 24575
ALLEN 1600 6550 26175
BLAKE 2850 9400 29025
20 SMITH 800 800 9550
ADAMS 1100 1900 10650
JONES 2975 4875 13625
SCOTT 3000 10875 19625
FORD 3000 10875 19625
10 MILLER 1300 1300 1300
CLARK 2450 3750 3750
KING 5000 8750 8750
RANK()
dense_rank()
【语法】RANK ( ) OVER ( [query_partition_clause] order_by_clause )
dense_RANK ( ) OVER ( [query_partition_clause] order_by_clause )
【功能】聚合函数RANK 和 dense_rank 主要的功能是计算一组数值中的排序值。
【参数】dense_rank与rank()用法相当,
【区别】dence_rank在并列关系是,相关等级不会跳过。rank则跳过
rank()是跳跃排序,有两个第二名时接下来就是第四名(同样是在各个分组内)
dense_rank()l是连续排序,有两个第二名时仍然跟着第三名。
【说明】Oracle分析函数
【示例】
聚合函数RANK 和 dense_rank 主要的功能是计算一组数值中的排序值。
在9i版本之前,只有分析功能(analytic ),即从一个查询结果中计算每一行的排序值,是基于order_by_clause子句中的value_exprs指定字段的。
其语法为:
RANK ( ) OVER ( [query_partition_clause] order_by_clause )
在9i版本新增加了合计功能(aggregate),即对给定的参数值在设定的排序查询中计算出其排序值。这些参数必须是常数或常值表达式,且必须和ORDER BY子句中的字段个数、位置、类型完全一致。
其语法为:
RANK ( expr [, expr]... ) WITHIN GROUP
( ORDER BY
expr [ DESC | ASC ] [NULLS { FIRST | LAST }]
[, expr [ DESC | ASC ] [NULLS { FIRST | LAST }]]...
)
例子1:
有表Table内容如下
COL1 COL2
1 1
2 1
3 2
3 1
4 1
4 2
5 2
5 2
6 2
分析功能:列出Col2分组后根据Col1排序,并生成数字列。比较实用于在成绩表中查出各科前几名的信息。
SELECT a.*,RANK() OVER(PARTITION BY col2 ORDER BY col1) "Rank" FROM table a;
结果如下:
COL1 COL2 Rank
1 1 1
2 1 2
3 1 3
4 1 4
3 2 1
4 2 2
5 2 3
5 2 3
6 2 5
例子2:
TABLE:A (科目,分数)
数学,80
语文,70
数学,90
数学,60
数学,100
语文,88
语文,65
语文,77
现在我想要的结果是:(即想要每门科目的前3名的分数)
数学,100
数学,90
数学,80
语文,88
语文,77
语文,70
那么语句就这么写:
select * from (select rank() over(partition by 科目 order by 分数 desc) rk,a.* from a) t
where t.rk<=3;
例子3:
合计功能:计算出数值(4,1)在Orade By Col1,Col2排序下的排序值,也就是col1=4,col2=1在排序以后的位置
SELECT RANK(4,3) WITHIN GROUP (ORDER BY col1,col2) "Rank" FROM table;
结果如下:
Rank
4
dense_rank与rank()用法相当,但是有一个区别:dence_rank在并列关系是,相关等级不会跳过。rank则跳过
例如:表
A B C
a liu wang
a jin shu
a cai kai
b yang du
b lin ying
b yao cai
b yang 99
例如:当rank时为:
select m.a,m.b,m.c,rank() over(partition by a order by b) liu from test3 m
A B C LIU
a cai kai 1
a jin shu 2
a liu wang 3
b lin ying 1
b yang du 2
b yang 99 2
b yao cai 4
而如果用dense_rank时为:
select m.a,m.b,m.c,dense_rank() over(partition by a order by b) liu from test3 m
A B C LIU
a cai kai 1
a jin shu 2
a liu wang 3
b lin ying 1
b yang du 2
b yang 99 2
b yao cai 3
ROW_NUMBER()
【语法】ROW_NUMBER() OVER (PARTITION BY COL1 ORDER BY COL2)
【功能】表示根据COL1分组,在分组内部根据 COL2排序,而这个值就表示每组内部排序后的顺序编号(组内连续的唯一的)
row_number() 返回的主要是“行”的信息,并没有排名
【参数】
【说明】Oracle分析函数
主要功能:用于取前几名,或者最后几名等
【示例】
表内容如下:
name | seqno | description
A | 1 | test
A | 2 | test
A | 3 | test
A | 4 | test
B | 1 | test
B | 2 | test
B | 3 | test
B | 4 | test
C | 1 | test
C | 2 | test
C | 3 | test
C | 4 | test
我想有一个sql语句,搜索的结果是
A | 1 | test
A | 2 | test
B | 1 | test
B | 2 | test
C | 1 | test
C | 2 | test
实现:
select name,seqno,description
from(select name,seqno,description,row_number() over (partition by name order by seqno) id
from table_name) where id<=3;
lag()和lead()
【语法】
lag(EXPR,<OFFSET>,<DEFAULT>)
LEAD(EXPR,<OFFSET>,<DEFAULT>)
【功能】表示根据COL1分组,在分组内部根据 COL2排序,而这个值就表示每组内部排序后的顺序编号(组内连续的唯一的)
lead () 下一个值 lag() 上一个值
【参数】
EXPR是从其他行返回的表达式
OFFSET是缺省为1 的正数,表示相对行数。希望检索的当前行分区的偏移量
DEFAULT是在OFFSET表示的数目超出了分组的范围时返回的值。
【说明】Oracle分析函数
【示例】
-- Create table
create table LEAD_TABLE
(
CASEID VARCHAR2(10),
STEPID VARCHAR2(10),
ACTIONDATE DATE
)
tablespace COLM_DATA
pctfree 10
initrans 1
maxtrans 255
storage
(
initial 64K
minextents 1
maxextents unlimited
);
insert into LEAD_TABLE values('Case1','Step1',to_date('20070101','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step2',to_date('20070102','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step3',to_date('20070103','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step4',to_date('20070104','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step5',to_date('20070105','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step4',to_date('20070106','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step6',to_date('20070101','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case1','Step1',to_date('20070201','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case2','Step2',to_date('20070202','yyyy-mm-dd'));
insert into LEAD_TABLE values('Case2','Step3',to_date('20070203','yyyy-mm-dd'));
commit;
结果如下:
Case1 Step1 2007-1-1 Step2 2007-1-2
Case1 Step2 2007-1-2 Step3 2007-1-3 Step1 2007-1-1
Case1 Step3 2007-1-3 Step4 2007-1-4 Step2 2007-1-2
Case1 Step4 2007-1-4 Step5 2007-1-5 Step3 2007-1-3
Case1 Step5 2007-1-5 Step4 2007-1-6 Step4 2007-1-4
Case1 Step4 2007-1-6 Step6 2007-1-7 Step5 2007-1-5
Case1 Step6 2007-1-7 Step4 2007-1-6
Case2 Step1 2007-2-1 Step2 2007-2-2
Case2 Step2 2007-2-2 Step3 2007-2-3 Step1 2007-2-1
Case2 Step3 2007-2-3 Step2 2007-2-2
还可以进一步统计一下两者的相差天数
select caseid,stepid,actiondate,nextactiondate,nextactiondate-actiondate datebetween from (
select caseid,stepid,actiondate,lead(stepid) over (partition by caseid order by actiondate) nextstepid,
lead(actiondate) over (partition by caseid order by actiondate) nextactiondate,
lag(stepid) over (partition by caseid order by actiondate) prestepid,
lag(actiondate) over (partition by caseid order by actiondate) preactiondate
from lead_table)
结果如下:
Case1 Step1 2007-1-1 2007-1-2 1
Case1 Step2 2007-1-2 2007-1-3 1
Case1 Step3 2007-1-3 2007-1-4 1
Case1 Step4 2007-1-4 2007-1-5 1
Case1 Step5 2007-1-5 2007-1-6 1
Case1 Step4 2007-1-6 2007-1-7 1
Case1 Step6 2007-1-7
Case2 Step1 2007-2-1 2007-2-2 1
Case2 Step2 2007-2-2 2007-2-3 1
Case2 Step3 2007-2-3
每一条记录都能连接到上/下一行的内容
lead () 下一个值 lag() 上一个值
select caseid,stepid,actiondate,lead(stepid) over (partition by caseid order by actiondate) nextstepid,
lead(actiondate) over (partition by caseid order by actiondate) nextactiondate,
lag(stepid) over (partition by caseid order by actiondate) prestepid,
lag(actiondate) over (partition by caseid order by actiondate) preactiondate
from lead_table
下面介绍用于计算分析函数结果的操作。
(1)WINDOW (BUFFER)
分析函数(Analytic Function)在窗口内存中进行聚集计算。
关键词释义
分析函数(Analytic Function):基于一组数据计算出聚集结果。这样一组数据就称为一个窗口(window),由分析函数中的OVER ( … )子句定义。
通过分析函数可以避免由复杂的执行操作(如关联)得出分析数据。示例如下:
HELLODBA.COM>exec sql_explain(‘select table_name, count(1) over (partition by table_name) cnt from
t_tables t’, ‘TYPICAL’);
Plan hash value: 4083014614
——————————————————————————–
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
——————————————————————————–
| 0 | SELECT STATEMENT | | 2070 | 37260 | 7 (0)| 00:00:07 |
| 1 | WINDOW BUFFER | | 2070 | 37260 | 7 (0)| 00:00:07 |
| 2 | INDEX FULL SCAN| T_TABLES_PK | 2070 | 37260 | 7 (0)| 00:00:07 |
——————————————————————————–
(2)WINDOW (SORT)
分析函数(Analytic Function)在窗口内存中对数据排序后,再进行聚集计算。
(3)WINDOW (NOSORT)
分析函数(Analytic Function)在窗口内存中对数据按照数据的物理顺序(或者数据结果有顺序要求,但数据物理顺序与逻辑顺序一致而无需再做排序)进行聚集计算。
(4)WINDOW (SORT PUSHED RANK)
子查询中的分析函数(Analytic Function)在窗口内存中对数据排序时,主查询中的谓词条件被推入子查询或视图中序列函数上,作为子查询或视图的过滤条件。示例如下:
HELLODBA.COM>exec sql_explain(‘select * from (select /*+index_ffs(T T_TABLES_IDX1)*/owner, row_number()
over (order by owner) rnum from t_tables t) where rnum<=10′, ‘TYPICAL’);
Plan hash value: 603754030
——————————————————————————————
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
——————————————————————————————
| 0 | SELECT STATEMENT | | 2070 | 62100 | 3 (34)| 00:00:04 |
|* 1 | VIEW | | 2070 | 62100 | 3 (34)| 00:00:04 |
|* 2 | WINDOW SORT PUSHED RANK| | 2070 | 12420 | 3 (34)| 00:00:04 |
| 3 | INDEX FAST FULL SCAN | T_TABLES_IDX1 | 2070 | 12420 | 2 (0)| 00:00:03 |
——————————————————————————————
Predicate Information (identified by operation id):
—————————————————
1 – filter(“RNUM”<=10)
2 – filter(ROW_NUMBER() OVER ( ORDER BY “OWNER”)<=10)
从以上执行计划中的谓词信息中可以看到PUSH之后的过滤条件。
(5)WINDOW (NOSORT STOPKEY)
分析函数(Analytic Function)在窗口内存中按照物理顺序处理数据时,处理特定记录数后停止。示例如下:
HELLODBA.COM>exec sql_explain(‘select * from (select /*+index(T T_TABLES_IDX1)*/owner, row_number()
over (order by owner) rnum from t_tables t) where rnum<=10′, ‘TYPICAL’);
Plan hash value: 2520140641
—————————————————————————————-
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
—————————————————————————————-
| 0 | SELECT STATEMENT | | 2070 | 62100 | 4 (0)| 00:00:04 |
|* 1 | VIEW | | 2070 | 62100 | 4 (0)| 00:00:04 |
|* 2 | WINDOW NOSORT STOPKEY| | 2070 | 12420 | 4 (0)| 00:00:04 |
| 3 | INDEX FULL SCAN | T_TABLES_IDX1 | 2070 | 12420 | 4 (0)| 00:00:04 |
—————————————————————————————-
Predicate Information (identified by operation id):
—————————————————
1 – filter(“RNUM”<=10)
2 – filter(ROW_NUMBER() OVER ( ORDER BY “OWNER”)<=10)
(6)WINDOW(IN SQL MODEL (SORT))
依据模型化SQL语句中的规则,建立窗口,对数据排序,进行聚集计算。示例如下:
HELLODBA.COM>begin
2 sql_explain('SELECT statistic#, s
3 FROM t_sesstat
4 MODEL RETURN UPDATED ROWS
5 PARTITION BY (statistic#)
6 DIMENSION BY (sid)
7 MEASURES (value v, 1 s)
8 RULES
9 (
10 s[any] = sum(v) over (partition by statistic#)
11 )','TYPICAL');
12 end;
13 /
Plan hash value: 3747799581
-----------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
-----------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 7280 | 65520 | 3 (0)| 00:00:04 |
| 1 | SQL MODEL ORDERED | | 7280 | 65520 | | |
| 2 | TABLE ACCESS FULL | T_SESSTAT | 7280 | 65520 | 3 (0)| 00:00:04 |
| 3 | WINDOW (IN SQL MODEL) SORT| | | | | |
---------------------------------------
原创文章,作者:ItWorker,如若转载,请注明出处:https://blog.ytso.com/7152.html