oracle 进阶之model子句

　　本博客是自己在学习和工作途中的积累与总结，仅供自己参考，也欢迎大家转载，转载时请注明出处。

　　http://www.cnblogs.com/king-xg/p/6692841.html

一，  model子句

　　 　　制作表格数据，用传统sql来实现的话，一般通过多个表的自联结实现，而model的出现则使得不用自联结就能实现表格，因为model拥有了跨行应用能力。

　　(1) 语法

　　　　MODEL
　　　　[]
　　　　[]
　　　　[MAIN ]
  　　　 [PARTITION BY ()]
  　　    DIMENSION BY ()
  　　    MEASURES ()
  　　　 []
  　　　 [RULES]
  　　　 (, ,.., )
  　　　 ::=
 　　　  ::= RETURN {ALL|UPDATED} ROWS
  　　　 ::=
  　　　 [IGNORE NAV | [KEEP NAV]
  　　　 [UNIQUE DIMENSION | UNIQUE SINGLE REFERENCE]
  　　　 ::=
  　　　 [UPDATE | UPSERT | UPSERT ALL]
  　　　 [AUTOMATIC ORDER | SEQUENTIAL ORDER]
  　　　 [ITERATE ()  [UNTIL ]]
  　　　 ::= REFERENCE ON ON ()
  　　　 DIMENSION BY () MEASURES ()

------------建表，初始化数据，才好讲下面的内容----------------------

-- 创建表
create table ademo(
       id number(18) primary key,
       year varchar2(4),
       week number(8),
       sale number(8,2),
       area varchar2(100)
);

-- 创建序列
create sequence seq_ademo_id
minvalue 1
start with 1
increment by 1
nomaxvalue
nocache
nocycle;

-- 创建触发器
create or replace trigger trigger_ademo_id
before insert on ademo for each row when (new.id is null)
begin
  select seq_ademo_id.nextval into :new.id from dual;
end;

-- 初始化数据
insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2000', 1, 52.12);

insert into  ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2001', 1, 110.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2001', 2, 110.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2001', 3, 1210.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2002', 1, 170.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2002', 2, 680.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('astiya', '2002', 3, 680.12);

insert into  ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2001', 1, 80.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2001', 2, 56.72);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2001', 3, 156.72);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2002', 1, 640.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2002', 2, 980.12);

insert into ademo (AREA, YEAR, WEEK, SALE)
values ('anter', '2002', 3, 1980.12);

/*delete from ademo;*/

-- 注释(这是我的个人习惯，不想麻烦的可以不加)
comment on table ademo is '测试经济类的表';
comment on column ademo.id is '主键';
comment on column ademo.year is '年份';
comment on column ademo.week is 'xxx周';
comment on column ademo.sale is '销售额';
comment on column ademo.area is '地区';

-- 展示数据
select * from ademo;

-- 例子1
select year,week,sale,area,up_sale
from ademo
model return updated rows    -- model 语句
partition by (area)　　　　　　-- 分组
dimension by (year,week)　　　-- 维度列
measures(sale,0 up_sale)　　　-- 度量值列
rules(　　　　　　　　　　　　　 -- 规则
    up_sale[year,week]=sale[cv(year),cv(week)]*10,
    up_sale[1999,1]=100.00　　
)order by year,week;

-- 例子2
select year,week,sale,area,up_sale
from ademo
model
partition by (area)
dimension by (year,week)
measures(sale,0 up_sale)
rules(
    up_sale[year,week]=sale[cv(year),cv(week)]*10,
    up_sale[1999,for week from 1 to 3 increment 1]=100.00
)order by year,week;

--------------------------------------------------------------------

　　(2) 规则

　　　　a. 位置标记

　　　　即指定确定的位置明确的维度列值，例如:例子1中的规则(rules)中的up_sale[1999,1]=100.00，明确指出，year=1999,week=1的up_sale列的值为100.00，

　　　　作用: 位置标记通常也叫UPSERT，即update and insert，当结果集中不存在则插入，数量随分组的数量而定；存在时，则更新数据，更新的数据条数同样与分组的组数相同。

　　　 b. 符号标记

　　　　即指定范围的度量列值，例如:例子2中，up_sale[1999,for week from 1 to 3 increment 1]=100.00,指出，week的范围是在1-3，增长步长为1，所以在每个组中添加了3个up_sale[1999,1..3]，共9个。

　　　　作用:只能更新数据

　　(3) model 返回更新后的行

　　　　在例子1中，model return updated rows 中，的“return updated rows”表示返回在本次操作中更新或插入的新纪录。默认返回所有符合条件的记录

　　(4) 在model的规则中是能够使用一般的聚合函数的，例如：count,sum,ave,stddev,PLAP。

　　(5) model 查找表,功能类似于表连接

　　

---- 查询表

-- 创建表(销售表)
create table product_cost(
       id number(18) primary key,
       year number(4),
       month number(2),
       pid number(18),
       countSum number(18)
);

comment on table product_cost is '产品销售表';
comment on column product_cost.id is '主键';
comment on column product_cost.year is '年份';
comment on column product_cost.month is '月份';
comment on column product_cost.pid is '产品id';
comment on column product_cost.countSum is '销售数量';

-- 创建表(产品表)
create table product(
       id number(18) primary key,
       pname varchar(100),
       price number(8,2)
);

comment on table product is '产品表';
comment on column product.id is '主键';
comment on column product.pname is '产品名称';
comment on column product.price is '单价';

-- 创建序列
create sequence seq_product_cost_id
minvalue 1
start with 1
increment by 1
nomaxvalue
nocache
nocycle;

create sequence seq_product_id
minvalue 1
start with 1
increment by 1
nomaxvalue
nocache
nocycle;

-- 创建触发器
create or replace trigger trigger_product_cost_id
before insert on product_cost for each row when (new.id is null)
begin
  select seq_product_cost_id.nextval into :new.id from dual;
end;

create or replace trigger trigger_product_id
before insert on product for each row when (new.id is null)
begin
  select seq_product_id.nextval into :new.id from dual;
end;

-- 初始化数据
insert into product (pname,price) values('i7-6700K','23');
insert into product (pname,price) values('i7-6600K','20');
insert into product (pname,price) values('i7-6500K','19');
insert into product (pname,price) values('i7-6400K','18');
insert into product (pname,price) values('i7-6300K','17');
insert into product (pname,price) values('i7-6200K','15');
insert into product (pname,price) values('i7-6100K','12');

delete from product;

select * from product;

insert into product_cost(year,month,pid,countSum) values(2000,1,1,500);
insert into product_cost(year,month,pid,countSum) values(2000,1,2,630);
insert into product_cost(year,month,pid,countSum) values(2000,1,3,1200);
insert into product_cost(year,month,pid,countSum) values(2000,1,4,320);
insert into product_cost(year,month,pid,countSum) values(2000,1,5,150);
insert into product_cost(year,month,pid,countSum) values(2000,1,6,250);
insert into product_cost(year,month,pid,countSum) values(2000,1,7,350);

insert into product_cost(year,month,pid,countSum) values(2000,2,1,1500);
insert into product_cost(year,month,pid,countSum) values(2000,2,2,1630);
insert into product_cost(year,month,pid,countSum) values(2000,2,3,200);
insert into product_cost(year,month,pid,countSum) values(2000,2,4,1320);
insert into product_cost(year,month,pid,countSum) values(2000,2,5,250);
insert into product_cost(year,month,pid,countSum) values(2000,2,6,350);
insert into product_cost(year,month,pid,countSum) values(2000,2,7,450);

insert into product_cost(year,month,pid,countSum) values(2000,3,1,520);
insert into product_cost(year,month,pid,countSum) values(2000,3,2,660);
insert into product_cost(year,month,pid,countSum) values(2000,3,3,1900);
insert into product_cost(year,month,pid,countSum) values(2000,3,4,300);
insert into product_cost(year,month,pid,countSum) values(2000,3,5,180);
insert into product_cost(year,month,pid,countSum) values(2000,3,6,210);
insert into product_cost(year,month,pid,countSum) values(2000,3,7,320);

insert into product_cost(year,month,pid,countSum) values(2000,4,1,1520);
insert into product_cost(year,month,pid,countSum) values(2000,4,2,1660);
insert into product_cost(year,month,pid,countSum) values(2000,4,3,2900);
insert into product_cost(year,month,pid,countSum) values(2000,4,4,1200);
insert into product_cost(year,month,pid,countSum) values(2000,4,5,980);
insert into product_cost(year,month,pid,countSum) values(2000,4,6,910);
insert into product_cost(year,month,pid,countSum) values(2000,4,7,620);

insert into product_cost(year,month,pid,countSum) values(2001,1,1,500);
insert into product_cost(year,month,pid,countSum) values(2001,1,2,630);
insert into product_cost(year,month,pid,countSum) values(2001,1,3,1200);
insert into product_cost(year,month,pid,countSum) values(2001,1,4,320);
insert into product_cost(year,month,pid,countSum) values(2001,1,5,150);
insert into product_cost(year,month,pid,countSum) values(2001,1,6,250);
insert into product_cost(year,month,pid,countSum) values(2001,1,7,350);

insert into product_cost(year,month,pid,countSum) values(2001,2,1,1500);
insert into product_cost(year,month,pid,countSum) values(2001,2,2,1630);
insert into product_cost(year,month,pid,countSum) values(2001,2,3,200);
insert into product_cost(year,month,pid,countSum) values(2001,2,4,1320);
insert into product_cost(year,month,pid,countSum) values(2001,2,5,250);
insert into product_cost(year,month,pid,countSum) values(2001,2,6,350);
insert into product_cost(year,month,pid,countSum) values(2001,2,7,450);

insert into product_cost(year,month,pid,countSum) values(2001,3,1,520);
insert into product_cost(year,month,pid,countSum) values(2001,3,2,660);
insert into product_cost(year,month,pid,countSum) values(2001,3,3,1900);
insert into product_cost(year,month,pid,countSum) values(2001,3,4,300);
insert into product_cost(year,month,pid,countSum) values(2001,3,5,180);
insert into product_cost(year,month,pid,countSum) values(2001,3,6,210);
insert into product_cost(year,month,pid,countSum) values(2001,3,7,320);

insert into product_cost(year,month,pid,countSum) values(2001,4,1,1520);
insert into product_cost(year,month,pid,countSum) values(2001,4,2,1660);
insert into product_cost(year,month,pid,countSum) values(2001,4,3,2900);
insert into product_cost(year,month,pid,countSum) values(2001,4,4,1200);
insert into product_cost(year,month,pid,countSum) values(2001,4,5,980);
insert into product_cost(year,month,pid,countSum) values(2001,4,6,910);
insert into product_cost(year,month,pid,countSum) values(2001,4,7,620);

select * from product_cost;
select * from product;

在传统sql实现:
   select pc.year as year,pc.month as month, p.pname as pname,pc.countSum as count, (pc.countSum * p.price) as sale
   from product_cost pc left join product p
   on pc.pid=p.id

　where year=2000 and month=4;

　-- 解释计划
    explain plan for
    select pc.year as year,pc.month as month, p.pname as pname,pc.countSum as count, (pc.countSum * p.price) as sale
    from product_cost pc left join product p
    on pc.pid=p.id where year=2000 and month=4;
    commit;

　-- 查看解释计划

　   select * from table(dbms_xplan.display);

　　

　　由于没添加索引所以全盘扫描。

　model查找表:
    select year,month,pid,pname,price,sale,countSum
    from product_cost
    model
    reference ref_pro on
    (
        select id,pname,price
        from product
    )
    dimension by (id)
    measures (pname,price)
    main main_selection
    partition by (year,month)
    dimension by (pid)
    measures(countSum,cast(' ' as varchar2(200))pname, cast(0 as number(18,2))sale, cast(0 as number(8,2))price)
    rules (
          pname[pid] =ref_pro.pname[cv(pid)],
          price[pid]=ref_pro.price[cv(pid)],
          countSum[pid]=countSum[cv(pid)],
          sale[pid]=price[cv(pid)]*countSum[cv(pid)]
    ) where year=2000 and month=4 order by year,month,pid;
    
    
    -- 解释计划
    explain plan for
    
     select year,month,pid,pname,price,sale,countSum
    from product_cost
    model
    reference ref_pro on
    (
        select id,pname,price
        from product
    )
    dimension by (id)
    measures (pname,price)
    main main_selection
    partition by (year,month)
    dimension by (pid)
    measures(countSum,cast(' ' as varchar2(200))pname, cast(0 as number(18,2))sale, cast(0 as number(8,2))price)
    rules (
          pname[pid] =ref_pro.pname[cv(pid)],
          price[pid]=ref_pro.price[cv(pid)],
          countSum[pid]=countSum[cv(pid)],
          sale[pid]=price[cv(pid)]*countSum[cv(pid)]
    ) where year=2000 and month=4 order by year,month,pid;
    commit;

　 -- 查看解释计划
    select * from table(dbms_xplan.display);
　

　两者相比较，model子句的性能会更好，即便在没有索引的情况下，model子句预期访问的字节数要小于传统的sql自联结，那这是为什么呢?

　其实这与model的内部分组机制有关，谓语中的字段含有分组(partition by)中的字段，所以，model就会仅仅访问谓语指定的分区，其他分区不管，这很大程度上提高了sql的性能。

　(5)谓语前推

　--谓语前推
   -- 内嵌视图
   select *
   from ( select year,week,sale,area,0 as new_sale from ademo)
   model return updated rows
   partition by (year,week)
   dimension by (area)
   measures (sale, 0 new_sale)
   rules (
         new_sale[area]=sale[cv(area)]*10
   )
   order by year,week;
   
   -- 成功将谓语推入视图
   explain plan for
   select *
   from ( select year,week,sale,area,0 as new_sale from ademo
   model return updated rows
   partition by (year,week)
   dimension by (area)
   measures (sale, 0 new_sale)
   rules (
         new_sale[area]=sale[cv(area)]*10
   )
   )where year=2001
   order by year,week;
   commit;
   -- 解释计划
select * from table(dbms_xplan.display);

　　在一开始全表扫描的时候就执行了过滤，减少了扫描的数据块的数，降低了加载的字节数。

　　然后看看下面推入失败的sql　　

-- 失败,在全表扫描完后的结果集上进行过滤，并未退入到视图
   explain plan for
   select *
   from ( select year,week,sale,area,0 as new_sale from ademo
   model return updated rows
   partition by (year,week)
   dimension by (area)
   measures (sale, 0 new_sale)
   rules (
         new_sale[area]=sale[cv(area)]*10
   )
   )where area='kinggardom'
   order by year,week;
   commit;

-- 解释计划
select * from table(dbms_xplan.display);

　　很明显，过滤实在view操作的时候进行，即在得到全包扫描后的结果集后进行过滤，无疑说明此次谓语前推失败。

　　-- 原因：在model中存在一种分区的机制，partition by是进行分区的判断依据，那么若果在外sql中存在与分区列匹配的列，则model子句就会只扫描匹配的分区，其他分区就不管了，如果不存在则，全表扫描或者说扫描所有分区
   　-- 结论: 谓语中，能被推入到视图中的仅仅只有分组中的字段(partition by(字段))

　

　　(6) 子查询因子化(小小的提一下，后期再出详细的笔记)

　　 -- 格式: with [alias] as () select ...
  　  -- 题目: 将同一年的一月前的sale进行对比，查看是增长还是下降了多少
  　  -- 神似内嵌视图

举例:

　

with t as (
    select year,month,pid,pname,price,sale,countSum
    from product_cost
    model
    reference ref_pro on
    (
        select id,pname,price
        from product
    )
    dimension by (id)
    measures (pname,price)
    main main_selection
    partition by (year,month)
    dimension by (pid)
    measures(countSum,cast(' ' as varchar2(200))pname, cast(0 as number(18,2))sale, cast(0 as number(8,2))price)
    rules (
          pname[pid] =ref_pro.pname[cv(pid)],
          price[pid]=ref_pro.price[cv(pid)],
          countSum[pid]=countSum[cv(pid)],
          sale[pid]=price[cv(pid)]*countSum[cv(pid)]
    )

  )
  select year,month,pname,sale,pre_sale,compare_pre_sale
  from t
  model
  partition by (pname)
  dimension by (year,month)
  measures (0 pre_sale,0 compare_pre_sale,sale)
  rules(
          pre_sale[year,month]=presentnnv(sale[cv(year),cv(month)-1],sale[cv(year),cv(month)-1],sale[cv(year),cv(month)]),
          compare_pre_sale[year,month]=sale[cv(year),cv(month)]-pre_sale[cv(year),cv(month)]

  )order by pname,month;
  

其实这个可以不用把规则分开写，因为数据并不复杂，对于复杂的数据，用这个还是很不错的选择。

小结: model用来制作表格数据比之传统的表联结来实现会是一个更好的选择，model子句能够提供更好的sql性能，提供更清晰的结构。