Over()視窗函式最常見的搭配有以下幾種：

1. rank(),dense_rank(),row_number() + over(partition by … order by …) 排名
2. sum(),avg(),count()聚合函式+over(partition by … order by …)
3. max(),min()+over(partition by … order by …) 最大值、最小值
4. first_value(),last_value() + over(partition by … order by …) 第一條、最後一條記錄
5. lag()，lead() + over(partition by … order by …) 偏移量

其中的partition by 是分組，order by 是排序。這裡的分組與group by 是不同的，最明顯的是group by會影響返回結果的條數，但是partition by 不會。

案例用到的建表語句：

create table LX_05_SALARY
(
  id              NUMBER,
  department_name VARCHAR2(100),
  sal             NUMBER,
  pay_date        DATE
)

插數語句：

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (1, 'A部門', 80000, to_date('10-01-2020', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (2, 'B部門', 60000, to_date('10-01-2020', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (3, 'C部門', 100000, to_date('10-01-2020', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (4, 'A部門', 70000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (5, 'B部門', 60000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (6, 'C部門', 80000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (7, 'C部門', 48000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (8, 'C部門', 92000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (9, 'B部門', 90000, to_date('10-12-2019', 'dd-mm-yyyy'));

insert into lx_05_salary (ID, DEPARTMENT_NAME, SAL, PAY_DATE)
values (10, 'B部門', 50000, to_date('10-12-2019', 'dd-mm-yyyy'));
 

具體的用法如下：

1. rank(),dense_rank(),row_number() + over(partition by … order by …) 排名：

 select a.department_name,
       a.id,
       a.sal,
       rank() over(partition by a.department_name order by a.sal desc) as rank排名,
       dense_rank() over(partition by a.department_name order by a.sal desc) as dense_rank排名,
       row_number() over(partition by a.department_name order by a.sal desc) as row_number排序
  from lx_05_salary a;
 

由上圖結果可以看出：
rank（）的排名如果有並列出現，下一名會跳過並列的名次。有兩個同時排名第二，那下一條就是第四名。
dense_rank()的排名與rank（）不同，出現並列不會跳過並列的名次，依然按順序排名。例如：兩個第二名，下一條還是第三名。
row_number()不會出現並列的情況，一直順序排下去。
知道這三者之間的區別之後，就需要在實際應用中選擇合適方式來使用。

2. sum(),avg(),count()聚合函式+over(partition by … order by …)

select a.department_name,a.id,a.sal,
       sum(a.sal)over(partition by a.department_name order by a.id ) as 部門內連續求和 ,
       sum(a.sal)over(partition by a.department_name) as 部門求和,
       round(a.sal/sum(a.sal)over(partition by a.department_name),4)*100 as 每人佔部門份額 ,
       sum(a.sal)over(order by a.department_name) as 部門連續求和,
       sum(a.sal)over(order by a.id) as 人員連續求和,
       sum(a.sal) over()  as 總計,
       round(a.sal/sum(a.sal) over(),4)*100 as 人員份額,
       round(sum(a.sal) over(partition by a.department_name)/sum(a.sal) over(),4)*100 as 部門份額 
  from lx_05_salary a 
  order by  a.department_name;

over（）中使用order by 會在分組內連續向下求和。不使用的話就只會在當前分組下求合計。avg()，count()的用法是一致的。

3. max(),min()+over(partition by … order by …) 最大值、最小值

       select a.department_name,
         a.id,
         a.sal,
         max(a.sal) over(partition by a.department_name order by a.sal desc) as max_desc,
         min(a.sal) over(partition by a.department_name order by a.sal desc) as min_desc_失效,
         min(a.sal) over(partition by a.department_name order by a.sal desc rows BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) as min_desc,
         max(a.sal) over(partition by a.department_name order by a.sal asc) as max_asc_失效, 
         max(a.sal) over(partition by a.department_name order by a.sal asc rows BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) as max_asc,
         min(a.sal) over(partition by a.department_name order by a.sal asc)  as min_asc
    from lx_05_salary a  

min(),max()在使用order by的時候會有無效的情況，要麼就去掉order by 要麼向上面例子使用（rows BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING）具體的可以參考這位大神的文章

4. first_value(),last_value() + over(partition by … order by …) 第一條、最後一條記錄。
   first_value(),last_value() 與max()和min()一樣，在升序或者降序排序的時候也會有無效的情況。

select a.department_name,
         a.id,
         a.sal,
         first_value(a.sal) over(partition by a.department_name order by a.sal desc) as first_value,
         last_value(a.sal) over(partition by a.department_name order by a.sal desc rows BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) as last_value,
         last_value(a.sal) over(partition by a.department_name order by a.sal asc)  as last_value_無效
    from lx_05_salary a;

5. lag()，lead() + over(partition by … order by …) 偏移量
   語法：
   lag(目標欄位,偏移量，預設值) + over(partition by … order by …)
   lead(目標欄位,偏移量，預設值) + over(partition by … order by …)

舉例：
沒有預設值的會取空

select a.*,
       lag(a.id, 1 ) over(partition by a.department_name order by a.id) as 同部門_上一id,
       lead(a.id, 1 ) over(partition by a.department_name order by a.id) as 同部門_下一id,
       lag(a.id,1) over(order by a.id) as 全部_上一id ,
       lead(a.id,1) over(order by a.id) as 全部_下一id  
  from lx_05_salary a
order by a.department_name,a.id;

如果不寫偏移量，預設是1，如下例：

select a.*,
       lag(a.id ) over(partition by a.department_name order by a.id) as 同部門_上一id,
       lead(a.id ) over(partition by a.department_name order by a.id) as 同部門_下一id,
       lag(a.id ) over(order by a.id) as 全部_上一id ,
       lead(a.id ) over(order by a.id) as 全部_下一id  
  from lx_05_salary a
order by a.department_name,a.id;

設定預設值的寫法：

select a.*,
       lag(a.id, 1,999999 ) over(partition by a.department_name order by a.id) as 同部門_上一id,
       lead(a.id, 1,999999  ) over(partition by a.department_name order by a.id) as 同部門_下一id,
       lag(a.id,1,999999 ) over(order by a.id) as 全部_上一id ,
       lead(a.id,1,999999 ) over(order by a.id) as 全部_下一id  
  from lx_05_salary a
order by a.department_name,a.id;