Oracle select 中带有 window 函数的运行总数（累计列）有什么问题？

Question

我有一个列中包含 0 或 1 的查询。出于演示目的，我将 1 替换为 77 以便更容易看到：

select dates.d the_date
, case TO_CHAR(dates.d, 'd') when '7' then 0 when '1' then 0 
    else 77 
  end as is_workday
from (SELECT (to_date('01.01.2019','dd.mm.yyyy') + (LEVEL -1)) 
      AS d FROM DUAL connect by level <=( 
      to_date('31.12.2020','dd.mm.yyyy')-(to_date('01.01.2019','dd.mm.yyyy')))
     ) as dates

如果此日期是否为工作日，则结果只是一个连续的日期列和一个标记。（在现实生活中我也计算假期，但这不是问题）：

| THE_DATE            | IS_WORKDAY |
| 2019-01-01 00:00:00 | 77         |
| 2019-01-02 00:00:00 | 77         |
| 2019-01-03 00:00:00 | 77         |
| 2019-01-04 00:00:00 | 77         |
| 2019-01-05 00:00:00 |  0         |
| 2019-01-06 00:00:00 |  0         |
| 2019-01-07 00:00:00 | 77         |
| 2019-01-08 00:00:00 | 77         |
| 2019-01-09 00:00:00 | 77         |
| 2019-01-10 00:00:00 | 77         |
....

我想在 is_workday 上加一个运行总数，意思是累计值。我确信 Oracles window 函数就是为此而设计的。

SELECT x.the_date
, x.is_workday
, sum(x.is_workday) over (
    partition by x.the_date          -- define the window
    order by x.the_date asc          -- order inside window
    rows between unbounded preceding -- sum to top
    and current row                  -- sum ending here
  ) as workdays_cumul
FROM (
  select dates.d the_date
  , case TO_CHAR(dates.d, 'd') when '7' then 0 when '1' then 0 
      else 77 
    end as is_workday
  from (SELECT (to_date('01.01.2019','dd.mm.yyyy') + (LEVEL -1)) 
        AS d FROM DUAL connect by level <=( 
        to_date('31.12.2020','dd.mm.yyyy')-(to_date('01.01.2019','dd.mm.yyyy')))
       ) as dates
) x
order by x.the_date
;

但我必须在这里遗漏一些东西，因为我没有得到运行总数，而只是价值本身。

| THE_DATE            | IS_WORKDAY | WORKDAYS_CUMUL |
| 2019-01-01 00:00:00 |  77        |  77            |
| 2019-01-02 00:00:00 |  77        |  77            |
| 2019-01-03 00:00:00 |  77        |  77            |
| 2019-01-04 00:00:00 |  77        |  77            |
| 2019-01-05 00:00:00 |   0        |   0            |
| 2019-01-06 00:00:00 |   0        |   0            |
| 2019-01-07 00:00:00 |  77        |  77            |
| 2019-01-08 00:00:00 |  77        |  77            |
....

显然应该是：

| THE_DATE            | IS_WORKDAY | WORKDAYS_CUMUL |
| 2019-01-01 00:00:00 |  77        |   77           |
| 2019-01-02 00:00:00 |  77        |  154           |
| 2019-01-03 00:00:00 |  77        |  231           |
...

我以为它会像这样：

sum(x.is_workday) -- 对 77 值
partition by x.the_date -- 使 windows/sections/parts 每行一行（在我的例子中）
order by x.the_date asc -- 按日期排序这些行
rows between unbounded preceding -- 第一行之间的总和...
and current row -- ...和当前行。

我在这里错过了什么？

Answer 1

删除查询的 PARTITION BY 子句，因为您希望 window 的范围是整个查询而不是每个单独的日期：

SELECT the_date
,      is_workday
,      sum(is_workday) over (
         ORDER BY the_date asc            -- order inside window
         ROWS BETWEEN unbounded preceding -- sum to top
                  AND current row         -- sum ending here
       ) as workdays_cumul
FROM (
  select d the_date
  ,      CASE
         WHEN d - TRUNC( d, 'IW' ) IN ( 5, 6 )
         THEN 0
         ELSE 1 
         END AS is_workday
  FROM (
    SELECT DATE '2019-01-01' + LEVEL -1 AS d
    FROM   DUAL
    connect by level <= DATE '2020-12-31' - DATE '2019-01-01'
  )
)
order by the_date;

你也可以使用日期文字，需要删除 table 别名中的 AS （并且实际上不需要任何 table 别名）并且可以使用之间的区别日期和其 ISO 周的开始 (d - TRUNC( d, 'IW' )) 作为查找星期六和星期日的方法，独立于 NLS_TERRITORY 会话参数。

您也可以删除 ROWS BETWEEN 子句，因为 ROWS BETWEEN unbounded preceding AND current row 是默认行为。但是，如果这是您需要的行为，那么您也可以保留它以证明 window 是您的预期结果。

输出:

THE_DATE  | IS_WORKDAY | WORKDAYS_CUMUL
:-------- | ---------: | -------------:
01-JAN-19 |          1 |              1
02-JAN-19 |          1 |              2
03-JAN-19 |          1 |              3
04-JAN-19 |          1 |              4
05-JAN-19 |          0 |              4
06-JAN-19 |          0 |              4
07-JAN-19 |          1 |              5
08-JAN-19 |          1 |              6
09-JAN-19 |          1 |              7
...
24-DEC-20 |          1 |            518
25-DEC-20 |          1 |            519
26-DEC-20 |          0 |            519
27-DEC-20 |          0 |            519
28-DEC-20 |          1 |            520
29-DEC-20 |          1 |            521
30-DEC-20 |          1 |            522

db<>fiddle here

Oracle select 中带有 window 函数的运行总数（累计列）有什么问题？

What is wrong with this running total (cumulative column) in Oracle select with a window function?

sql

oracle

cumulative-sum

window-functions

Oracle select 中带有 window 函数的 运行 总数（累计列）有什么问题？

What is wrong with this running total (cumulative column) in Oracle select with a window function?

sql

oracle

cumulative-sum

window-functions

Oracle select 中带有 window 函数的运行总数（累计列）有什么问题？