SQL 查询使用中断事件获取运行总计

Question

There are cities with reports of rainy day or sunny day. Each flag value represents a rain event(flag=1 Sunny day flag=0 Rainy day). Each row I need to produce a result how many days transpired before the rain inclusive of the day it rained. For example for the city "A" below on 1/10/2017 the expected result is 2 counting 1/05 and 1/10 as it rained on 1/10. I tried with CTE which works for some rows but not all of them. Each rain event resets the running totals. So each rainy day has its own running total after the last rain event, and doesn't accumulate from previous rainy days.

I have the column with expected result in this table below.

CREATE TABLE #Sample
  ([Date] datetime,
   [Name] char(1),
   [flag] bit,
   [expected_result] int
   );
GO

--drop table #Sample

INSERT INTO #Sample
VALUES
  ('1/1/2017','A',0,1),
  ('1/2/2017','B',0,1),
  ('1/3/2017','C',1,0),
  ('1/4/2017','D',0,1),
  ('1/5/2017','A',1,0),
  ('1/6/2017','C',1,0),
  ('1/7/2017','C',1,0),
  ('1/8/2017','B',0,1),
  ('1/9/2017','D',1,0),
  ('1/11/2017','C',0,4),
  ('1/10/2017','A',0,2),
  ('1/11/2017','A',0,1),
  ('1/12/2017','A',1,0),
  ('1/13/2017','A',1,0),
  ('1/14/2017','A',0,3),
  ('1/15/2017','D',0,2),
  ('1/16/2017','D',1,0);
GO

SELECT * FROM #Sample
order by Name,Date

I tried the below with no success

with t as 
(
SELECT S1.Date, S1.Name,
       --s1.[Count] AS [Count],
       S1.flag,
       SUM(CAST(S1.flag AS INT)) OVER(PARTITION BY S1.Name,S1.flag Order BY s1.[Name], S1.Date) AS Sunny_Days,
       [expected_result]
FROM #Sample S1
)
select  t.*, 
CASE WHEN flag=0 AND LAG(flag,1,0) OVER(PARTITION BY Name Order BY [Name], Date)=1 THEN
LAG(Sunny_Days,1,0)  OVER(PARTITION BY Name Order BY [Name], Date)+1
ELSE
0
END as Days_Transpired_Before_Rain from t
order by Name,Date

Answer 1

The way that you have described this is rather tricky. A simpler way is to assign a group to rows based on the number of rainy days that are on or after each row.

Then, each rainy day is in a group with all the preceding non-rainy days. Just count the size of the group to get the number you want.

In SQL:

select s.*,
       (case when flag = 0 then count(*) over (partition by name, grp) 
             else 0
        end) as calculated_result
from (select s.*,
             sum(case when flag = 0 then 1 else 0 end) over (partition by name order by date desc) as grp
      from sample s
     ) s
order by date

Here is a db<>fiddle.