Christoffer
Christoffer

Reputation: 347

SQL Server Calculate cumulative sum / conditional running total depending on different flags

I have a table in SQL Server with data looking like this example.

ID  Flag    Art.No  Amount
1   U   A1000   -100
2   U   B2000   -5
3   V   B2000   900
4   U   B2000   -10
5   I   B2000   50
6   U   B2000   -20
7   U   A1000   -50
8   I   A1000   1000
9   V   A1000   3600
10  U   A1000   -500
11  U   A1000   -100
12  U   A1000   -2000
13  I   A1000   2000
14  U   A1000   -1000
15  I   C3000   10000
16  U   C3000   -4000
17  U   B2000   -5
18  U   B2000   -5
19  I   B2000   40
20  V   B2000   200
21  U   A1000   -500
22  U   B2000   -50
23  U   C3000   -1000

I want to calculate ackumulated value based on the transactions. My problem is that the table contains 3 types of transactions.

  1. Flag U - Sales
  2. Flag I - Incoming goods
  3. Flag V - Stocktaking

When Flag U and I appears the amount represent the change When Flag V appears the amount represent the total amount when stocktaking

In words I want to find the latest V-transaction for each unice Art.No and then add or subtract U and I transactions to get a cummulativ sum for each row. If there is no V-transaction go through the whole dataset.

I have made examples with expected result for each Art.No

A1000

ID  Flag    Art.No  Amount  A1000 Example
1   U   A1000   -100    
7   U   A1000   -50 
8   I   A1000   1000    
9   V   A1000   3600    3600
10  U   A1000   -500    3100
11  U   A1000   -100    3000
12  U   A1000   -2000   1000
13  I   A1000   2000    3000
14  U   A1000   -1000   2000
21  U   A1000   -500    1500

B2000

ID  Flag    Art.No  Amount  B2000 Example
2   U   B2000   -5  
3   V   B2000   900 
4   U   B2000   -10 
5   I   B2000   50  
6   U   B2000   -20 
17  U   B2000   -5  
18  U   B2000   -5  
19  I   B2000   40  
20  V   B2000   200 200
22  U   B2000   -50 150

C3000

ID  Flag    Art.No  Amount  C3000 Example
15  I   C3000   10000   10000
16  U   C3000   -4000   6000
23  U   C3000   -1000   5000

Edit: To get more history in the dataset there would be nice to have values before the latest V-transaction like this

B2000

ID  Flag    Art.No  Amount  B2000 Example
2   U   B2000   -5  150
3   V   B2000   900 140
4   U   B2000   -10 140
5   I   B2000   50  190
6   U   B2000   -20 170
17  U   B2000   -5  165
18  U   B2000   -5  160
19  I   B2000   40  200
20  V   B2000   200 200
22  U   B2000   -50 150

Where each I and U transaction is taken in consideration but V-transactions is ignored.

Upvotes: 7

Views: 8590

Answers (1)

dnoeth
dnoeth

Reputation: 60482

with cte as
 (
   select *,
      -- find the latest 'V' ID per ArtNo
      max(case when Flag = 'V' then ID end) 
      over (partition by ArtNo) as Last_V_ID
   from myTable
 )
select *,
   -- cumulative sum, but ignore all rows before the latest 'V' ID
   -- includes rows when there's no 'V' ID for this ArtNo
   sum(case when ID < Last_V_ID then null else Amount end)
   over (partition by ArtNo
         order by ID
         rows unbounded preceding)
from cte
order by ArtNo, ID

See Fiddle

Edit:

To include the data before the last stocktaking and to ignore all previous stocktakings you can use this approach:

with cte as
 (
   select *,
      -- find the latest 'V' ID per ArtNo
      max(case when Flag = 'V' then ID end) 
      over (partition by ArtNo) as Last_V_ID
   from [dbo].[Warehouse]
 )
select *,
   -- cumulative sum, but ignore all rows before the latest 'V' ID
   -- includes rows when there's no 'V' ID for this ArtNo
   sum(case when ID < Last_V_ID then null else Amount end)
   over (partition by ArtNo
         order by ID
         rows unbounded preceding)
   -- calculate in-stock based on last 'V' ID, discarding all previous 'V' rows
  ,sum(case when (ID < Last_V_ID and Flag <> 'V')  then -Amount 
            when ID = Last_V_ID then Amount 
       end)
   over (partition by ArtNo
         order by ID 
         rows between 1 following and unbounded following)
from cte
order by ArtNo, ID

Both calculations are mutually exlusive, so you can easily combine them using COALESCE.

See Fiddle

Upvotes: 9

Related Questions