Calculate sum of maximum and minimum value for repeated group of entries in MySQL8

Question

LogDateAndTime	BatchDate	TagLetter	Totaliser	ExpectedResult
10-11-2020 09:06:14	10-11-2020 08:29:55	A	6319	31
10-11-2020 09:06:24	10-11-2020 08:29:55	A	6337	31
10-11-2020 09:08:14	10-11-2020 08:29:55	B	6355	31
10-11-2020 09:08:24	10-11-2020 08:29:55	B	6372	31
10-11-2020 09:08:34	10-11-2020 08:29:55	B	6378	31
10-11-2020 09:08:44	10-11-2020 08:29:55	A	6383	31
10-11-2020 09:09:14	10-11-2020 08:29:55	A	6388	31
10-11-2020 09:09:24	10-11-2020 08:29:55	A	6396	31
10-11-2020 09:09:34	10-11-2020 08:29:55	B	6409	31
10-11-2020 09:09:44	10-11-2020 08:29:55	B	6426	31
10-11-2020 09:10:24	10-11-2020 08:29:55	B	6442	31

The above table has LogDateAndTime(Primary_Key) column with unique datetime entries. The BatchDate column holds same datetime values throughout the entire batch. I need to calculate the sum of MAX(Totaliser)-MIN(Totaliser) for each instance of TagLetter=A so that i should ignore values in TagLetter=B. My ExpectedResult in this case will be SUM[(6337-6319)+(6396-6383)]=31. I tried below query, but did not get the expected result.

SELECT SUM(
           CASE 
             WHEN TagLetter='A' THEN MAX(Totaliser)-MIN(Totaliser) 
             ELSE 0.0 
           END
          ) OVER (PARTITION BY BatchDate) AS ExpectedResult

In this case, it is calculating 6396-6319=77 which is not the expected result. Could someone help me to get the result right?

Answer 1

First create the groups of consecutive 'A's with window functions LAG() and SUM() and then aggregate on these groups:

WITH cte AS (
  SELECT DISTINCT SUM(MAX(Totaliser) - MIN(Totaliser)) OVER () ExpectedResult
  FROM (
    SELECT *, SUM(flag) OVER (ORDER BY LogDateAndTime) grp
    FROM (
      SELECT *, LAG(TagLetter, 1, '') OVER (ORDER BY LogDateAndTime) <> 'A' flag
      FROM tablename 
    ) t
    WHERE TagLetter = 'A'
  ) t
  GROUP BY grp
)
SELECT t.*, c.ExpectedResult
FROM tablename t CROSS JOIN cte c

Or if you want results for each BatchDate:

WITH cte AS (
  SELECT DISTINCT BatchDate,
         SUM(MAX(Totaliser) - MIN(Totaliser)) OVER () ExpectedResult
  FROM (
    SELECT *, SUM(flag) OVER (PARTITION BY BatchDate ORDER BY LogDateAndTime) grp
    FROM (
      SELECT *, LAG(TagLetter, 1, '') OVER (PARTITION BY BatchDate ORDER BY LogDateAndTime) <> 'A' flag
      FROM tablename 
    ) t
    WHERE TagLetter = 'A'
  ) t
  GROUP BY BatchDate, grp
)
SELECT t.*, c.ExpectedResult
FROM tablename t LEFT JOIN cte c
ON c.BatchDate = t.BatchDate

See the demo.
Results:

> LogDateAndTime      | BatchDate           | TagLetter | Totaliser | ExpectedResult
> :------------------ | :------------------ | :-------- | --------: | -------------:
> 10-11-2020 09:06:14 | 10-11-2020 08:29:55 | A         |      6319 |             31
> 10-11-2020 09:06:24 | 10-11-2020 08:29:55 | A         |      6337 |             31
> 10-11-2020 09:08:14 | 10-11-2020 08:29:55 | B         |      6355 |             31
> 10-11-2020 09:08:24 | 10-11-2020 08:29:55 | B         |      6372 |             31
> 10-11-2020 09:08:34 | 10-11-2020 08:29:55 | B         |      6378 |             31
> 10-11-2020 09:08:44 | 10-11-2020 08:29:55 | A         |      6383 |             31
> 10-11-2020 09:09:14 | 10-11-2020 08:29:55 | A         |      6388 |             31
> 10-11-2020 09:09:24 | 10-11-2020 08:29:55 | A         |      6396 |             31
> 10-11-2020 09:09:34 | 10-11-2020 08:29:55 | B         |      6409 |             31
> 10-11-2020 09:09:44 | 10-11-2020 08:29:55 | B         |      6426 |             31
> 10-11-2020 09:10:24 | 10-11-2020 08:29:55 | B         |      6442 |             31