Selecting fields that are not in GROUP BY [duplicate] - sql

Question

This question already has answers here:
Get top 1 row of each group
(19 answers)
Closed last year.
I am working with group by in SQL Server. I want to select fields that shouldn't be part of group by.
I want to select DishId but it shouldn't be in group by clause. If I add the DishId into group by it repeats the range as shown following.
SELECT v.range, COUNT(*) AS 'occurrences',COUNT(DishID) * SUM(OrderQty) AS [TotalOrders],po.DishID
FROM ( SELECT PgrId,CASE WHEN DATEDIFF(YEAR, p.PgrDOB, GETDATE()) >= 0 AND DATEDIFF(YEAR, p.PgrDOB, GETDATE()) < 10 THEN '0-9'
WHEN DATEDIFF(YEAR, p.PgrDOB, GETDATE()) >= 10 AND DATEDIFF(YEAR, p.PgrDOB, GETDATE()) < 20 THEN '10-19'
WHEN DATEDIFF(YEAR, p.PgrDOB, GETDATE()) >= 20 AND DATEDIFF(YEAR, p.PgrDOB, GETDATE()) < 30 THEN '20-29'
WHEN DATEDIFF(YEAR, p.PgrDOB, GETDATE()) >= 30 AND DATEDIFF(YEAR, p.PgrDOB, GETDATE()) < 40 THEN '30-39'
ELSE '40+'
END AS 'range'
FROM Passenger p
) v inner join PassengerOrder po on v.PgrId = po.PgrID
GROUP BY v.range,po.DishID

Answer 1

Sounds like you want the highest dish in each age range, ordered by total sales descending. There are many other improvements that could be made here but the simplest way is to just generate a row number per range ordered by total sales descending, and wrap that in a CTE, filtered by the first row number.
;WITH cte AS
(
SELECT v.range,
COUNT(*) AS occurrences,
COUNT(DishID) * SUM(OrderQty) AS TotalOrders,
po.DishID,
rn = ROW_NUMBER() OVER (PARTITION BY v.range
ORDER BY COUNT(DishID) * SUM(OrderQty) DESC)
FROM
( SELECT PgrId,
CASE WHEN p.PgrDOB < DATEADD(YEAR, -40, GETDATE()) THEN '40+'
WHEN p.PgrDOB < DATEADD(YEAR, -30, GETDATE()) THEN '30-39'
WHEN p.PgrDOB < DATEADD(YEAR, -20, GETDATE()) THEN '20-29'
WHEN p.PgrDOB < DATEADD(YEAR, -10, GETDATE()) THEN '10-19'
ELSE '0-9'
END AS range
FROM dbo.Passenger p
) v inner join dbo.PassengerOrder po on v.PgrId = po.PgrID
GROUP BY v.range,po.DishID
)
SELECT range, occurrences, TotalOrders, DishID
FROM cte
WHERE rn = 1
ORDER BY TotalOrders DESC;
Output (shown in this db<>fiddle):
range
occurrences
TotalOrders
DishID
40+
2
220
dsh0000001
30-39
2
84
dsh0000001
10-19
1
11
dsh0000001
20-29
1
1
dsh0000001
0-9
1
1
dsh0000001
I fixed a couple of other things, too:
Always use schema name for objects.
Don't use 'single quotes' to delimit column aliases; this makes them too easy to confuse with string literals. Use [square brackets] instead, but don't delimit at all unless you need to. (I talk a little about that here.)
Your DATEDIFF calculation was not accurate to a person's birthday, nor is mine, but if you flip the order you can make a much less complex CASE expression, and when you move functions away from the column, you make it more likely that the query could benefit from a current (or future) index. Lots more at Dating Responsibly.