Count Distinct with Subquery throws error - sql-server

Question

I have an sql query where I count receipts. I want to count a) all receipts, b) receipts where the related customer has registered in the same year/month, c) receipts where the related customer has visited the store at least 4 times. The query is grouped over year and month. I hope it is comprehensible.
SELECT t.Year, t.Month,
COUNT(DISTINCT fs.Receipt) AS NumberOfGuests, //fs.Receipt is Receipt Number
COUNT(DISTINCT
(CASE
WHEN YEAR(c.RegistrationDate) = t.Year AND MONTH(c.RegistrationDate) = t.Month
THEN fs.Receipt END)) AS NumberOfNewGuests,
COUNT(DISTINCT
(CASE WHEN
(
SELECT COUNT(DISTINCT fs_sub.Receipt)
FROM Dimension_Time AS t_sub
LEFT OUTER JOIN Fact_Sales AS fs_sub ON t_sub.ID = fs_sub.Time AND fs_sub.Store = #storeID
WHERE t_sub.Time = t.Time AND fs_sub.CustomerID = fs.CustomerID
) > 3
THEN fs.Receipt END)) AS NumberOfRegularGuests
FROM Dimension_Time AS t
LEFT OUTER JOIN Fact_Sales AS fs ON t.ID = fs.Time AND fs.Store = #storeID
LEFT OUTER JOIN Dimension_Customer AS c ON fs.Customer = c.ID
WHERE (t.Time >= DATEFROMPARTS(2021 - 1, 5, 1)) AND (t.Time <= EOMONTH(DATEFROMPARTS(2021, 5, 1)))
GROUP BY t.Month, t.Year
ORDER BY t.Year, t.Month
NumberOfGuests and NumberOfNewGuests works fine, but with NumberOfRegularGuests, I get an error:
Cannot perform an aggregate function on an expression containing an aggregate or a subquery.
Is it possible to build the query another way to avoid the error?

Answer 1

You can put the subquery into an APPLY:
SELECT t.Year, t.Month,
COUNT(DISTINCT fs.Receipt) AS NumberOfGuests, //fs.Receipt is Receipt Number
COUNT(DISTINCT
(CASE
WHEN YEAR(c.RegistrationDate) = t.Year AND MONTH(c.RegistrationDate) = t.Month
THEN fs.Receipt END)) AS NumberOfNewGuests,
COUNT(DISTINCT
(CASE WHEN t2.countReceipts > 3 THEN fs.Receipt END)
) AS NumberOfRegularGuests
FROM Dimension_Time AS t
LEFT OUTER JOIN Fact_Sales AS fs ON t.ID = fs.Time AND fs.Store = #storeID
LEFT OUTER JOIN Dimension_Customer AS c ON fs.Customer = c.ID
OUTER APPLY (
SELECT COUNT(DISTINCT fs_sub.Receipt) AS countReceipts
FROM Dimension_Time AS t_sub
LEFT OUTER JOIN Fact_Sales AS fs_sub ON t_sub.ID = fs_sub.Time AND fs_sub.Store = #storeID
WHERE t_sub.Time = t.Time AND fs_sub.CustomerID = fs.CustomerID
) t2
WHERE (t.Time >= DATEFROMPARTS(2021 - 1, 5, 1)) AND (t.Time <= EOMONTH(DATEFROMPARTS(2021, 5, 1)))
GROUP BY t.Month, t.Year
ORDER BY t.Year, t.Month;
But a subquery of this type is usually pretty inefficient (although this can depend on indexing as well as cardinalities ie how much of Fact_Sales is actually being used after filtering on dates). It can be replace by a windowed aggregate.
SELECT t.Year, t.Month,
COUNT(DISTINCT fs.Receipt) AS NumberOfGuests, //fs.Receipt is Receipt Number
COUNT(DISTINCT
(CASE
WHEN YEAR(c.RegistrationDate) = t.Year AND MONTH(c.RegistrationDate) = t.Month
THEN fs.Receipt END)) AS NumberOfNewGuests,
COUNT(DISTINCT
(CASE WHEN fs.countReceipts > 3 THEN fs.Receipt END)
) AS NumberOfRegularGuests
FROM Dimension_Time AS t
LEFT OUTER JOIN (
SELECT *,
COUNT(*) OVER (PARTITION BY fs.CustomerID) AS countReceipts
FROM Fact_Sales
) AS fs ON t.ID = fs.Time AND fs.Store = #storeID
LEFT OUTER JOIN Dimension_Customer AS c ON fs.Customer = c.ID
WHERE (t.Time >= DATEFROMPARTS(2021 - 1, 5, 1)) AND (t.Time <= EOMONTH(DATEFROMPARTS(2021, 5, 1)))
GROUP BY t.Month, t.Year
ORDER BY t.Year, t.Month;
It's unclear why you have COUNT(DISTINCT as opposed to just COUNT, this is usually an indication of a poorly thought-out join.
If it is indeed the case that there are multiple Fact_Sales with the same Receipt then you would need to change the window aggregate, as you cannot use COUNT(DISTINCT...) OVER...
SELECT t.Year, t.Month,
COUNT(DISTINCT fs.Receipt) AS NumberOfGuests, //fs.Receipt is Receipt Number
COUNT(DISTINCT
(CASE
WHEN YEAR(c.RegistrationDate) = t.Year AND MONTH(c.RegistrationDate) = t.Month
THEN fs.Receipt END)) AS NumberOfNewGuests,
COUNT(DISTINCT
(CASE WHEN fs.countReceipts > 3 THEN fs.Receipt END)
) AS NumberOfRegularGuests
FROM Dimension_Time AS t
LEFT OUTER JOIN (
SELECT *,
MAX(rn) OVER (PARTITION BY fs.CustomerID) AS countReceipts
FROM (
SELECT *,
DENSE_RANK() OVER (PARTITION BY fs.CustomerID ORDER BY fs.Receipt) AS rn
FROM Fact_Sales
) AS fs
) AS fs ON t.ID = fs.Time AND fs.Store = #storeID
LEFT OUTER JOIN Dimension_Customer AS c ON fs.Customer = c.ID
WHERE (t.Time >= DATEFROMPARTS(2021 - 1, 5, 1)) AND (t.Time <= EOMONTH(DATEFROMPARTS(2021, 5, 1)))
GROUP BY t.Month, t.Year
ORDER BY t.Year, t.Month;