SQL server pivot count function works incorrectly if used with inner query - sql

Question

I think there is a bug in my SQL-server pivot function because I can't find any other explanation.
I am running a simple pivot and using the exact syntax shown in MSDN.
but the pivot answer shows the exact same number for all columns with the value equivalence of the count of all weeks!!!
(left is query result and right is what I want)
here is my query:
SELECT
*
FROM (SELECT
r.cutomer_id
,c.[Week]
,r.id
FROM r
JOIN c
ON r.Create_date = c.Date
WHERE Is_ride = 1
AND ((Create_date_int BETWEEN 20190302 AND 20190319)
OR (Create_date_int BETWEEN 20190406 AND 20190426))) p
PIVOT
(
COUNT(id)
FOR [Week] IN
([9], [10], [11], [12], [14], [15], [16], [17])
) AS pvt
here are some test data representing the "p" output (i just changed id numbers, weeks are the same number you get from the query)
DROP TABLE IF EXISTS #t
CREATE TABLE #t (
customer_id INT
,WEEK INT
,id INT
)
INSERT #t (customer_id, WEEK, id)
VALUES (12032, 10, 8607)
, (43551, 10, 8721051)
, (55025, 10, 81200)
, (198874, 10, 861362)
, (99675, 10, 867081)
, (19387, 10, 863656)
, (12526, 10, 8603706)
, (19503, 10, 860924)
, (37597, 10, 860909)
, (136019, 10, 8610674);
so I thought there is something wrong with my query but then I changed the query to this:
SELECT
r.cutomer_id
,c.[Week]
,r.id INTO #t
FROM r
JOIN c
ON r.Create_date = c.Date
WHERE Is_ride = 1
AND ((Create_date_int BETWEEN 20190302 AND 20190319)
OR (Create_date_int BETWEEN 20190406 AND 20190426))
SELECT
*
FROM #t
PIVOT
(
COUNT(id)
FOR [Week] IN
([9], [10], [11], [12], [14], [15], [16], [17])
) AS pvt
and it works just fine!
also if I remove the r.id column from select and change it to count(week) it works fine!!!
also if I only change my where to
WHERE Is_ride = 1
AND ((Create_date_int BETWEEN 20190302 AND 20190319)
OR (Create_date_int BETWEEN 20190406 AND 20190426))
and passenger_id in (43551,12032,136019)
) p
it works fine!!!
can someone give me an explanation?

Answer 1

I have no idea why you're getting the result shown on your images. There's possibly some information missing.
Here's an MVCE with code commented to create a larger set of test data.
CREATE TABLE r(
id int identity,
customer_id int,
create_date date,
Create_date_int AS CONVERT( int, CONVERT( char(8), create_date, 112)),
is_ride bit)
INSERT INTO r(customer_id, create_date, is_ride)
SELECT customer_id, '20190307', 1
FROM (VALUES(1),(2),(3),(4),(5),(6),(7),(8),(9),(10))x(customer_id)
--SELECT TOP 100000
-- ABS(CHECKSUM(NEWID())) % 10,
-- DATEADD( dd, ABS(CHECKSUM(NEWID())) % 120, '2019'),
-- 1
--FROM sys.all_columns a, sys.all_columns b;
CREATE TABLE c(
[Date] date,
[Week] AS DATEPART( wk, [Date])
)
INSERT INTO c([Date])
SELECT DATEADD( dd, ROW_NUMBER() OVER( ORDER BY (SELECT NULL))-1, '2019')
FROM sys.all_columns;
From here, I could run your original query that gives the expected results, just 1's in week 10.
I can also run an alternative query using cross tabs that is my preferred method to pivot data.
SELECT
r.customer_id
,COUNT( CASE WHEN c.[Week] = 9 THEN r.id END) AS wk9
,COUNT( CASE WHEN c.[Week] = 10 THEN r.id END) AS wk10
,COUNT( CASE WHEN c.[Week] = 11 THEN r.id END) AS wk11
,COUNT( CASE WHEN c.[Week] = 12 THEN r.id END) AS wk12
,COUNT( CASE WHEN c.[Week] = 14 THEN r.id END) AS wk14
,COUNT( CASE WHEN c.[Week] = 15 THEN r.id END) AS wk15
,COUNT( CASE WHEN c.[Week] = 16 THEN r.id END) AS wk16
,COUNT( CASE WHEN c.[Week] = 17 THEN r.id END) AS wk17
FROM r
JOIN c ON r.Create_date = c.Date
WHERE Is_ride = 1
AND (Create_date_int BETWEEN 20190302 AND 20190319
OR Create_date_int BETWEEN 20190406 AND 20190426)
GROUP BY r.customer_id
ORDER BY r.customer_id;
That also gives the correct information.
Both of these methods might benefit from preaggregation to improve their performance.
WITH ctePreAggregate AS(
SELECT
r.customer_id
,c.[Week]
,COUNT(*) AS weeklycount
FROM r
JOIN c
ON r.Create_date = c.Date
WHERE Is_ride = 1
AND (Create_date_int BETWEEN 20190302 AND 20190319
OR Create_date_int BETWEEN 20190406 AND 20190426)
GROUP BY r.customer_id
,c.[Week]
)
SELECT
*
FROM ctePreAggregate
PIVOT
(
SUM(weeklycount)
FOR [Week] IN
([9], [10], [11], [12], [14], [15], [16], [17])
) AS pvt
ORDER BY customer_id;
WITH ctePreAggregate AS(
SELECT
r.customer_id
,c.[Week]
,COUNT(*) AS weeklycount
FROM r
JOIN c
ON r.Create_date = c.Date
WHERE Is_ride = 1
AND (Create_date_int BETWEEN 20190302 AND 20190319
OR Create_date_int BETWEEN 20190406 AND 20190426)
GROUP BY r.customer_id
,c.[Week]
)
SELECT
customer_id
,SUM( CASE WHEN [Week] = 9 THEN weeklycount ELSE 0 END) AS wk9
,SUM( CASE WHEN [Week] = 10 THEN weeklycount ELSE 0 END) AS wk10
,SUM( CASE WHEN [Week] = 11 THEN weeklycount ELSE 0 END) AS wk11
,SUM( CASE WHEN [Week] = 12 THEN weeklycount ELSE 0 END) AS wk12
,SUM( CASE WHEN [Week] = 14 THEN weeklycount ELSE 0 END) AS wk14
,SUM( CASE WHEN [Week] = 15 THEN weeklycount ELSE 0 END) AS wk15
,SUM( CASE WHEN [Week] = 16 THEN weeklycount ELSE 0 END) AS wk16
,SUM( CASE WHEN [Week] = 17 THEN weeklycount ELSE 0 END) AS wk17
FROM ctePreAggregate
GROUP BY customer_id
ORDER BY customer_id;
The only issue is that you would need to add a series of ISNULL() in the column list to show NULLs instead of zeros for the PIVOT query.