SQL join on table twice not bringing expected results - sql

Question

I have two tables (extraneous columns removed to exemplify the issue):
-People-
PID | CarID1 | CarID2
----------------------
1 | 1 | 3
2 | 5 | NULL
3 | 1 | NULL
4 | NULL | 1
-Cars-
CarID
-----
1
3
5
I'm creating a view based on the CarID so using:
SELECT
c.CarID,
COUNT(p.PID) AS pCount
FROM
Cars c
LEFT JOIN People p ON p.CarID1 = c.CarID OR p.CarID2 = c.CarID
Group By c.CarID
Brings back the expected results:
CarID | pCount
--------------
1 | 3
3 | 1
5 | 1
The issue being that on a table with 1000+ car id's and 25,000 people, this can take a long time (taking out the OR clause means it takes milliseconds)
So I was trying to do it another way like this:
SELECT
c.CarID,
COUNT(p1.PID) AS pCount1,
COUNT(p2.PID) AS pCount2
FROM
Cars c
LEFT JOIN People p1 ON p1.CarID1 = c.CarID
LEFT JOIN People p2 ON p2.CarID2 = c.CarID
Group By c.CarID
It's many times quicker, but because CarID 1 exists in both CarID1 and CarID2 I'm getting this:
CarID | pCount1 | pCount2
-------------------------
1 | 3 | 3
3 | 0 | 1
5 | 1 | 0
When I would expect this:
CarID | pCount1 | pCount2
-------------------------
1 | 2 | 1
3 | 0 | 1
5 | 1 | 0
And I could just sum the pCount1 and pCount2
Is there any way I can achieve the results of the first query using the 2nd method? I'm presuming the GROUP BY clause has something to do with it, but not sure how to omit it.

Answer 1

How about unpivoting the columns and then joining:
SELECT v.CarID, COUNT(p.PID) AS pCount
FROM People p CROSS APPLY
(VALUES (p.CarID1), (p.CarID2)) v(CarID) JOIN
Cars c
ON v.CarID = c.CarId
WHERE v.CarID IS NOT NULL
GROUP BY v.CarID;
If you want to keep cars even with no people, then you can express this as a LEFT JOIN:
SELECT c.CarID, COUNT(p.PID) AS pCount
FROM Cars c LEFT JOIN
(People p CROSS APPLY
(VALUES (p.CarID1), (p.CarID2)) v(CarID)
)
ON v.CarID = c.CarId
GROUP BY c.CarID;
Here is a db<>fiddle.

Answer 2

Is the p.CarID1 a Primary Key?
If so it would explain that a join on the carID1 is fast but on the carID2 it's slow.
Try creating an Index on CarID2 and see if that solves your performance issues.
The index would turn it from a full table scan into an index lookup. Which is a lot faster.
CREATE NONCLUSTERED INDEX CarId2Index
ON p.CarID2;
If that solves it you can keep your query as it is.
Alternatively you can send us the query explain plan so we can see what is slowing it down.

Answer 3

Try using SUM with condition like below.
SELECT
c.CarID,
SUM(IIF(p1.PID IS NULL, 0, 1)) AS pCount1,
SUM(IIF(p2.PID IS NULL, 0, 1)) AS pCount2
FROM
Cars c
LEFT JOIN People p1 ON p1.CarID1 = c.CarID
LEFT JOIN People p2 ON p2.CarID2 = c.CarID
Group By c.CarID

Answer 4

Try with COALESCE function:
SELECT
c.CarID,
COUNT(p.PID) AS pCount
FROM
Cars c
LEFT JOIN People p ON COALESCE(p.CarID1, p.CarID2) = c.CarID
Group By c.CarID