Distributing Records Evenly From One Table to Another - sql

Question

I have 3 tables:
Users
-----
UserID (varchar)
Active (bit)
Refunds_Upload
--------------
BorrowerNumber (varchar)
Refunds
-------
BorrowerNumber
UserID
I first select all of the UserID values where Active = 1.
I need to insert the records from Refunds_Upload to Refunds but I need to insert the same (or as close as possible) number of records for each Active UserID.
For example, if Refunds_Upload has 20 records and the Users table has 5 people where Active = 1, then I would need to insert 4 records per UserID into table Refunds.
End Result would be:
BorrowerNumber UserID
105 Fred
110 Fred
111 Fred
115 Fred
120 Billy
122 Billy
123 Billy
125 Billy
130 Lucius
131 Lucius
133 Lucius
135 Lucius
138 Lucy
139 Lucy
140 Lucy
141 Lucy
142 Grady
143 Grady
144 Grady
145 Grady
Of course, it won't always come to an even number of records per User so I need to account for that as well.

Answer 1

First run this and check it returns something like what you want to insert, before you uncomment the insert and actually carry it out..
--INSERT INTO Refunds
SELECT
numbered_u.UserID,
numbered_ru.BorrowerNumber
FROM
(SELECT u.*, ROW_NUMBER() OVER(ORDER BY UserID) - 1 as rown, SUM(CAST(Active as INT)) OVER() as count_users FROM Users u WHERE active=1) numbered_u
INNER JOIN
(SELECT ru.*, ROW_NUMBER() OVER(ORDER BY BorrowerNumber) - 1 as rown, COUNT(*) OVER() as count_ru FROM Refund_Uploads ru) numbered_ru
ON
ROUND(CAST(numbered_ru.rown AS FLOAT) / (count_ru / count_users)) = numbered_u.rown
The logic:
We number every interesting (active=1) row in users and we also count them all. This should return us all 5 users, numbered 0 to 4 and with a ctr that is 5 on each row.
Then we join them to a similarly numbered list of Refund_Uploads (say 20). Similarly, those rows will be numbered 0 to 19 for mathematical reasons that become apparent later. We also count all these rows too
And we then join these two datasets together but the condition is a range of values rather than exact values. The logic is "refund_upload row number, divided by the_count_of_rows_there_should_be_per_user" (i.e. 0..19 / (20/5) ) = user_row_number. Hopefully thus refund rows 0 to 3, associate with user 0, refund rows 4 thru 7 associate with user 1.. etc
It's a little hard to debug without full data - I feel it might need a few +1 / -1 tweaks here and there.
I originally used FLOOR but switched to using ROUND, as I think this might work for distributing sets of numbers where there isn't a whole number of divisions in Refund/User e.g. your 240/13 example.. Hopefully some users will have 18 rows and some 19