Question

Assume (highly anonymized):

Create Table myTable
(
ID INT PK,
INDEXNUMBER INT,
VERSION INT,
Data VARCHAR(MAX)
)

This table is used to store mutually exclusive data.
For example:

100 1 1 BOB
217 1 2 JOHN
319 1 3 GEORGE
420 7 1 MARY
415 7 2 SUSAN

In this case, I need to randomly pick ONE of BOB, JOHN or GEORGE and ONE of MARY or SUSAN.

I’m happy with either the ID or the INDEXNUMBER/VERSION pair.

If it helps to think about it, it’s like picking a single shift of a hockey team from a table containing a roster:

Pick 1 Center from 3 available,
Pick 1 Left Wing from 5 available, etc.

I’ve been playing with NEWID() and MAX/MIN (Cast NEWID to varchar first) but I keep getting hung up on the GROUP BY. If I GROUP BY ID, then max is operating on a single row at a time, yielding the entire table.

If I GROUP BY INDEXNUMBER, VERSION I get a similar result (The pair being unique).

What I need to do is GROUP BY INDEXNUMBER (excluding ID from the query entirely) yet somehow retrieve the VERSION.

Answer 1

Partition by the INDEXNUMBER (I’m assuming that you need one from each though it’s not specifically stated) and order by NEWID()

SELECT ID
FROM (
    SELECT 
        Row_Number() OVER (PARTITION BY INDEXNUMBER ORDER BY NEWID()) Sort,
        ID, 
        Data
    FROM myTable
) s
WHERE Sort = 1

Sql Fiddle