Question

I have a SQL Server 2008 database where I would like to be able to do the following on one table.

The table has multiple columns that should be unique based on a combination of two columns.
We will call them [ID1] and [ID2], then there is a key we will call [index] and a value that could be duplicated called [ID3] and a datetime value called [start].

So here is the dilemma, within the scope of the table there are should only be one increasing value of [index] for every [ID1 ]and [ID2] combination, the three form a natural PK from the client DB that is being warehoused in one consolidated server database.
ID3 is indicative of a value that was used to determine when a row was stored in the client DB so there may be duplicates in the server

[ID1]   [ID2]   [index]     [ID3]   [start]         [other1]    [other2]
1   1   1       1   01/01/2000 01:00:00 5       6
1   1   2       2   01/01/2000 01:00:01 4       2
1   1   3       3   01/01/2000 01:00:02 5       2
1   1   4       3   01/01/2000 01:00:03 5       2
1   1   5       4   01/01/2000 01:00:04 4       6

What I want is a query that will return rows that are unique combinations of [ID3] and the [other1] & [other2] columns, unique to the [ID1], [ID2] key, with that I would like the first [start] that fits that criteria, essentially ignoring further occurrences of the same distinct clause.

From the above table it would return …

[ID1]   [ID2]   [index]     [ID3]   [start]         [other1]    [other2]
1   1   1       1   01/01/2000 01:00:00 5       6
1   1   2       2   01/01/2000 01:00:01 4       2
1   1   3       3   01/01/2000 01:00:02 5       2
1   1   5       4   01/01/2000 01:00:04 4       6

The second row with the [ID3] of the value 3 would be ignored, as would any other that had duplicates of [ID3]

The point I cannot seem to get to is the first value of each distinct combination, because the distinct does not allow me to select the value of the other columns, a group by would require some aggregate function.

Answer 1

;with a as
(
SELECT [ID1],[ID2],[index],[ID3],[start],[other1],[other2], 
rn = row_number() over (partition by ID1, ID2, ID3 order by start)
FROM yourtable
)
select 
[ID1],[ID2],[index],[ID3],[start],[other1],[other2], 
from a
where rn = 1