Question

I’m up against a mssql database, having a SQL query like…

SELECT id, type, start, stop, one, two, three, four
FROM a
UNION ALL
SELECT id, type, start, stop, one, two, three, four
FROM b
UNION ALL
SELECT id, type, start, stop, one, two, three, four
FROM c
ORDER BY type ASC

Resulting in…

row |  id  type  start       stop         one   two    three   four
----+--------------------------------------------------------------
 1  |  1   a     2010-01-01  2010-01-31   100   1000   1000    100
 2  |  1   a     2010-02-01  2010-12-31   100   500    500     50
 3  |  1   b     2010-01-01  2010-01-31   100   NULL   NULL    100
 4  |  1   b     2010-01-01  2010-12-31   100   NULL   NULL    100
 5  |  1   c     2010-01-01  2010-01-31   0     NULL   NULL    100
 6  |  1   c     2010-01-01  2010-12-31   0     NULL   NULL    100

However, I would much rather prefer the following outcome…

row |  id  type  start       stop         one   two    three   four
----+--------------------------------------------------------------
 1  |  1   a     2010-01-01  2010-01-31   100   1000   1000    100
 2  |  1   a     2010-02-01  2010-12-31   100   500    500     50
 4  |  1   b     2010-01-01  2010-12-31   100   NULL   NULL    100
 6  |  1   c     2010-01-01  2010-12-31   0     NULL   NULL    100

That is, eliminating row 3 and 5, since they are dupes to row 4 and 6 in every way but the stop-column, AND whereas the unfortunate row having to lowest value in the excluding stop-column is to be removed.

How can I accomplish this?
I’ve been thinking something like…

SELECT * FROM (
    SELECT id, type, start, stop, one, two, three, four
    FROM a
    UNION ALL
    SELECT id, type, start, stop, one, two, three, four
    FROM b
    UNION ALL
    SELECT id, type, start, stop, one, two, three, four
    FROM c
    ORDER BY type ASC
) AS types
GROUP BY ... HAVING ???

I need guidance, please help.

(And no, I’m in no position to change any conditions, I’ve got to work with the given situation.)

Answer 1

Similar questions have been asked and answered. For example: Select uniques, and one of the doubles

And your situation is even simpler (if I understood your problem description correctly):

select id, type, start, max(stop), one, two, three, four
    from (...) types
    group by id, type, start, one, two, three, four
    order by ...

In place of (...) you put your selects from a, b and c.
Just leave out order by clause.

Or, if instead of (id, type, start)->(one, two, three, four) you have (id, type, start, stop)->(one, two, three, four) (meaning you have to chose other columns that correspond
to max(stop)), this query usually results in sensible execution plan:

select id, type, start, stop, one, two, three, four
    from (...) types
    where stop = (select max(stop)
                  from (...) t2
                  where t2.id = types.id
                        and t2.type = types.type
                        and t2.start = types.start)

but it depends on how data is distributed among your source tables and what indexes are present. In some cases solutions from link above might still be better.