Question

I have table with statistical data.

I’m trying to count the records grouped by a version, the problem is that the records have another criteria (ref) and should counted only once (per ref).

stats sample data to illustrate the issue:

id stat_date  ref version
-------------------------
1  2012-01-25   1  A
2  2012-01-25   2  B
3  2012-01-25   3  A
4  2012-01-26   8  B
5  2012-01-26   2  B
6  2012-01-26   3  B  <-- version has been updated for ref=3

Simple counting would return

SELECT COUNT(*),version FROM stat GROUP BY version

1,A
5,B

The problem here is that only the last record with ref=3 (id=6) should be counted and (id=3) has to be ignored.
So the question is, how can I filter row (id=3) from the query?

I can’t figure out what I should insert as condition in the subquery

SELECT COUNT(*),version FROM stats
WHERE  stat_date BETWEEN "2012-01-25"  AND "2012-01-26" 
AND id = (SELECT MAX(id) FROM stats WHERE <condition>)
GROUP BY 2

The expected result would be:

1,A (since id=3 is ignored)
3,B (since the first id=2 is ignored and only id=5 is taken into account)

Answer 1

Try:

SELECT COUNT(*),version 
FROM stats s1
WHERE stat_date BETWEEN "2012-01-25"  AND "2012-01-26" AND
      not exists (SELECT null 
                  FROM stats s2 
                  WHERE s1.ref = s2.ref and s2.id>s1.id)
GROUP BY version

EDIT: Not guaranteed to always work:

SELECT COUNT(*),version from
(select * from
 (select * from stats 
  where stat_date BETWEEN "2012-01-25"  AND "2012-01-26" 
  order by ref, id desc) s0
 group by ref) s1
group by version