Question

Table foobar is, for clarity, structured and has data as follows:

id, action_dt, status_id
1, '02-JUL-10', 'x'
1, '02-JUL-10', '2'
1, '02-JUL-10', NULL
2, '02-JUL-10', 'a'
2, '02-JUL-10', 'b'
3, '02-JUL-10', 'k'
3, '02-JUL-10', NULL
3, '03-JUL-10', 'k'
3, '03-JUL-10', NULL

I need a query that gets IDs such that for each ID a NULL value and a NOT NULL value exists per day. So, in the example dataset above, the query needs to return:

'02-JUL-10', 1
'02-JUL-10', 3
'03-JUL-10', 3

Yes, it can be done using something like:

SELECT
    nulls.action_dt
    , nulls.id 

FROM        (SELECT 
                action_dt
                , id 
            FROM        foobar 
            WHERE       status_id IS NULL
            GROUP BY    action_dt)   nulls

INNER JOIN (SELECT
                action_dt
                , id
            FROM        foobar 
            WHERE       status_id IS NOT NULL
            GROUP BY    action_dt)    non_nulls     ON nulls.action_dt = non_nulls.action_dt 
                                                        AND nulls.id = non_nulls.id

but as you can see, among other things, two subqueries and another iteration for the join…

The query I’ve been working on and have hopes for is of the form:

SELECT
    action_dt
    , id
FROM
    foobar
GROUP BY
    action_dt
    , id
    , CASE WHEN status_id IS NOT NULL THEN 1 ELSE 0 END
HAVING
    COUNT(prim_card_nb) > 1

but it doesn’t quite return what I need (as you know, the HAVING clause applies to the underlying data that is being queried). Any ideas?

After all this, it seems a solution would be to have the above query in a subquery and filter it down that way, such as:

SELECT
    action_dt
    , id
FROM        (SELECT
                action_dt
                , id
            FROM
                foobar
            GROUP BY
                action_dt
                , id
                , CASE WHEN status_id IS NOT NULL THEN 1 ELSE 0 END
            ) repeat_ids_per_day
GROUP BY
    action_dt
    , id
HAVING
    COUNT(id) > 1

but I feel it can be better…

Answer 1

Your idea is sound: in such a case you don’t need a subquery, an aggregate is sufficient and should be more efficient. This should work:

SQL> SELECT action_dt, id
  2    FROM foobar
  3   GROUP BY action_dt, ID
  4  HAVING COUNT(DISTINCT CASE WHEN status_id IS NULL THEN 1 ELSE 0 END) > 1;

ACTION_DT         ID
--------- ----------
02-JUL-10          1
02-JUL-10          3
03-JUL-10          3