Question

I’m having an issue with a query, I’m using postgresql 9.2.

I have created a view which returns a key value, a timestamp, and an interval that look like this:

select * from foo.a;  -- My view and the results

 k  |         ts1            |  tm
 1  | 2011-12-08 06:46:00+00 | 01:00:00
 1  | 2011-12-08 09:14:00+00 | 01:00:00
 2  | 2011-12-08 02:00:00+00 | 02:00:00
 2  | 2011-12-08 05:00:00+00 | 02:00:00
 4  | 2011-12-08 10:06:00+00 | 02:00:00
 4  | 2011-12-08 13:31:00+00 | 02:00:00
 1  | 2011-12-11 03:00:00+00 | 01:00:00
 1  | 2011-12-11 04:49:00+00 | 02:00:00
 4  | 2011-12-11 10:06:00+00 | 02:00:00
 4  | 2011-12-11 13:31:00+00 | 02:00:00

I have about 8,000 rows in my results. Now, I want to sum the interval’s and group by key, so I do this:

select k,
       sum(tm) as tm
  from foo.a
group by k;

As expected my results are:

k   |  tm
1   | 5:00
2   | 4:00
4   | 8:00

I want to take it a step further and do this (the error – ERROR: subquery uses ungrouped column ‘ts1’ from outer query).

select k,
       sum(tm) as tm,
       (select sum(tm) where ts1 > '2011-12-09')
        -- (select sum(tm) where ts1 > '2012-01-01')
        -- I need to the above on 5 different dates. 
  from foo.a
  group by k,

Desired results are:

k    |   tm   |  tm1
1    |  5:00  |  3:00
2    |  4:00  |  
4    |  8:00  |  4:00

If I add ts1 into my group clause, it doesn’t return desired results, naturally – because I don’t want to group by ts1, only k. I can use ts1 in an aggregate function and the query will run with undesirable results.

This works:

select foo.a.k,
       bar.tm1,
       sum(tm)
  from foo.a
  left join
      (select k,
              sum(tm) as tm1
         from foo.a
        where foo.a.ts1 > '2011-12-09'
        group by k) as bar
  on foo.a.k = bar.k
  group by foo.a.k,
           bar.tm1;

But querying the same view for the same information seems wrong and unnecessary. Especially, since I have to do this 5 times.

My question: Is there a work-around for performing a correlated-query on an ungrouped column? If not, what is a better approach to achieving my desired results?

Answer 1

Is this what you want?

select k,
       sum(tm) as tm,
       sum(case when ts1 > '2012-01-01' then tm else 0 end) as tm_20120101
from foo.a
group by k