Question

I have the following table in my postgreSQL 8.3.14 database

     timestamp           status
2012-03-12 19:15:01        f
2012-03-12 19:15:02        f
2012-03-12 19:15:05        f
2012-03-12 19:17:01        t
-- END OF SLOT ONE (change from f to t)
2012-03-12 19:20:01        f
2012-03-12 19:25:01        f
2012-03-12 19:27:01        f
2012-03-12 20:15:01        t
-- END OF SLOT TWO (change from f to t)

No I want to get the following result:

• slot 1 duration (2012-03-12 19:17:01 – 2012-03-12 19:15:01)
• slot 2 duration (2012-03-12 20:15:01 – 2012-03-12 19:20:01)

So every time the status value changes from f to t, a new slot begins. Now I want to get all slots and the duration of every slot. The timestamp is not every second or minute or so, the difference between two entries is more or less random.

I thought about subqueries and grouping, but I have no idea how to solve this, at least whether it is possible at all. So my question is: is that possible 🙂 And if it is, how to start? I’m stuck because of the f-t changing issue…

Answer 1

If you have a modern version of PG you can do what you want using window functions (in my example instead of timestamps I have integers, but it doesn’t matter):
I’m working on a non-window function version…

db=> select * from test1;
 time | status
------+--------
1 | f
2 | f
3 | f
4 | t
5 | f
6 | f
7 | f
8 | t
9 | f
10 | t

(8 rows)

Here is the query with window functions:

 WITH a AS (
 SELECT time,rank() OVER (ORDER BY time) 
      FROM (
               SELECT status, time, lag(status,1)  OVER (ORDER BY time) AS ls
            FROM test1 
            ) AS x
             WHERE status AND NOT ls 
            -- select where the current status is true, the previous is false
       ),  -- a is now the endings of the slots
 b AS  (
     SELECT time, rank() OVER (ORDER BY time)
       FROM (
             SELECT  status, time, LAG(status,1)         
               OVER  (ORDER BY time)  AS ls
             FROM test1 
            ) AS x 
            WHERE (NOT status AND ls) OR (NOT status AND ls IS NULL) 
        )   -- b is now the beginning of the slots
    SELECT b.time as time1, a.time as time2 FROM a,b WHERE a.rank=b.rank;

  time1 | time2
------+------
1 |    4
5 |    8
9 |   10

Here is the version of the query for PG without window functions (it is actually shorter, but less clear)

  SELECT min(time) as time1, time2 FROM 
     (
         SELECT time, status,
          (
           SELECT time FROM test1 AS y
               WHERE y.time >= x.time
               ORDER BY (x.status = y.status)::int, y.time ASC LIMIT 1
          ) AS time2 
        FROM test1 AS x WHERE NOT status
     ) AS y GROUP BY y.time2 ORDER BY time2;