Checking for X continuous periods for a certain item

Question

Hi,
simplified, I have a table like this:
ROWNUM VEHICULE_ID FROM(DD/MM/YYYY) TILL(DD/MM/YYYY)
==============================================
1 1 10/09/2009 02/02/2010
2 2 01/01/2010 31/12/2010
3 1 01/01/2010 31/12/2010
4 3 01/10/2009 31/03/2010
5 1 01/08/2009 01/03/2010
6 1 01/02/2010 01/09/2010
7 1 01/03/2010 30/06/2010
Now, I have to determine all the vehicules that have at least 3 continuous periods between 01/01/2010 and 30/06/2010.
When I apply this question to the above data, then it should return vehicule_id 1.
The 3 continuous periods for vehicule_id 1 are:
A)
rownum 3: 01/01/2010 - 31/12/2010 ==> 01/01/2010 - 30/06/2010: OK
B)
rownum 5: 01/08/2009 - 01/03/2010 ==> 01/01/2010 - 01/03/2010: OK
+ rownum 6: 01/02/2010 - 01/09/2010 ==> 02/03/2010 - 30/06/2010: OK
C)
rownum 1: 10/09/2009 - 02/02/2010 ==> 01/01/2010 - 02/02/2010
+ rownum 7: 01/03/2010 - 30/06/2010 ==> 01/03/2010 - 30/06/2010: OK
+ rownum 6: 01/02/2010 - 01/09/2010 ==> 03/02/2010 - 28/02/2010: OK ==> This period was not yet used in the periods for B), so we could still use it here.
The best solution I can find for this is to create a cursor and check if for every date between 01/01/2010 and 30/06/2010 at least 3 diffent periods are valid, but this not very fast and generic if the period should change.
Does anybody have an idea how you could get this information via sql or pl/sql?
Thanks in advance for any help about this,
Geert

Answer 1

Hi, Geert,
You can use analytic functions to find how many registrations were active at any point in time.
In the query below, s_cnt is the number of registrations active at the beginning of the period of interest (this is the same for every row with the same vehicule_id), and e_cnt is the number of other registrations remaining active at the time each registration ends. (If the number of concurrent registrations ever falls below 3, it will do so at a time when some registration ends, so it is only necessary to test those points.) The vehicule_ids you want are those where e_cnt is at least 3, and e_cnt is never less than 3 during the period of interest.
VARIABLE     period_sdat     VARCHAR2 (10)
VARIABLE     period_edat     VARCHAR2 (10)
VARIABLE     min_cnt          NUMBER
EXEC     :period_sdat := '01/01/2010';
EXEC     :period_edat := '30/06/2010';
EXEC     :min_cnt := 3;
WITH     got_cnt     AS
     SELECT     t.*
     ,     COUNT ( CASE
                    WHEN  sdat <= TO_DATE (:period_sdat, 'DD/MM/YYYY')
                    THEN  1
               END
                )       OVER ( PARTITION BY  vehicule_id)     AS s_cnt
     ,     COUNT (1) OVER ( PARTITION BY  vehicule_id
                     ORDER BY      sdat
                     RANGE BETWEEN UNBOUNDED    PRECEDING
                           AND     edat - sdat  FOLLOWING
           - COUNT (1) OVER ( PARTITION BY  vehicule_id
                      ORDER BY      edat
                      )                    AS e_cnt
     FROM     t
     WHERE     sdat     <= TO_DATE (:period_edat, 'DD/MM/YYYY')
     AND     edat     >= TO_DATE (:period_sdat, 'DD/MM/YYYY')
SELECT       vehicule_id
FROM       got_cnt
WHERE       s_cnt     >= :min_cnt
GROUP BY  vehicule_id
HAVING       SUM ( CASE
               WHEN  e_cnt < :min_cnt
               AND   edat  < TO_DATE (:period_edat, 'DD/MM/YYYY')
               THEN  1
               ELSE  0
          END
           ) = 0
;The tricky part here is computing e_cnt. This is the number of prior registrations begun, minus the number of prior registrations ended. The number of prior registrations ended is a straightforward analytic function, but the number of prior registrations begun requires a windowing clause, because "prior" here means "prior to edat", not sdat, which is used for the ORDER BY.
Edited by: Frank Kulash on Apr 2, 2010 11:22 AM