LEAD with Partition

Question

Hi,
I am stuck on a problem to implement LEAD/LAG with partition.
Below is the example and expected Result
create table trd(
     key number,
     book number,
     prd_key number,
     direction varchar2(2),
     trdtime date,
     price number)
insert into trd values(1234,115,133864,'B','17-07-2013 18:18:00',108.859375);
insert into trd values(1235,115,133864,'S','17-07-2013 18:18:00',108.859375);
insert into trd values(1245,115,133864,'S','17-07-2013 18:18:00',108.859375);
insert into trd values(1236,115,133864,'B','15-07-2013 18:18:00',108.872261598);
insert into trd values(1237,115,133864,'S','15-07-2013 18:18:00',108.866752398);
insert into trd values(1247,115,133864,'S','15-07-2013 18:18:00',108.866752398);
insert into trd values(1238,115,133864,'S','14-07-2013 18:18:00',107.86);
insert into trd values(1239,115,133864,'S','14-07-2013 18:17:00',108.86);
insert into trd values(1240,115,133864,'B','14-07-2013 18:12:00',109.86);
insert into trd values(1241,115,133864,'B','14-07-2013 18:17:00',110.86);
I need to return the value something like this:
Key
Book
Prd_Key
Direction
TrdTime
Price
NextPrice
1234
115
133864
B
7/17/2013 6:18:00 PM
108.859375
108.866752398
1235
115
133864
S
7/17/2013 6:18:00 PM
108.859375
108.872261598
1245
115
133864
S
7/17/2013 6:18:00 PM
108.859375
108.872261598
1236
115
133864
B
7/15/2013 6:18:00 PM
108.872261598
108.86
1237
115
133864
S
7/15/2013 6:18:00 PM
108.866752398
110.86
1247
115
133864
S
7/15/2013 6:18:00 PM
108.866752398
110.86
1238
115
133864
S
7/14/2013 6:18:00 PM
107.86
110.86
1239
115
133864
S
7/14/2013 6:17:00 PM
108.86
109.86
1240
115
133864
B
7/14/2013 6:12:00 PM
109.86
NULL
1241
115
133864
B
7/14/2013 6:17:00 PM
110.86
NULL
The logic to embed is :
For each record, Need to get the OPPOSITE direction's and Existing TrdTime > Other records TrdTime.
For example: for key 1237, the Direction is S and TrdTime is 7/15/2013 6:18:00 PM.
There are following records returned for this record:
1240 and 1241 both having opposite side ; 'B' and existing record TrdTime > these two records.
But the TrdTime of 1241 is selected since it is ordered by nearest and highest TrdTime.
How Can I implement this functionality.
I was thinking of doing it using LEAD function and partition.
I cannot use Cursors since tables are not indexed and there are over 5 mil records.
Any suggestions please.

Answer 1

Hi,
Here's a way to do it using the analytic LAST_VALUE function.
Let me know how this compares to the previous solution on your real data.
SELECT    t.*
,         CASE
              WHEN  direction = 'B'
              THEN  LAST_VALUE ( CASE
                                     WHEN  direction = 'S'
                                     THEN  price
                                END
                                IGNORE NULLS
                               ) OVER ( PARTITION BY  book
                                        ,             prd_key
                                        ORDER BY      trdtime
                                        RANGE BETWEEN UNBOUNDED PRECEDING
                                              AND      1E-6 PRECEDING
              WHEN  direction = 'S'
              THEN  LAST_VALUE ( CASE
                                     WHEN  direction = 'B'
                                     THEN  price
                                 END
                                 IGNORE NULLS
                               ) OVER ( PARTITION BY  book
                                        ,             prd_key
                                        ORDER BY      trdtime
                                        RANGE BETWEEN UNBOUNDED PRECEDING
                                              AND      1E-6 PRECEDING
          END AS nextprice
FROM      trd  t
--WHERE   ... -- If you want any filtering, put it here
ORDER BY  trdtime  DESC
,        direction
,        key
The magic number 1E-6 is an arbitrary number, less than the granularity of trdtime.  That is, we want to look at trdtimes up to and including 1 second (roughly .000012 days) in the past.  1E-6 is clearly less than that.