A Job for 'PARTION BY' Analytical Function?

Question

Hi,
I'm still a little fuzzy on using partitions but this looks like a possible candidate to me.
I need to count the number of different customers that visit an office in a single day. If a customer visits an office more than once in a single day that counts as 1.
Input
OFFICE CUSTOMER TRAN_DATE
1     11     1-Apr-09
1     11     1-Apr-09
1     11     1-Apr-09
1     11     2-Apr-09
2     22     2-Apr-09
2     22     2-Apr-09
2     33     2-Apr-09
select a.office as "OFFICE", a.customer AS "CUSTOMER", a.tran_date AS "TRAN_DATE", COUNT(*)
FROM
(SELECT 1 AS "OFFICE", 11 AS "CUSTOMER", '01-APR-2009' AS "TRAN_DATE" FROM DUAL
UNION ALL
SELECT 1 , 11 , '01-APR-2009' FROM DUAL
UNION ALL
SELECT 1 , 11 , '01-APR-2009' FROM DUAL
UNION ALL
SELECT 1 , 11 , '02-APR-2009' FROM DUAL
UNION ALL
SELECT 2 , 22 , '02-APR-2009' FROM DUAL
UNION ALL
SELECT 2 , 22 , '02-APR-2009' FROM DUAL
UNION ALL
SELECT 2 , 33 , '02-APR-2009' FROM DUAL
) a;
Desired Result
1     1-Apr-09     1
1     2-Apr-09     1
2     2-Apr-09     2
Is this possible with partitions, do I need to use subqueries, or some other methid?
Thank You in Advance for Your Help,
Lou
Edited by: Wind In Face on Apr 15, 2009 1:34 PM

Answer 1

"I wanted to use PARTITION BY instead of what John suggested because it is my understanding that PARTION BY will be faster"
It may be, or it may not be. As Frank pointed out analytic functions have their uses, and aggregate functions have theis. In some places, those uses do overlap, but not always. You query is equivalent to mine, that is, it returns the same resultset in this case, however, there are some differences.
For the relatively small amount of data I generated, it is probably not significant, but the analytic version does two sorts (one unique) while my aggregate version does only one.
SQL> CREATE TABLE test (office NUMBER, customer NUMBER, tran_dt DATE);
Table created.
SQL> INSERT /*+ APPEND */ INTO test
  2  SELECT MOD(rownum, 10)+1, MOD(rownum, 121)+1, TRUNC(sysdate+MOD(rownum, 42))
  3  FROM all_objects;
18135 rows created.
SQL> COMMIT;
Commit complete.
SQL> SELECT office, tran_dt, COUNT(DISTINCT customer) cust_count
  2  FROM test
  3  GROUP BY office, tran_dt;
210 rows selected.
Execution Plan
Plan hash value: 2407667464
| Id  | Operation          | Name |
|   0 | SELECT STATEMENT   |      |
|   1 |  SORT GROUP BY     |      |
|   2 |   TABLE ACCESS FULL| TEST |
Statistics
          0  recursive calls
          0  db block gets
         27  consistent gets
          0  physical reads
          0  redo size
       6061  bytes sent via SQL*Net to client
        631  bytes received via SQL*Net from client
         15  SQL*Net roundtrips to/from client
          1  sorts (memory)
          0  sorts (disk)
        210  rows processed
SQL> SELECT DISTINCT office office, tran_dt tran_date,
  2         COUNT(DISTINCT customer) OVER(PARTITION BY office, tran_dt) cust_count
  3  FROM test;
210 rows selected.
Execution Plan
Plan hash value: 1303194651
| Id  | Operation           | Name |
|   0 | SELECT STATEMENT    |      |
|   1 |  SORT UNIQUE        |      |
|   2 |   WINDOW SORT       |      |
|   3 |    TABLE ACCESS FULL| TEST |
Statistics
          0  recursive calls
          0  db block gets
         27  consistent gets
          0  physical reads
          0  redo size
       6063  bytes sent via SQL*Net to client
        631  bytes received via SQL*Net from client
         15  SQL*Net roundtrips to/from client
          2  sorts (memory)
          0  sorts (disk)
        210  rows processedFor a larger resultset, this could make a significant difference.
In general, my preference is use the simplest construct that will work.
John