SQL Tuning. (union all)

Question

Hi,
I am working in Oracle 10g R2 /RAC - Solaris
I have sql which is getting executed more than 1million times during peak hours.
+++++++++++++++++++
SQL Statement
+++++++++++++++++++
select location_id -- (,longitude, latitude)
from ph_location pl
where (longitude between :long1 and :long2)
and (latitude between :lat1 and :lat2)
and catalog_id = :catalog
and location_id in (select location_id
from ph_location_cats
where category_id = :cat
union all
select location_id
from ph_location_prd
where product_id = :prod)
and location_id not in (select location_id
from ph_loc_prd_excluded
where product_id = :prod);
XPLAN
SQL> SET LINESIZE 130
SET PAGESIZE 0
SELECT *
FROM   TABLE(DBMS_XPLAN.DISPLAY);
SQL>   Plan hash value: 3391784670
| Id  | Operation                     | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT              |                        |     1 |    85 |    23   (0)| 00:00:01 |
|*  1 |  FILTER                       |                        |       |       |            |          |
|   2 |   NESTED LOOPS ANTI           |                        |     1 |    85 |    18   (0)| 00:00:01 |
|*  3 |    TABLE ACCESS BY INDEX ROWID| PH_LOCATION            |     1 |    56 |    17   (0)| 00:00:01 |
|*  4 |     INDEX RANGE SCAN          | ID_PH_LOCATION_CALO    |    96 |       |    13   (0)| 00:00:01 |
|   5 |      UNION-ALL                |                        |       |       |            |          |
|*  6 |       INDEX UNIQUE SCAN       | PK_PH_LOCATION_CATS    |     1 |    38 |     3   (0)| 00:00:01 |
|*  7 |       INDEX UNIQUE SCAN       | PK_PH_LOCATION_PRD     |     1 |    30 |     2   (0)| 00:00:01 |
|*  8 |    INDEX UNIQUE SCAN          | PK_PH_LOC_PRD_EXCLUDED |   370 | 10730 |     1   (0)| 00:00:01 |
Predicate Information (identified by operation id):
   1 - filter(TO_NUMBER(:LAT1)<=TO_NUMBER(:LAT2) AND TO_NUMBER(:LONG1)<=TO_NUMBER(:LONG2))
   3 - filter("LONGITUDE">=TO_NUMBER(:LONG1) AND "LONGITUDE"<=TO_NUMBER(:LONG2) AND
              "LATITUDE">=TO_NUMBER(:LAT1) AND "LATITUDE"<=TO_NUMBER(:LAT2))
   4 - access("CATALOG_ID"=:CATALOG)
       filter( EXISTS ( (SELECT /*+ */ "LOCATION_ID" FROM "PH_LOCATION_CATS" "PH_LOCATION_CATS"
              WHERE "CATEGORY_ID"=:CAT AND "LOCATION_ID"=:B1) UNION ALL  (SELECT /*+ */ "LOCATION_ID" FROM
              "PH_LOCATION_PRD" "PH_LOCATION_PRD" WHERE "PRODUCT_ID"=:PROD AND "LOCATION_ID"=:B2)))
   6 - access("LOCATION_ID"=:B1 AND "CATEGORY_ID"=:CAT)
   7 - access("LOCATION_ID"=:B1 AND "PRODUCT_ID"=:PROD)
   8 - access("LOCATION_ID"="LOCATION_ID" AND "PRODUCT_ID"=:PROD)
29 rows selected.
How to tune this query? Is there any way to re-write this sql without unionall? If so, How? Please help.
thanks
Raj

Answer 1

Hi
Thanks for your reply.
The cost of this query is more compared to original query posted.
{noformat}
SQL> explain plan for select location_id -- (,longitude, latitude)
  2  from ph_location pl
  3  where (longitude between :long1 and :long2)
  4  and (latitude between :lat1 and :lat2)
  5  and catalog_id = :catalog
  6  and (
  7  location_id in (select location_id
  8  from ph_location_cats
  9  where category_id = :cat
10  )
11  or
12  location_id in (
13  select location_id
from ph_location_prd
14   15  where product_id = :prod
16  )
17  )
18  and location_id not in (select location_id
19  from ph_loc_prd_excluded
20  where product_id = :prod);
Explained.
SQL> select * from table(dbms_xplan.display());
PLAN_TABLE_OUTPUT
Plan hash value: 2662773144
| Id  | Operation                      | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT               |                        |     1 |    71 |    82   (0)| 00:00:01 |
|*  1 |  FILTER                        |                        |       |       |            |          |
|*  2 |   FILTER                       |                        |       |       |            |          |
|   3 |    NESTED LOOPS ANTI           |                        |     1 |    71 |    82   (0)| 00:00:01 |
|*  4 |     TABLE ACCESS BY INDEX ROWID| PH_LOCATION            |     1 |    42 |    81   (0)| 00:00:01 |
|*  5 |      INDEX RANGE SCAN          | ID_PH_LOCATION_CI      |  1923 |       |    16   (0)| 00:00:01 |
|*  6 |     INDEX UNIQUE SCAN          | PK_PH_LOC_PRD_EXCLUDED |   370 | 10730 |     1   (0)| 00:00:01 |
|*  7 |   INDEX UNIQUE SCAN            | PK_PH_LOCATION_CATS    |     1 |    38 |     3   (0)| 00:00:01 |
|*  8 |    INDEX UNIQUE SCAN           | PK_PH_LOCATION_PRD     |     1 |    30 |     2   (0)| 00:00:01 |
Predicate Information (identified by operation id):
   1 - filter( EXISTS (SELECT /*+ */ 0 FROM "PH_LOCATION_CATS" "PH_LOCATION_CATS" WHERE
              "CATEGORY_ID"=:CAT AND "LOCATION_ID"=:B1) OR  EXISTS (SELECT /*+ */ 0 FROM "PH_LOCATION_PRD"
              "PH_LOCATION_PRD" WHERE "PRODUCT_ID"=:PROD AND "LOCATION_ID"=:B2))
   2 - filter(TO_NUMBER(:LAT1)<=TO_NUMBER(:LAT2) AND TO_NUMBER(:LONG1)<=TO_NUMBER(:LONG2))
   4 - filter("LONGITUDE">=TO_NUMBER(:LONG1) AND "LONGITUDE"<=TO_NUMBER(:LONG2) AND
              "LATITUDE">=TO_NUMBER(:LAT1) AND "LATITUDE"<=TO_NUMBER(:LAT2))
   5 - access("CATALOG_ID"=:CATALOG)
   6 - access("LOCATION_ID"="LOCATION_ID" AND "PRODUCT_ID"=:PROD)
   7 - access("LOCATION_ID"=:B1 AND "CATEGORY_ID"=:CAT)
   8 - access("LOCATION_ID"=:B1 AND "PRODUCT_ID"=:PROD)
29 rows selected.
{noformat}