SQL & PL/SQL

Hi people !

I need some help because I have an issue with a query that don t use the right Indexes as it should

First of all, I have mainly three tables :
ORDER : Table that contains description for each Order (approximately 1 000 000 Records)
ORDER_MVTS : Table that contains the tasks made (called movements) to set up each Orders
with quantity of packages prepared for each product (approximately 10 000 000 Records)
PRODUCT : Tables that contains the products (approximately 50 000 Records)


When I launch the query with hard coded values, it brings back response very fast
because it uses the right index (ORDER_DHR_VALID) which represent the date and hour of the order
(with format 'DD/MM/YYYY HH24:MI:SS'). The selectivity for this index is good.

NB 1: I have to use the trick " >= Trunc(date) and < trunc(date) +1 " to filter on a simple date because
the index contains hour and minutes (I know it wasn't probably a bright idea at conception time).
NB 2: The index on ORDER_MVTS.PRODUCT_CODE is'nt discriminating enough because there is'nt enough different products.
It's the same for index on CUSTOMER_CODE and on MVT_TYPE so only the index on ORDER.DHR_VALID is good.


Here is the correct explain plan when I execute the query with hard coded values :

-----
SELECT SUM(ORDER_MVTS.NB_PACKAGE)
FROM ORDER_MVTS, PRODUCT, ORDER
WHERE ORDER.DHR_VALID >= TRUNC(to_date('14/11/2008 10:04:56','DD/MM/YYYY HH24:MI:SS'))
AND ORDER.DHR_VALID < TRUNC(to_date('14/11/2008 10:04:56','DD/MM/YYYY HH24:MI:SS')) + 1
AND ORDER_MVTS.MVT_TYPE = 'DELIVERY'
AND PRODUCT.CODE = ORDER_MVTS.PRODUCT_CODE
AND ORDER_MVTS.ORDER_CODE = ORDER.CODE
AND ORDER.CUSTOMER_CODE = 'ADIDAS'
AND PRODUCT.CODE = 1234

Rows Row Source Operation

1 SORT AGGREGATE
2 NESTED LOOPS
4 NESTED LOOPS
2 INDEX UNIQUE SCAN (object id 378548) --> PRODUCT_PK
4 TABLE ACCESS BY INDEX ROWID ORDER
777 INDEX RANGE SCAN (object id 378119) --> ORDER_DHR_VALID
2 TABLE ACCESS BY INDEX ROWID ORDER_MVTS
30 INDEX RANGE SCAN (object id 377784) --> ORDER_MVTS_ORDER_FK
-----

Now the problem is when the query is used in a Cursor with bind variables.
It seems like Oracle don't use index on ORDER.DHR_VALID because he can't figure out that he have
to actually filter on a short period of time (only one day).
So Oracle uses the index on ORDER_MVTS.PRODUCT_CODE which is'nt a bright idea (it takes 10 secondes instead of just one)

Here is the bad explain plan :

-----
Rows Row Source Operation

1 SORT AGGREGATE
2 NESTED LOOPS
722 NESTED LOOPS
2 INDEX UNIQUE SCAN (object id 378548) --> PRODUCT_PK
722 TABLE ACCESS BY INDEX ROWID ORDER_MVTS
1790 INDEX RANGE SCAN (object id 377777) --> ORDER_MVTS_PRODUCT_FK
2 TABLE ACCESS BY INDEX ROWID ORDER
1442 INDEX UNIQUE SCAN (object id 378439) --> ORDER_PK
-----

Now I have found two solutions to this problem :
1) using a Hint to force the use of index on ORDER.DHR_VALID (with /*+ INDEX(ORDER ORDER_DHR_VALID) */ )
2) Using Dynamic SQL and keeping the date hard coded (but not the other values except mvt_type)

For example :

-----
QUERY :=
'SELECT SUM(ORDER_MVTS.NB_PACKAGE)
FROM ORDER_MVTS, PRODUCT, ORDER
WHERE ORDER.DHR_VALID >= TRUNC(TO_DATE('''||To_char(P_DTE_VAL,'DD/MM/YYYY')||''',''DD/MM/YYYY'')) '||
AND ORDER.DHR_VALID < TRUNC(TO_DATE('''||To_char(P_DTE_VAL,'DD/MM/YYYY')||''',''DD/MM/YYYY'')) + 1 '||
AND ORDER_MVTS.MVT_TYPE = 'DELIVERY'
AND PRODUCT.CODE = ORDER_MVTS.PRODUCT_CODE
AND ORDER_MVTS.ORDER_CODE = ORDER.CODE
AND ORDER.CUSTOMER_CODE = :CUSTOMER
AND PRODUCT.CODE = :CODE ';
-----

These two solutions work but Number 1 is bad in theory because it uses a Hint
and Number 2 may be difficult to code.

So my question is : Does someone knows another solution to force the use of index ORDER_DHR_VALID that can be simple and reliable.

Thank you very much for support

Edited by: remaï on Apr 1, 2009 4:08 PM