Bind Variable Peeking Not Working

Question

Hello,
I'm having an issue with a query I'm running in a stored proc.
The BILLING_TRANS_LOG table is partitioned on orig_process_date and when I run the query with literals the plan makes use of the partitions, but when using values from the procedure parameters or bind variables the plan does not use them and takes a long time to return.
  SELECT btl.billing_trans_id 
FROM compas.billing_trans_log btl, compas.billing_trans_log_alloc btla 
WHERE
btl.billing_trans_id = btla.billing_trans_id 
AND btl.bank_routing_number = '123456789' 
AND btl.bank_account_number = '222222222''
--AND btl.orig_process_date BETWEEN '01-sep-2012' AND '07-sep-2012' 
AND btl.orig_process_date BETWEEN :p_start_date AND :p_stop_date 
AND btla.total_amount = 214.5;
Does anyone know what I might do to force bind variable peeking so that the plan uses the partitions every time, or some other approach that would ensure the use of them?
Thanks,
Christine

Answer 1

Dominic,
Your suggestion of local non prefixed index seems interesting. I have tested it
create table billing_trans_log
     (billing_trans_id     number
     ,orig_process_date    date
     ,bank_routing_number  number
     ,bank_account_number  number
PARTITION BY RANGE (orig_process_date)
PARTITION P_20121201 VALUES LESS THAN (TO_DATE(' 2012-12-01 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN')) ,
PARTITION P_20121202 VALUES LESS THAN (TO_DATE(' 2012-12-02 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN')) ,
PARTITION P_20121203 VALUES LESS THAN (TO_DATE(' 2012-12-03 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN')) ,
PARTITION P_20121204 VALUES LESS THAN (TO_DATE(' 2012-12-04 00:00:00', 'SYYYY-MM-DD HH24:MI:SS', 'NLS_CALENDAR=GREGORIAN')) 
alter table billing_trans_log add constraint PK_BILLING_TRANS_LOG primary key (billing_trans_id);
create table billing_trans_log_alloc (billing_trans_id number,total_amount number);
create index IDX_BILLING_TRANS_LOG_ALLOC on billing_trans_log_alloc(billing_trans_id);
SELECT btl.billing_trans_id
FROM billing_trans_log btl
  , billing_trans_log_alloc btla
WHERE
btl.billing_trans_id = btla.billing_trans_id
AND btl.bank_routing_number = '123456789'
AND btl.bank_account_number = '222222222'
AND btl.orig_process_date BETWEEN to_date('01122012','ddmmyyyy') AND to_date('02122012','ddmmyyyy')
AND btla.total_amount = 214.5;
| Id  | Operation                    | Name                        | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT             |                             |       |       |     2 (100)|          |       |       |
|   1 |  NESTED LOOPS                |                             |       |       |            |          |       |       |
|   2 |   NESTED LOOPS               |                             |     1 |    74 |     2   (0)| 00:00:01 |       |       |
|   3 |    PARTITION RANGE ITERATOR  |                             |     1 |    48 |     2   (0)| 00:00:01 |     2 |     3 |
|*  4 |     TABLE ACCESS FULL        | BILLING_TRANS_LOG           |     1 |    48 |     2   (0)| 00:00:01 |     2 |     3 |
|*  5 |    INDEX RANGE SCAN          | IDX_BILLING_TRANS_LOG_ALLOC |     1 |       |     0   (0)|          |       |       |
|*  6 |   TABLE ACCESS BY INDEX ROWID| BILLING_TRANS_LOG_ALLOC     |     1 |    26 |     0   (0)|          |       |       |
Predicate Information (identified by operation id):
   4 - filter(("BTL"."BANK_ROUTING_NUMBER"=123456789 AND "BTL"."BANK_ACCOUNT_NUMBER"=222222222 AND
              "BTL"."ORIG_PROCESS_DATE"<=TO_DATE(' 2012-12-02 00:00:00', 'syyyy-mm-dd hh24:mi:ss')))
   5 - access("BTL"."BILLING_TRANS_ID"="BTLA"."BILLING_TRANS_ID")
   6 - filter("BTLA"."TOTAL_AMOUNT"=214.5)
Lets create the suggested local non prefixed index and re-execute the same query again
create index local_non_prefixed_mho_ind on billing_trans_log (bank_routing_number,bank_account_number) local;
| Id  | Operation                            | Name                        | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT                     |                             |       |       |     1 (100)|          |       |       |
|   1 |  NESTED LOOPS                        |                             |       |       |            |          |       |       |
|   2 |   NESTED LOOPS                       |                             |     1 |    74 |     1   (0)| 00:00:01 |       |       |
|   3 |    PARTITION RANGE ITERATOR          |                             |     1 |    48 |     1   (0)| 00:00:01 |     2 |     3 |
|*  4 |     TABLE ACCESS BY LOCAL INDEX ROWID| BILLING_TRANS_LOG           |     1 |    48 |     1   (0)| 00:00:01 |     2 |     3 |
|*  5 |      INDEX RANGE SCAN                | LOCAL_NON_PREFIXED_MHO_IND  |     1 |       |     1   (0)| 00:00:01 |     2 |     3 |
|*  6 |    INDEX RANGE SCAN                  | IDX_BILLING_TRANS_LOG_ALLOC |     1 |       |     0   (0)|          |       |       |
|*  7 |   TABLE ACCESS BY INDEX ROWID        | BILLING_TRANS_LOG_ALLOC     |     1 |    26 |     0   (0)|          |       |       |
Predicate Information (identified by operation id):
   4 - filter("BTL"."ORIG_PROCESS_DATE"<=TO_DATE(' 2012-12-02 00:00:00', 'syyyy-mm-dd hh24:mi:ss'))
   5 - access("BTL"."BANK_ROUTING_NUMBER"=123456789 AND "BTL"."BANK_ACCOUNT_NUMBER"=222222222)
   6 - access("BTL"."BILLING_TRANS_ID"="BTLA"."BILLING_TRANS_ID")
   7 - filter("BTLA"."TOTAL_AMOUNT"=214.5)
It seems to be a  better plan.
The local non prefixed index is suitable here because the query itself is doing a partition pruning (see this article for details about the attention one should make when creating a local non prefixed index).
I pushed a little bit the test further and replaced the local non prefixed index by a local prefixed one and re-executed the query again
drop index local_non_prefixed_mho_ind;
create index local_prefixed_mho_ind on billing_trans_log
   (orig_process_date, bank_routing_number,bank_account_number) local;
| Id  | Operation                            | Name                        | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT                     |                             |     1 |    74 |     0   (0)| 00:00:01 |       |       |
|   1 |  NESTED LOOPS                        |                             |       |       |            |          |       |       |
|   2 |   NESTED LOOPS                       |                             |     1 |    74 |     0   (0)| 00:00:01 |       |       |
|   3 |    PARTITION RANGE ITERATOR          |                             |     1 |    48 |     0   (0)| 00:00:01 |     2 |     3 |
|   4 |     TABLE ACCESS BY LOCAL INDEX ROWID| BILLING_TRANS_LOG           |     1 |    48 |     0   (0)| 00:00:01 |     2 |     3 |
|*  5 |      INDEX RANGE SCAN                | LOCAL_PREFIXED_MHO_IND      |     1 |       |     0   (0)| 00:00:01 |     2 |     3 |
|*  6 |    INDEX RANGE SCAN                  | IDX_BILLING_TRANS_LOG_ALLOC |     1 |       |     0   (0)| 00:00:01 |       |       |
|*  7 |   TABLE ACCESS BY INDEX ROWID        | BILLING_TRANS_LOG_ALLOC     |     1 |    26 |     0   (0)| 00:00:01 |       |       |
Predicate Information (identified by operation id):
   5 - access("BTL"."ORIG_PROCESS_DATE">=TO_DATE(' 2012-12-01 00:00:00', 'syyyy-mm-dd hh24:mi:ss') AND
              "BTL"."BANK_ROUTING_NUMBER"=123456789 AND "BTL"."BANK_ACCOUNT_NUMBER"=222222222 AND "BTL"."ORIG_PROCESS_DATE"<=TO_DATE('
              2012-12-02 00:00:00', 'syyyy-mm-dd hh24:mi:ss'))
       filter("BTL"."BANK_ROUTING_NUMBER"=123456789 AND "BTL"."BANK_ACCOUNT_NUMBER"=222222222)
   6 - access("BTL"."BILLING_TRANS_ID"="BTLA"."BILLING_TRANS_ID")
   7 - filter("BTLA"."TOTAL_AMOUNT"=214.5)
What I gained here is that there is no filter on the table BILLING_TRANS_LOG. The index became very precise having only an access predicate  
Of course the test should be conducted under meaningful data volume to have an exact feeling about the new indexes
Best regards
Mohamed Houri