RE: Why partition pruning is not happening?

From: Polarski, Bernard <Bernard.Polarski_at_atosorigin.com>
Date: Fri, 5 Jan 2007 14:03:06 +0100
Message-ID: <25D4919915CCF742A88EE3366D6D913D113D185D@mailserver1>

Partitioned on FE_DIA with an additional filter on ID_MES.

When the CBO creates the plan, how should he knows that ID_MES is always a logical subset within FE_DIA?

There is not transitive predicate between FE_DIA and ID_MES and ID_MES is not organized as a ranged sub-partition of FE_DIA.

As it is :

a11.fe_dia = a12.fe_dia is a join condition that could be satisfied by all partitions and

a12.id_mes = 200611 Is a filter that could be true in any partitions.

Bernard Polarski

From: LS Cheng [mailto:exriscer_at_gmail.com] Sent: vrijdag 5 januari 2007 13:19
To: ORACLE-L
Subject: Why partition pruning is not happening?

I have a DWH query which IMHO should do partition prunning but it does not. The query is as follows:

SELECT   a11.id_tp_busq id_tp_busq,
         a14.dt_busqueda dt_busqueda,
         a11.id_site id_site,
         a13.de_site de_site,
         SUM (a11.nt_busqueda) wjxbfs1
    FROM prm_fedia a12,
         prm_site a13,
         prm_tibusqbrs a14,
         prh_bqbusq a11     -- PARTITIONED BY FE_DIA 
   WHERE a11.fe_dia = a12.fe_dia
     AND a11.id_site = a13.id_site
     AND a11.id_tp_busq = a14.id_tp_busq
     AND a12.id_mes = 200611
     AND a13.id_site = 1
     AND a11.tx_termino LIKE 'Without usage%'

GROUP BY a11.id_tp_busq, a14.dt_busqueda, a11.id_site, a13.de_site

| 0 | SELECT STATEMENT | |

2 |   154 |   113 |       |       | 


|   1 |  SORT GROUP BY                         |                   |

2 |   154 |   113 |       |       |


|   2 |   NESTED LOOPS                         |                   |

2 |   154 |   110 |       |       | 


|   3 |    NESTED LOOPS                        |                   |

2 |   130 |   108 |       |       |


|   4 |     NESTED LOOPS                       |                   |

2 |   104 |   106 |       |       | 


|   5 |      TABLE ACCESS BY INDEX ROWID       | PRM_SITE          |

1 |    17 |     1 |       |       |
|*  6 |       INDEX UNIQUE SCAN                | PRM_SITE_PK       |
221 |       |       |       |       | 


|   7 |      PARTITION RANGE ALL               |                   |

|       |       |     1 |    25 |

|*  8 |       TABLE ACCESS BY LOCAL INDEX ROWID| PRH_BQBUSQ        |
2 |    70 |   105 |     1 |    25 | 
|*  9 |        INDEX RANGE SCAN                | PRH_BQBUSQ_LN3    |
100 |       |    50 |     1 |    25 |
|* 10 |     TABLE ACCESS BY INDEX ROWID        | PRM_FEDIA         |
1 |    13 |     1 |       |       | 
|* 11 |      INDEX UNIQUE SCAN                 | PRM_FEDIA_PK      |
85 |       |       |       |       |


|  12 |    TABLE ACCESS BY INDEX ROWID         | PRM_TIBUSQBRS     |

1 |    12 |     1 |       |       | 
|* 13 |     INDEX UNIQUE SCAN                  | PRM_TIBUSQBRS_PK  |
1 |       |       |       |       |
------------------------------------------------------------------------
------------------------------------

Predicate Information (identified by operation id):

   6 - access("A13"."ID_SITE"=1)
   8 - filter("A11"."ID_SITE"=1) 
   9 - access("A11"."TX_TERMINO" LIKE 'Without usage%')
       filter("A11"."TX_TERMINO" LIKE 'Without usage%')
  10 - filter("A12"."ID_MES"=200611)

11 - access("A11"."FE_DIA"="A12"."FE_DIA") 13 - access("A11"."ID_TP_BUSQ"="A14"."ID_TP_BUSQ")

I was expecting the condition a11.fe_dia = a12.fe_dia eliminates partitions and only reads 2006 November Partition. Any clues why it is not happening...?

This is 9.2.0.4 running on HPUX

Cheers

--
LSC 





--
http://www.freelists.org/webpage/oracle-l

Received on Fri Jan 05 2007 - 07:03:06 CST