FBI and Histogrammes

Question

Hi,
Database configuration:
SQL> select * from v$version;
BANNER
Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bi
PL/SQL Release 10.2.0.4.0 - Production
CORE    10.2.0.4.0      Production
TNS for Solaris: Version 10.2.0.4.0 - Production
NLSRTL Version 10.2.0.4.0 - ProductionI have bad performance in a production database with the following query:
SELECT X.NUBIX,X.NUFDP,X.COMAR,X.COINT,X.NUCPT,X.CDCOF,NVL(S.RGCOD,R.RGCOD),NVL(S.RGCID,R.RGCID),TO_DATE('16112009','DDMMYYYY'),
X.COINI,X.CTFLU,X.NUPRO,X.NCCOF,X.CTCOF,X.MHCOF,X.MGCOF,X.MHPND,X.MGPND,X.TXCLO
FROM VCOAT3 X,TYPSCR S,RGCCAL R
WHERE X.COMAR IN ('ICE') AND S.CTCOF=X.CTCOF AND S.COLAN='A' AND R.COMAR(+)=X.COMAR AND R.COINT(+)=X.COINT AND NVL(R.NUCPT(+),'#*#')=NVL(X.NUCPT,'#*#') VCOAT3 is a view defined by this script below:
SELECT /*+ OPT_PARAM('optimizer_index_cost_adj' 100) OPT_PARAM('optimizer_index_caching' 0) */
L.NUBIX,L.NUFDP,
L.COMAR,
DECODE(X.NCCOF,5,L.COINT,6,L.COINT,7,L.COINT,8,L.COINI,9,L.COINI,10,L.COINI,11,NULL   ,12,D.COINA,13,D.COINV,14,L.COINT),
DECODE(X.NCCOF,5,L.NUCPT,6,L.NUCPT,7,L.NUCPT,8,L.NUCPI,9,L.NUCPI,10,L.NUCPI,11,NULL   ,12,NULL   ,13,NULL   ,14,L.NUCPT),
DECODE(X.NCCOF,5,L.COINI,6,L.COINI,7,L.COINI,8,NULL   ,9,NULL   ,10,NULL   ,11,NULL   ,12,NULL   ,13,NULL   ,14,L.COINI),
DECODE(X.NCCOF,5 ,'D',
               6 ,'D',
               7 ,'D',
               8 ,DECODE(I.CTINT,'I','S','T'),
               9 ,DECODE(I.CTINT,'I','S','T'),
               10,DECODE(I.CTINT,'I','S','T'),
               11,'N',
               12,'A',
               13,'O',
               14,'L'),
L.DANEG,L.CSENS,L.QTCCP,L.MTULP,L.CSOPT,L.CNACT,L.CMECH,L.CAECH,L.MTSNA,L.NUCON,
X.NUPRO,X.NCCOF,X.CTCOF,X.CDCOF,
X.MHCOF,
X.MGCOF,
X.MHPND,
X.MHPND*(1+X.TTCOF/100),
X.TXCLO
FROM
LIGPOR L,
HISCRD X,
IHSDEP D,
INTERV I
WHERE
X.NUBIX=L.NUBIX    AND
X.NUFDP=L.NUFDP    AND
SIGN(X.MHPND) IN (-1,1) AND
D.NUBIX=L.NUBIX    AND
D.NUFDP=L.NUFDP    AND
I.COINT=L.COINT;here is the execution plan:
| Id  | Operation                    | Name   | Rows  | Bytes |TempSpc| Cost (%CPU)| Time     | Pstart| Pstop
|
|   0 | SELECT STATEMENT             |        |    49M|  9979M|       |   418K  (6)| 00:26:15 |       |
|
|*  1 |  HASH JOIN                   |        |    49M|  9979M|       |   418K  (6)| 00:26:15 |       |
|
|*  2 |   TABLE ACCESS FULL          | TYPSCR |    31 |   682 |       |     3   (0)| 00:00:01 |       |
|
|*  3 |   HASH JOIN RIGHT OUTER      |        |    34M|  6350M|       |   417K  (6)| 00:26:11 |       |
|
|   4 |    TABLE ACCESS FULL         | RGCCAL |  1826 | 60258 |       |     5   (0)| 00:00:01 |       |
|
|   5 |    VIEW                      | VCOAT3 |    34M|  5253M|       |   416K  (6)| 00:26:09 |       |
|
|*  6 |     HASH JOIN                |        |    34M|  3890M|    56M|   416K  (6)| 00:26:09 |       |
|
|*  7 |      HASH JOIN               |        |   678K|    49M|       |   153K  (4)| 00:09:37 |       |
|
|   8 |       TABLE ACCESS FULL      | INTERV |  1254 | 12540 |       |    13   (0)| 00:00:01 |       |
|
|*  9 |       HASH JOIN              |        |   678K|    42M|    36M|   153K  (4)| 00:09:37 |       |
|
|  10 |        PARTITION RANGE SINGLE|        |   689K|    28M|       | 10554   (3)| 00:00:40 |    16 |    16
|
|* 11 |         TABLE ACCESS FULL    | LIGPOR |   689K|    28M|       | 10554   (3)| 00:00:40 |    16 |    16
|
|  12 |        TABLE ACCESS FULL     | IHSDEP |    15M|   325M|       |   125K  (4)| 00:07:53 |       |
|
|* 13 |      TABLE ACCESS FULL       | HISCRD |    54M|  2150M|       |   174K  (8)| 00:10:58 |       |
|
Predicate Information (identified by operation id):
   1 - access("S"."CTCOF"="X"."CTCOF")
   2 - filter("S"."COLAN"='A')
   3 - access("R"."COMAR"(+)="X"."COMAR" AND "R"."COINT"(+)="X"."COINT" AND
              NVL("R"."NUCPT"(+),'#*#')=NVL("X"."NUCPT",'#*#'))
   6 - access("X"."NUBIX"="L"."NUBIX" AND "X"."NUFDP"="L"."NUFDP")
   7 - access("I"."COINT"="L"."COINT")
   9 - access("D"."NUBIX"="L"."NUBIX" AND "D"."NUFDP"="L"."NUFDP")
  11 - filter("L"."COMAR"='ICE')
  13 - filter(SIGN("X"."MHPND")=(-1) OR SIGN("X"."MHPND")=1)
Note
   - 'PLAN_TABLE' is old version
   - dynamic sampling used for this statementThe problem is the FTS performed on HISCRD table.
The HISCRD table contains 100M of rows and 99.88% of the table have MHPN=0.
Thats's why I have used the following filter predicate:
AND
SIGN(X.MHPND) IN (-1,1)I have created an FBI for this column :
CREATE INDEX HISCRD2 ON HISCRD (SIGN("MHPND")); Statistics exists for this column and a histogramme is calculated:
SQL> select INDEX_NAME,INDEX_TYPE,BLEVEL,CLUSTERING_FACTOR,STATUS,NUM_ROWS,LAST_ANALYZED
  2  from user_indexes
  3  where TABLE_NAME='HISCRD';
INDEX_NAME                     INDEX_TYPE                      BLEVEL CLUSTERING_FACTOR STATUS     NUM_ROWS LAST_ANALYZED
HISCRD2                        FUNCTION-BASED NORMAL                2           1032922 VALID     115850590 15/11/2009 00:30:1
2
HISCRD1                        NORMAL                               3          43339640 VALID     108514554 15/11/2009 00:30:1
4
SQL> select density,NUM_DISTINCT,num_nulls, num_buckets, histogram
  2  from user_tab_columns
  3  where table_name = 'HISCRD' and column_name in ('MHPND');
   DENSITY NUM_DISTINCT  NUM_NULLS NUM_BUCKETS HISTOGRAM
4.5427E-09            5          0           5 FREQUENCY
SQL> select num_rows, sample_size, blocks from user_tables where table_name = 'HISCRD';
  NUM_ROWS SAMPLE_SIZE     BLOCKS
109981839     5987368     732097How does the CBO estimate that there's going to be 54M rows returned in HISCRD table after it has applied the filter predicate SIGN(X.MHPND) IN (-1,1):
SQL> select count(1) from hiscrd where SIGN(MHPND) IN (1);
  COUNT(1)
    127451
SQL> select * from table(dbms_xplan.display_cursor);
PLAN_TABLE_OUTPUT
SQL_ID  80whqnuycnc5x, child number 0
select count(:"SYS_B_0") from hiscrd where SIGN(MHPND) IN (:"SYS_B_1")
Plan hash value: 3957584767
| Id  | Operation         | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |         |       |       | 10743 (100)|          |
|   1 |  SORT AGGREGATE   |         |     1 |     3 |            |          |
|*  2 |   INDEX RANGE SCAN| HISCRD2 |    54M|   157M| 10743   (3)| 00:00:41 |
Predicate Information (identified by operation id):
   2 - access("HISCRD"."SYS_NC00017$"=:SYS_B_1)
19 rows selected.
SQL> select count(1) from hiscrd where SIGN(MHPND) IN (0);
  COUNT(1)
110380609
SQL> select * from table(dbms_xplan.display_cursor);
PLAN_TABLE_OUTPUT
SQL_ID  80whqnuycnc5x, child number 0
select count(:"SYS_B_0") from hiscrd where SIGN(MHPND) IN (:"SYS_B_1")
Plan hash value: 3957584767
| Id  | Operation         | Name    | Rows  | Bytes | Cost (%CPU)| Time     |
|   0 | SELECT STATEMENT  |         |       |       | 10743 (100)|          |
|   1 |  SORT AGGREGATE   |         |     1 |     3 |            |          |
|*  2 |   INDEX RANGE SCAN| HISCRD2 |    54M|   157M| 10743   (3)| 00:00:41 |
Predicate Information (identified by operation id):
   2 - access("HISCRD"."SYS_NC00017$"=:SYS_B_1)Why the CBO doesn't consider histogram in the MHPND column?

Answer 1

Timur Akhmadeev wrote:
Another oddity is why CBO decided to not merge the view's subquery - even though it is a costed tranformation in 10g, usually CBO prefers to merge subqueries.The join is an outer join, and the view being joined is a join view. This combination is one of the restrictions listed as blocking view merging. (It is a case where a push_pred() hint could be used to force the optimizer into using "join predicate pushdown").
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
http://www.jlcomp.demon.co.uk
To post code, statspack/AWR report, execution plans or trace files, start and end the section with the tag {noformat}{noformat} (lowercase, curly brackets, no spaces) so that the text appears in fixed format.
"Science is more than a body of knowledge; it is a way of thinking"
Carl Sagan