Partition pruning, Nested loops

Question

Hi,
I am having a problem with getting partition pruning in a query. I've managed to dumb down the problem to two tables and a minimal query (see below).
Basically, I have a partitioned table "Yearly Facts", and a helper table "Current Year" that always contain 1 row. The sole purpose of this one row is to tell what the curent year is, what the previous year was and what the next year will be. (In the real problem, there are non-standard timeperiods, so one cannot just calculate previous and next by adding/subtracting 1 as would be possible in this example).
The following query is executed the way I want.
It performs a scan on current_year, and then nested loop into the facts. And partition pruning was happening.
select sum(decode(a.year_key, b.curr_year, some_measure)) as curr_year_measure
  from yearly_fact_t a
      ,current_year  b
where a.year_key = b.curr_year;
| Id  | Operation                  | Name          | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT           |               |     1 |    39 |     4   (0)| 00:00:01 |       |    |
|   1 |  SORT AGGREGATE            |               |     1 |    39 |            |          |       |    |
|   2 |   NESTED LOOPS             |               |     1 |    39 |     4   (0)| 00:00:01 |       |    |
|   3 |    INDEX FAST FULL SCAN    | SYS_C00247890 |     1 |    13 |     2   (0)| 00:00:01 |       |    |
|   4 |    PARTITION RANGE ITERATOR|               |     1 |    26 |     2   (0)| 00:00:01 |   KEY |   KEY |
|*  5 |     TABLE ACCESS FULL      | YEARLY_FACT_T |     1 |    26 |     2   (0)| 00:00:01 |   KEY |   KEY |
Predicate Information (identified by operation id):
   5 - filter("A"."YEAR_KEY"="B"."CURR_YEAR")The following query is where I have my problem.
The basic xplan is the same, but for some reason the cbo gave up and decide to scan all partitions, which was not very scalable on production data :)
I would have thought that the plan would be the same.
select sum(decode(a.year_key, b.curr_year, some_measure)) as curr_year_measure
      ,sum(decode(a.year_key, b.prev_year, some_measure)) as prev_year_measure
  from yearly_fact_t a
      ,current_year  b
where a.year_key = b.curr_year
    or a.year_key = b.prev_year;
| Id  | Operation             | Name          | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT      |               |     1 |    52 |    13   (0)| 00:00:01 |       |       |
|   1 |  SORT AGGREGATE       |               |     1 |    52 |            |          |       |       |
|   2 |   NESTED LOOPS        |               |     4 |   208 |    13   (0)| 00:00:01 |       |       |
|   3 |    TABLE ACCESS FULL  | CURRENT_YEAR  |     1 |    26 |     4   (0)| 00:00:01 |       |       |
|   4 |    PARTITION RANGE ALL|               |     4 |   104 |     9   (0)| 00:00:01 |     1 |     6 |
|*  5 |     TABLE ACCESS FULL | YEARLY_FACT_T |     4 |   104 |     9   (0)| 00:00:01 |     1 |     6 |
Predicate Information (identified by operation id):
   5 - filter("A"."YEAR_KEY"="B"."CURR_YEAR" OR "A"."YEAR_KEY"="B"."PREV_YEAR")
-- drop table yearly_fact_t purge;
-- drop table current_year  purge;
create table current_year(
   curr_year number(4) not null
  ,prev_year number(4) not null
  ,next_year number(4) not null
  ,primary key(curr_year)
  ,unique(prev_year)
  ,unique(next_year)
insert into current_year(curr_year, prev_year, next_year) values(2010, 2009, 2011);
commit;
create table yearly_fact_t(
   year_key     number(4) not null
  ,some_dim_key number    not null
  ,some_measure number    not null
partition by range(year_key)(
   partition p2007 values less than(2008)
  ,partition p2008 values less than(2009)
  ,partition p2009 values less than(2010)
  ,partition p2010 values less than(2011)
  ,partition p2011 values less than(2012) 
  ,partition pmax  values less than(maxvalue)
insert into yearly_fact_t(year_key, some_dim_key, some_measure) values(2007,1, 10);
insert into yearly_fact_t(year_key, some_dim_key, some_measure) values(2008,1, 20);
insert into yearly_fact_t(year_key, some_dim_key, some_measure) values(2009,1, 30);
insert into yearly_fact_t(year_key, some_dim_key, some_measure) values(2010,1, 40);
commit; What can I do to get partition pruning to happen in the second query?
Or even better, what is it in my query that prevents it from happening?
We're running Oracle Database 10g Enterprise Edition Release 10.2.0.4.0 - 64bit.
Best regards
Ronnie

Answer 1

Hi Ronnie,
Check whether your statistics are up to date.
SQL> set autotrace traceonly exp
SQL> select sum(decode(a.year_key, b.curr_year, some_measure)) as curr_year_measure
  2        ,sum(decode(a.year_key, b.prev_year, some_measure)) as prev_year_measure
  3    from yearly_fact_t a
  4        ,current_year  b
  5   where a.year_key = b.curr_year
  6      or a.year_key = b.prev_year;
Execution Plan
Plan hash value: 229094315
| Id  | Operation             | Name          | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT      |               |     1 |    52 |    10   (0)| 00:00:01 |       |       |
|   1 |  SORT AGGREGATE       |               |     1 |    52 |            |          |       |       |
|   2 |   NESTED LOOPS        |               |     4 |   208 |    10   (0)| 00:00:01 |       |       |
|   3 |    TABLE ACCESS FULL  | CURRENT_YEAR  |     1 |    26 |     3   (0)| 00:00:01 |       |       |
|   4 |    PARTITION RANGE ALL|               |     4 |   104 |     7   (0)| 00:00:01 |     1 |     6 |
|*  5 |     TABLE ACCESS FULL | YEARLY_FACT_T |     4 |   104 |     7   (0)| 00:00:01 |     1 |     6 |
Predicate Information (identified by operation id):
   5 - filter("A"."YEAR_KEY"="B"."CURR_YEAR" OR "A"."YEAR_KEY"="B"."PREV_YEAR")
Note
   - dynamic sampling used for this statement
SQL> set autotrace off
SQL> exec dbms_stats.gather_table_stats(user, 'current_year', estimate_percent=>100, cascade=>true, method_opt
=>'for all columns size 1');
PL/SQL procedure successfully completed.
SQL>
SQL> exec dbms_stats.gather_table_stats(user, 'yearly_fact_t', estimate_percent=>100, cascade=>true, method_op
t=>'for all columns size 1');
PL/SQL procedure successfully completed.
SQL> set autotrace traceonly exp
SQL> select sum(decode(a.year_key, b.curr_year, some_measure)) as curr_year_measure
  2        ,sum(decode(a.year_key, b.prev_year, some_measure)) as prev_year_measure
  3    from yearly_fact_t a
  4        ,current_year  b
  5   where a.year_key = b.curr_year
  6      or a.year_key = b.prev_year;
Execution Plan
Plan hash value: 2253546831
| Id  | Operation                   | Name          | Rows  | Bytes | Cost (%CPU)| Time     | Pstart| Pstop |
|   0 | SELECT STATEMENT            |               |     1 |    15 |     8   (0)| 00:00:01 |       |       |
|   1 |  SORT AGGREGATE             |               |     1 |    15 |            |          |       |       |
|   2 |   CONCATENATION             |               |       |       |            |          |       |       |
|   3 |    NESTED LOOPS             |               |     1 |    15 |     4   (0)| 00:00:01 |       |       |
|   4 |     TABLE ACCESS FULL       | CURRENT_YEAR  |     1 |     8 |     3   (0)| 00:00:01 |       |       |
|   5 |     PARTITION RANGE ITERATOR|               |     1 |     7 |     1   (0)| 00:00:01 |   KEY |   KEY |
|*  6 |      TABLE ACCESS FULL      | YEARLY_FACT_T |     1 |     7 |     1   (0)| 00:00:01 |   KEY |   KEY |
|   7 |    NESTED LOOPS             |               |     1 |    15 |     4   (0)| 00:00:01 |       |       |
|   8 |     TABLE ACCESS FULL       | CURRENT_YEAR  |     1 |     8 |     3   (0)| 00:00:01 |       |       |
|   9 |     PARTITION RANGE ITERATOR|               |     1 |     7 |     1   (0)| 00:00:01 |   KEY |   KEY |
|* 10 |      TABLE ACCESS FULL      | YEARLY_FACT_T |     1 |     7 |     1   (0)| 00:00:01 |   KEY |   KEY |
Predicate Information (identified by operation id):
   6 - filter("A"."YEAR_KEY"="B"."PREV_YEAR")
  10 - filter("A"."YEAR_KEY"="B"."CURR_YEAR" AND LNNVL("A"."YEAR_KEY"="B"."PREV_YEAR"))
SQL> set autotrace off
SQL>Asif Momen
http://momendba.blogspot.com