How to create outline from perfstat.stats$SQL_PLAN table

Question

How do I create outline (Plan stability in 10gR2database) from perfstat.stats$SQL_PLAN table?
Current V$SQL_PLAN does not contain good but but week older snapshots have good plan.
I can't modify query, it's vendor query so I need to create outline/plan stability.
Thanks for help.

Answer 1

user7478143 wrote:
Thanks for reply.
I'm not saying I find it out doc but here is situation.
I have query which change execution plan week ago.
I have perfstat enable and I found out good execution plan from perfstat.stat$SQL_PLAN view.
If it's our query I can modify query and able to get good execution plan but it's vendor supplied query (which is Oracle - it used to be Portal now Oracle acquired it) so I can't modified it.
All I'm trying to do is how do I generate good plan and assigned to vendor query's hash_value/sql_id.If DBMS_ADVANCED_REWRITE is not an option for you and you want to use an outline, there is one significant problem. The Statspack captured data does not include quite enough information, in particular the OTHER_XML column, and the access predicates. So, what can you do? Note that this script is from page 214 of the book that I co-authored. Let's say you execute this SQL statement:
SQL> SELECT
  2    SQL_ID,
  3    COUNT(DISTINCT PLAN_HASH_VALUE) C
  4  FROM
  5    PERFSTAT.STATS$SQL_PLAN_USAGE
  6  GROUP BY
  7    SQL_ID
  8  HAVING
  9    COUNT(DISTINCT PLAN_HASH_VALUE)>1;
SQL_ID         C
0fr8zhn4ymu3v  2
0h6b2sajwb74n  2
1gu8t96d0bdmu  2
39m4sx9k63ba2  2
4b57myt9mpz37  3
52tr7ay85qwn0  5
…Interesting, SQL_ID has 2 different execution plans, let's take a look at the execution plans:
SQL> SET LINESIZE 150
SQL> SET PAGESIZE 10000
SQL> SPOOL StatspackPlan.txt
SQL> SELECT /*+ ORDERED */
  2    T.*
  3  FROM
  4    (SELECT DISTINCT
  5       PLAN_HASH_VALUE
  6     FROM
  7       PERFSTAT.STATS$SQL_PLAN_USAGE
  8     WHERE
  9       SQL_ID='0fr8zhn4ymu3v'
10     ORDER BY
11       PLAN_HASH_VALUE) SPU,
12    TABLE(DBMS_XPLAN.DISPLAY(
13      'PERFSTAT.STATS$SQL_PLAN',
14      NULL,
15      'TYPICAL -PREDICATE -NOTE',
16      'PLAN_HASH_VALUE='||SPU.PLAN_HASH_VALUE)) T;
| Id  | Operation                   | Name       | Rows  | Bytes | Cost (%CPU)| Time     |
|*  0 | SELECT STATEMENT            |            |       |       |     2 (100)|          |
|*  1 |  TABLE ACCESS BY INDEX ROWID| OPQTYPE$   |     1 |    27 |     2   (0)| 00:00:01 |
|*  2 |   INDEX RANGE SCAN          | I_OPQTYPE1 |     1 |       |     1   (0)| 00:00:01 |
| Id  | Operation             | Name     | Cost  |
|*  0 | SELECT STATEMENT      |          |       |
|*  1 |  SORT ORDER BY        |          |     0 |
|*  2 |   TABLE ACCESS CLUSTER| OPQTYPE$ |       |
|*  3 |    INDEX UNIQUE SCAN  | I_OBJ#   |       |
--------------------------------------------------As can be seen by the above, in one case the I_OBJ# index was used, and in another case the I_OPQTYPE1 index was used. Let's assume that I_OPQTYPE1 is the most efficient access path, so a hint like this needs to be inserted (assuming that the table does not have an alias in the SQL statement:
/*+ INDEX(OPQTYPE$ I_OPQTYPE1) */Now what (obviously, we would not want to actually fix the above plan for an internal SQL statement)? Take a look at the following, which shows how to build a private outline for the unhinted version of the SQL statement, an outline for a hinted version, and then swap the resulting outlines, followed by building a public outline from the hacked outline:
http://hoopercharles.wordpress.com/2009/12/18/tracking-performance-problems-inserting-a-hint-into-sql-in-a-compiled-program/
Charles Hooper
Co-author of "Expert Oracle Practices: Oracle Database Administration from the Oak Table"
http://hoopercharles.wordpress.com/
IT Manager/Oracle DBA
K&M Machine-Fabricating, Inc.