控制执行计划之-SPM BASELINE（五）

修正执行计划

baseline不仅仅可以用来稳固执行计划，还可以像SQL PROFILE、SQL PATCH一样用来修正执行计划，本节会介绍如何通过baseline来修正执行计划。首先构建测试需要的表和索引。

test@DLSP>create table t as select * from dba_objects;   Table created.   test@DLSP>create index t_oi_ind on t(object_id);   Index created.   test@DLSP>begin                                                                                   2     dbms_stats.gather_table_stats(ownname          =>'test',                             3                                   tabname          => 't',                             4                                   no_invalidate    => FALSE,                             5                                   estimate_percent => 100,                               6                                   force            => true,                              7                                   degree         => 5,                                   8                                   method_opt       => 'for all columns size 1',       9                                   cascade          => true);                            10   end;                                                                                  11   /                                                                                        PL/SQL procedure successfully completed.

上面的代码创建了测试表t，字段object_id存在索引，分析表的统计信息，不收集直方图。

test@DLSP>select /*+ full(t) */count(object_name) from t where object_id=2;   COUNT(OBJECT_NAME) ------------------                  1 test@DLSP>select * from table(dbms_xplan.display_cursor);   PLAN_TABLE_OUTPUT ------------------------------------------------------------------------------- SQL_ID  bb0mwx9rcbptu, child number 0 ------------------------------------- select /*+ full(t) */count(object_name) from t where object_id=2   Plan hash value: 2966233522   --------------------------------------------------------------------------- | Id  | Operation          | Name | Rows  | Bytes | Cost (%CPU)| Time     | --------------------------------------------------------------------------- |   0 | SELECT STATEMENT   |      |       |       |    49 (100)|          | |   1 |  SORT AGGREGATE    |      |     1 |    24 |            |          | |*  2 |   TABLE ACCESS FULL| T    |     1 |    24 |    49   (0)| 00:00:01 | ---------------------------------------------------------------------------   Predicate Information (identified by operation id): ---------------------------------------------------      2 - filter("OBJECT_ID"=2)

假如我们想修正这个SQL的执行计划，让查询走上索引扫描。我们可以先构造一个SQL，让构造的SQL使用到索引扫描：

test@DLSP>select count(object_name) from t where object_id=2;   COUNT(OBJECT_NAME) ------------------                  1 test@DLSP>select * from table(dbms_xplan.display_cursor);   PLAN_TABLE_OUTPUT ------------------------------------------------------------------------------- SQL_ID  1k6p0yds2rjgp, child number 0 ------------------------------------- select count(object_name) from t where object_id=2   Plan hash value: 3312862475   ----------------------------------------------------------------------------------------- | Id  | Operation                    | Name     | Rows  | Bytes | Cost (%CPU)| Time     | ----------------------------------------------------------------------------------------- |   0 | SELECT STATEMENT             |          |       |       |     2 (100)|          | |   1 |  SORT AGGREGATE              |          |     1 |    24 |            |          | |   2 |   TABLE ACCESS BY INDEX ROWID| T        |     1 |    24 |     2   (0)| 00:00:01 | |*  3 |    INDEX RANGE SCAN          | T_OI_IND |     1 |       |     1   (0)| 00:00:01 | -----------------------------------------------------------------------------------------   Predicate Information (identified by operation id): ---------------------------------------------------      3 - access("OBJECT_ID"=2)

上面我们已经构造了一个SQL，执行计划走的是索引扫描。我们现在可以交换2个SQL的执行计划了。

test@DLSP>declare   2    m_clob clob;   3  begin   4    select sql_fulltext   5      into m_clob   6      from v$sql   7     where sql_id = 'bb0mwx9rcbptu' --------------原始sql的  sql_id   8       and child_number = 0; -----------------为了让sql只返回一行，也可以rownum=1代替   9    dbms_output.put_line(m_clob);  10    dbms_output.put_line(dbms_spm.load_plans_from_cursor_cache(        sql_id          => '1k6p0yds2rjgp', ------------构造SQL的 SQL_ID  11    plan_hash_value => 3312862475, -------------------构造SQL的 PLAN_HASH_VALUE  12    sql_text        => m_clob,  13    fixed           => 'YES', ---------------------禁止演化baseline  14    enabled         => 'YES'));  15   16  end;  17   18  /

通过SPM包的load_plans_from_cursor_cache函数我们索引扫描的执行计划转嫁到了我们原始的SQL上。我们看看再次查询有没有使用到索引扫描：

test@DLSP>select /*+ full(t) */count(object_name) from t where object_id=2;   COUNT(OBJECT_NAME) ------------------                  1   test@DLSP>select * from table(dbms_xplan.display_cursor(null,null));   PLAN_TABLE_OUTPUT ------------------------------------------------------------------------------- SQL_ID  bb0mwx9rcbptu, child number 2 ------------------------------------- select /*+ full(t) */count(object_name) from t where object_id=2   Plan hash value: 3312862475   ----------------------------------------------------------------------------------------- | Id  | Operation                    | Name     | Rows  | Bytes | Cost (%CPU)| Time     | ----------------------------------------------------------------------------------------- |   0 | SELECT STATEMENT             |          |       |       |     2 (100)|          | |   1 |  SORT AGGREGATE              |          |     1 |    24 |            |          | |   2 |   TABLE ACCESS BY INDEX ROWID| T        |     1 |    24 |     2   (0)| 00:00:01 | |*  3 |    INDEX RANGE SCAN          | T_OI_IND |     1 |       |     1   (0)| 00:00:01 | -----------------------------------------------------------------------------------------   Predicate Information (identified by operation id): ---------------------------------------------------      3 - access("OBJECT_ID"=2)   Note -----    - SQL plan baseline SQL_PLAN_3uq7qskvgtv1s4f379100 used for this statement

来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/22034023/viewspace-1241483/，如需转载，请注明出处，否则将追究法律责任。

转载于:http://blog.itpub.net/22034023/viewspace-1241483/