本文为基数反馈(Cardinality Feedback 以后简称CFB)功能的第二部分,主要介绍CFB有效时的状况例子,以及CFB处理流程。
关于CFB无效时的状况例子,以及CFB概述请参考前篇文章:
基数反馈 (Cardinality Feedback)(一)
下面我们在11.2.0.4的环境中也就是CFB有效的情况下,看看执行的情况: (我们依然使用Oracle数据库提供的样例Schema OE 及其表PRODUCT_INFORMATION和ORDER_ITEMS进行测试。)
1.首先确认相关表的统计信息和表的数据量。(基于11.2.0.4版本测试)
--统计信息能够反映出表中的数据量。
SQL> select TABLE_NAME,NUM_ROWS,BLOCKS from user_tables where TABLE_NAME in ('PRODUCT_INFORMATION','ORDER_ITEMS');
TABLE_NAME NUM_ROWS BLOCKS
-------------------- ---------- ----------
ORDER_ITEMS 665 5
PRODUCT_INFORMATION 288 13
Elapsed: 00:00:00.04
SQL> select count(*) from ORDER_ITEMS;
COUNT(*)
----------
665
Elapsed: 00:00:00.02
SQL> select count(*) from PRODUCT_INFORMATION;
COUNT(*)
----------
288
Elapsed: 00:00:00.01
SQL>
2.设定环境参数statistics_level为ALL,以便能够通过dbms_xplan.display_cursor函数查看SQL文根据统计信息估算出的访问数据行数和SQL执行时的实际值。
SQL> alter session set statistics_level=all;
Session altered.
Elapsed: 00:00:00.01
3.第一次执行SQL文
SQL>
SQL> SELECT o.order_id, v.product_name
2 FROM orders o,
3 ( SELECT order_id, product_name
4 FROM order_items o, product_information p
5 WHERE p.product_id = o.product_id
6 AND list_price < 50
7 AND min_price < 40 ) v
8 WHERE o.order_id = v.order_id
9 ;
ORDER_ID PRODUCT_NAME
---------- --------------------
2403 Battery - EL
...
2450 Plastic Stock - W/HD
269 rows selected.
Elapsed: 00:00:00.22
SQL>
4.查看第一次执行后的执行计划。
SQL> select * from table(dbms_xplan.display_cursor(format=>'typical iostats last -cost -bytes'));
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------
SQL_ID bmh5hb8331u33, child number 0
-------------------------------------
SELECT o.order_id, v.product_name FROM orders o, ( SELECT ...
o.order_id = v.order_id
Plan hash value: 1906736282
---------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | E-Time | A-Rows | A-Time | Buffers | Reads |
---------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 269 |00:00:00.17 | 1337 | 20 |
| 1 | NESTED LOOPS | | 1 | 1 | 00:00:01 | 269 |00:00:00.17 | 1337 | 20 |
| 2 | MERGE JOIN CARTESIAN| | 1 | 4 | 00:00:01 | 9135 |00:00:00.06 | 33 | 15 |
|* 3 | TABLE ACCESS FULL | PRODUCT_INFORMATION | 1 | ★ 1 | 00:00:01 | ★ 87 |00:00:00.01 | 32 | 14 |
| 4 | BUFFER SORT | | 87 | 105 | 00:00:01 | 9135 |00:00:00.02 | 1 | 1 |
| 5 | INDEX FULL SCAN | ORDER_PK | 1 | 105 | 00:00:01 | 105 |00:00:00.01 | 1 | 1 |
|* 6 | INDEX UNIQUE SCAN | ORDER_ITEMS_UK | 9135 | 1 | | 269 |00:00:00.05 | 1304 | 5 |
---------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
3 - filter(("MIN_PRICE"<40 AND "LIST_PRICE"<50))
6 - access("O"."ORDER_ID"="ORDER_ID" AND "P"."PRODUCT_ID"="O"."PRODUCT_ID")
28 rows selected.
Elapsed: 00:00:00.19
我们发现和10.2.0.5环境一样,由于访问条件(“MIN_PRICE”<40 AND “LIST_PRICE”<50)的影响,优化器认为PRODUCT_INFORMATION表的预估行数(E-Rows)为1,优化器基于预估基数在选择表PRODUCT_INFORMATION和ORDER_ITEMS结合的最优执行计划时,选择了MERGE JOIN CARTESIAN的结合方式。
5.查看动态视图V$sql和v$SQL_SHARED_CURSOR
SQL> ---sql_id:bmh5hb8331u33
SQL> select sql_id,child_number, executions, buffer_gets,plan_hash_value
2 from v$sql
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER EXECUTIONS BUFFER_GETS PLAN_HASH_VALUE
------------- ------------ ---------- ----------- ---------------
bmh5hb8331u33 0 1 1604 1906736282
Elapsed: 00:00:00.01
SQL>
SQL> select sql_id, child_number, USE_FEEDBACK_STATS
2 from V$SQL_SHARED_CURSOR
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER U
------------- ------------ -
bmh5hb8331u33 0 Y
Elapsed: 00:00:00.04
SQL>
我们发现V$SQL_SHARED_CURSOR的USE_FEEDBACK_STATS列标记为Y。 (USE_FEEDBACK_STATS列是在11.2.0.4 的版本上新追加的列,用于标示当根据统计信息估算出的基数(Computed cardinality)和SQL执行时的实际值差距很大时,下次执行时重新生成执行计划)
6.我们再次次执行相同的SQL文
---第二次执行
SQL> SELECT o.order_id, v.product_name
...
8 WHERE o.order_id = v.order_id
9 ;
SQL>
7.再次查看执行计划
SQL> select * from table(dbms_xplan.display_cursor(format=>'typical iostats last -cost -bytes'));
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------
SQL_ID bmh5hb8331u33, child number 1
-------------------------------------
SELECT o.order_id, v.product_name FROM orders o, ( SELECT
order_id, product_name FROM order_items o,
product_information p WHERE p.product_id = o.product_id
AND list_price < 50 AND min_price < 40 ) v WHERE
o.order_id = v.order_id
Plan hash value: 35479787
----------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Starts | E-Rows | E-Time | A-Rows | A-Time | Buffers | Reads |
----------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | | | 269 |00:00:00.01 | 61 | 1 |
| 1 | NESTED LOOPS | | 1 | 313 | 00:00:01 | 269 |00:00:00.01 | 61 | 1 |
|* 2 | HASH JOIN | | 1 | 313 | 00:00:01 | 269 |00:00:00.01 | 40 | 1 |
|* 3 | TABLE ACCESS FULL | PRODUCT_INFORMATION | 1 | ★87 | 00:00:01 | ★87 |00:00:00.01 | 15 | 0 |
| 4 | INDEX FAST FULL SCAN| ORDER_ITEMS_UK | 1 | 665 | 00:00:01 | 665 |00:00:00.01 | 25 | 1 |
|* 5 | INDEX UNIQUE SCAN | ORDER_PK | 269 | 1 | | 269 |00:00:00.01 | 21 | 0 |
----------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
2 - access("P"."PRODUCT_ID"="O"."PRODUCT_ID")
3 - filter(("MIN_PRICE"<40 AND "LIST_PRICE"<50))
5 - access("O"."ORDER_ID"="ORDER_ID")
Note
-----
- cardinality feedback used for this statement ★
32 rows selected.
Elapsed: 00:00:00.03
我们发现SQL文进行了硬解析,并且表PRODUCT_INFORMATION的预估信息(E-Rows)调整为第一次执行时收集的实际值(87),用于优化器选择执行计划。因此,优化器基于调整后预估基数在选择表PRODUCT_INFORMATION和ORDER_ITEMS结合的最优执行计划时,选择了HASH JOIN的结合方式,从而更有效的执行了SQL文。
8.再次查看动态视图V和SQL_SHARED_CURSOR
SQL> ---sql_id:bmh5hb8331u33
SQL> select sql_id,child_number, executions, buffer_gets,plan_hash_value,is_shareable
2 from v$sql
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER EXECUTIONS BUFFER_GETS PLAN_HASH_VALUE I
------------- ------------ ---------- ----------- --------------- -
bmh5hb8331u33 0 1 1604 1906736282 N★
bmh5hb8331u33 1 1 61 35479787 Y★
Elapsed: 00:00:00.02
SQL>
SQL> select sql_id, child_number, USE_FEEDBACK_STATS
2 from V$SQL_SHARED_CURSOR
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER U
------------- ------------ -
bmh5hb8331u33 0 Y
bmh5hb8331u33 1 N
Elapsed: 00:00:00.00
通过视图V$SQL我们发现,新生成的游标CHILD#1比以前的游标CHILD#1会使用更少的BUFFER_GETS,效率更高。并且以前游标CHILD#0的is_shareable列标记为N,不在被共享。 新生成的游标CHILD#1的is_shareable列标记为Y,供以后的执行重用。
9.再多次执行SQL文
--第三次执行
SQL> SELECT o.order_id, v.product_name
...
8 WHERE o.order_id = v.order_id
9 ;
--第四次执行
SQL> SELECT o.order_id, v.product_name
...
8 WHERE o.order_id = v.order_id
9 ;
ORDER_ID PRODUCT_NAME
---------- --------------------
2403 Battery - EL
...
2401 SPNIX3.3 AU
269 rows selected.
Elapsed: 00:00:00.05
SQL>
--查看执行计划
SQL> set line 200
SQL> set pagesize 9999
SQL>
SQL> select * from table(dbms_xplan.display_cursor(format=>'typical iostats last -cost -bytes'));
PLAN_TABLE_OUTPUT
---------------------------------------------------------------------------------------------
SQL_ID bmh5hb8331u33, child number 1
-------------------------------------
...
Note
-----
- cardinality feedback used for this statement
32 rows selected.
Elapsed: 00:00:00.02
SQL>
--查看动态视图
SQL>
SQL> ---sql_id:bmh5hb8331u33
SQL> select sql_id,child_number, executions, buffer_gets,plan_hash_value,is_shareable
2 from v$sql
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER EXECUTIONS BUFFER_GETS PLAN_HASH_VALUE I
------------- ------------ ---------- ----------- --------------- -
bmh5hb8331u33 0 1 1604 1906736282 N
bmh5hb8331u33 1 3 183 35479787 Y
Elapsed: 00:00:00.00
SQL>
SQL> select sql_id, child_number, USE_FEEDBACK_STATS
2 from V$SQL_SHARED_CURSOR
3 where sql_id = 'bmh5hb8331u33';
SQL_ID CHILD_NUMBER U
------------- ------------ -
bmh5hb8331u33 0 Y
bmh5hb8331u33 1 N
Elapsed: 00:00:00.00
我们发现以后的执行都会变成软解析,使用第二次产生的执行计划。 通过CFB功能使优化器能够在以后的执行中选择更优的执行计划,从得到更好的执行效率。
下面通过以下流程图来总体的回顾一下CFB的处理过程。
在下列情况CBO可能无法估算出准确的Cardinality,Oracle会启用CFB功能:
・没有收集表的统计信息,并且dynamic sampling 也没有开启;
・ 一个表的查询条件涉及多列,但却没有收集扩展的统计信息(extended statistics)
・ 查询条件复杂(比如条件有函数)
针对上述情况,Oracle会采取如下的CFB流程处理:
1. SQL文第一次执行时,Oracle会监控操作的实际行数(A-Row),然后对比CBO估算的行数(E-Row)。
2. 如果两个值相差很大,就记录实际行数(A-Row),做上标记。
下次执行时再次进行硬解析,根据实际行数来重新生成执行计划。
3. 如果两个值相差不大,CBO就不再监控这条SQL语句。
针对版本12c的一些情况我们将在以后的章节中进行介绍。
https://blogs.oracle.com/optimizer/entry/cardinality_feedback
Cardinality Feedback
Statistics (Cardinality) Feedback - Frequently Asked Questions (Doc ID 1344937.1)