Oracle優化器之基數反饋（CardinalityFeedback）功能

阿新 • • 發佈：2018-12-10

概述

在Oracle 11gR2的版本上推出了基數反饋（Cardinality Feedback 以後簡稱CFB）功能，通過這個特性，對於某些查詢在第一次執行時，如果CBO發現根據統計資訊估算出的基數(Computed cardinality)和SQL執行時的實際值差距很大的情況發生時，在SQL下次執行時，會根據實際值調整基數，重新生成執行計劃。

另外，基數反饋（CFB）在12c版本上得到更進一步的擴充套件改稱為統計反饋（Statistics Feedback），成為12c自動重新優化（Automatic Reoptimization）的一部分。
關於這統計反饋（Statistics Feedback）中擴充套件的內容和12c自動重新優化（Automatic Reoptimization）的內容，將在以後的章節中進行介紹。

下面我們將通過幾個例子來了解一下CFB功能。

例子1（CFB無效）

首先我們在10.2.0.5的環境中也就是CFB無效的情況下，看看執行的情況：
（我們使用了Oracle資料庫提供的樣例Schema OE 及其表PRODUCT_INFORMATION和ORDER_ITEMS進行測試。）

1.首先確認相關表的統計資訊和表的資料量。（基於10.2.0.5版本測試）

--統計資訊能夠反映出表中的資料量。
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 
 
SQL> select count(*) from ORDER_ITEMS;
 
  COUNT(*)
----------
       665
 
SQL> select count(*) from PRODUCT_INFORMATION;
 
  COUNT(*)
----------
       288

2.設定環境引數statistics_level為ALL，以便能夠通過dbms_xplan.display_cursor函式檢視SQL文根據統計資訊估算出的訪問資料行數和SQL執行時的實際值。

SQL> alter session set statistics_level=all;
 
Session altered.

3.第一次執行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        
---------- --------------------
      2354 Sound Card STD      
...
      2457 Graphics - DIK+     
 
269 rows selected.

4.檢視第一次執行後的執行計劃

SQL> select * from table(dbms_xplan.display_cursor(NULL, NULL,'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 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: 1906736282    
 
---------------------------------------------------------------------------------------------------------------------------
| Id  | Operation             | Name                | Starts | E-Rows | E-Time   | A-Rows |   A-Time   | Buffers | Reads  |
---------------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |                     |      1 |        |          |    269 |00:00:00.44 |    9189 |     20 |
|   1 |  NESTED LOOPS         |                     |      1 |      1 | 00:00:01 |    269 |00:00:00.44 |    9189 |     20 |
|   2 |   MERGE JOIN CARTESIAN|                     |      1 |      4 | 00:00:01 |   9135 |00:00:00.17 |      35 |     15 |
|*  3 |    TABLE ACCESS FULL  | PRODUCT_INFORMATION |      1 |      1 ★| 00:00:01 |     87 ★|00:00:00.01 |      34 |     14 |
|   4 |    BUFFER SORT        |                     |     87 |    105 | 00:00:01 |   9135 |00:00:00.07 |       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 |00:00:00.18 |    9154 |      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")        
 
 
26 rows selected.
 
SQL> ---檢視V$SQL的統計資訊。
SQL> select sql_id,child_number, executions, buffer_gets,plan_hash_value
  2  from v$sql
  3  where sql_id = 'bmh5hb8331u33';
 
SQL_ID        CHILD_NUMBER EXECUTIONS BUFFER_GETS PLAN_HASH_VALUE
------------- ------------ ---------- ----------- ---------------
bmh5hb8331u33            0          1        9701      1906736282

我們發現由於訪問條件(“MIN_PRICE”<40 AND “LIST_PRICE”<50)的影響，優化器認為PRODUCT_INFORMATION表的預估行數（E-Rows）為1，優化器基於預估基數在選擇表PRODUCT_INFORMATION和ORDER_ITEMS結合的最優執行計劃時，選擇了MERGE JOIN CARTESIAN的結合方式。
但實際實際訪問行數（A-Time：87），因此由於預估基數不準，很有可能導致選擇的執行計劃不是最優的。

5.我們再多次執行相同的SQL文

---第二次執行
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        
---------- --------------------
      2354 Sound Card STD      
...
      2457 Graphics - DIK+     
 
269 rows selected.
 
SQL> 
---第二次執行的執行計劃
SQL> select * from table(dbms_xplan.display_cursor(NULL, NULL,'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 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: 1906736282    
 
------------------------------------------------------------------------------------------------------------------
| Id  | Operation             | Name                | Starts | E-Rows | E-Time   | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |                     |      1 |        |          |    269 |00:00:00.45 |    9189 |
|   1 |  NESTED LOOPS         |                     |      1 |      1 | 00:00:01 |    269 |00:00:00.45 |    9189 |
|   2 |   MERGE JOIN CARTESIAN|                     |      1 |      4 | 00:00:01 |   9135 |00:00:00.17 |      35 |
|*  3 |    TABLE ACCESS FULL  | PRODUCT_INFORMATION |      1 |      1 | 00:00:01 |     87 |00:00:00.01 |      34 |
|   4 |    BUFFER SORT        |                     |     87 |    105 | 00:00:01 |   9135 |00:00:00.06 |       1 |
|   5 |     INDEX FULL SCAN   | ORDER_PK            |      1 |    105 | 00:00:01 |    105 |00:00:00.01 |       1 |
|*  6 |   INDEX UNIQUE SCAN   | ORDER_ITEMS_UK      |   9135 |      1 |          |    269 |00:00:00.18 |    9154 |
------------------------------------------------------------------------------------------------------------------
 
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")        
 
 
26 rows selected.
 
---第三次執行的執行計劃
 
    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        
---------- --------------------
      2354 Sound Card STD      
...
      2457 Graphics - DIK+     
 
269 rows selected.
 
--第三次執行的執行計劃    
SQL> select * from table(dbms_xplan.display_cursor(NULL, NULL,'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 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: 1906736282    
 
------------------------------------------------------------------------------------------------------------------
| Id  | Operation             | Name                | Starts | E-Rows | E-Time   | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT      |                     |      1 |        |          |    269 |00:00:00.45 |    9189 |
|   1 |  NESTED LOOPS         |                     |      1 |      1 | 00:00:01 |    269 |00:00:00.45 |    9189 |
|   2 |   MERGE JOIN CARTESIAN|                     |      1 |      4 | 00:00:01 |   9135 |00:00:00.16 |      35 |
|*  3 |    TABLE ACCESS FULL  | PRODUCT_INFORMATION |      1 |      1 | 00:00:01 |     87 |00:00:00.01 |      34 |
|   4 |    BUFFER SORT        |                     |     87 |    105 | 00:00:01 |   9135 |00:00:00.07 |       1 |
|   5 |     INDEX FULL SCAN   | ORDER_PK            |      1 |    105 | 00:00:01 |    105 |00:00:00.01 |       1 |
|*  6 |   INDEX UNIQUE SCAN   | ORDER_ITEMS_UK      |   9135 |      1 |          |    269 |00:00:00.19 |    9154 |
------------------------------------------------------------------------------------------------------------------
 
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")        
 
 
26 rows selected.

我們發現，雖然根據統計資訊估算出的基數(Computed cardinality)和SQL執行時的實際值不同，但是以後的執行過程中，SQL文依然會利用以前的執行計劃（軟解析）。
在這個情況下，很有可能由於最初優化器沒有選擇最優的執行計劃，在以後的重複執行中也得不到改進而導致效率問題。

例子2（CFB有效）

下面我們在11.2.0.4的環境中也就是CFB有效的情況下，看看執行的情況：
（我們依然使用Oracle資料庫提供的樣例Schema OE 及其表PRODUCT_INFORMATION和ORDER_ITEMS進行測試。）

1.首先確認相關表的統計資訊和表的資料量。（基於11.2.0.4版本測試）

--統計資訊能夠反映出表中的資料量。
    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.設定環境引數statistics_level為ALL，以便能夠通過dbms_xplan.display_cursor函式檢視SQL文根據統計資訊估算出的訪問資料行數和SQL執行時的實際值。

SQL> alter session set statistics_level=all;
 
Session altered.
 
Elapsed: 00:00:00.01

3.第一次執行SQL

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.檢視第一次執行後的執行計劃

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                                                              
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: 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

我們發現和10.2.0.5環境一樣，由於訪問條件(“MIN_PRICE”<40 AND “LIST_PRICE”<50)的影響，優化器認為PRODUCT_INFORMATION表的預估行數（E-Rows）為1，優化器基於預估基數在選擇表PRODUCT_INFORMATION和ORDER_ITEMS結合的最優執行計劃時，選擇了MERGE JOIN CARTESIAN的結合方式。

5.檢視動態檢視VSQL和VSQL_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>

我們發現V$SQL_SHARED_CURSOR的USE_FEEDBACK_STATS列標記為Y。
（USE_FEEDBACK_STATS列是在11.2.0.4 的版本上新追加的列，用於標示當根據統計資訊估算出的基數(Computed cardinality)和SQL執行時的實際值差距很大時，下次執行時重新生成執行計劃）

6.我們再次次執行相同的SQL文

---第二次執行
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                                     
...
      2401 SPNIX3.3 AU                                      
 
269 rows selected.
 
Elapsed: 00:00:00.03★
SQL>

我們發現執行時間變短了。

7.再次檢視執行計劃

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

我們發現SQL文進行了硬解析，並且表PRODUCT_INFORMATION的預估資訊（E-Rows）調整為第一次執行時收集的實際值（87），用於優化器選擇執行計劃。因此，優化器基於調整後預估基數在選擇表PRODUCT_INFORMATION和ORDER_ITEMS結合的最優執行計劃時，選擇了HASH JOIN的結合方式，從而更有效的執行了SQL文。

8.再次檢視動態檢視VSQL和VSQL_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

通過檢視V$SQL我們發現，新生成的遊標CHILD#1比以前的遊標CHILD#1會使用更少的BUFFER_GETS，效率更高。並且以前遊標CHILD#0的is_shareable列標記為N，不在被共享。
新生成的遊標CHILD#1的is_shareable列標記為Y，供以後的執行重用。

9.再多次執行SQL文

--第三次執行
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     
...
      2401 SPNIX3.3 AU      
 
269 rows selected.
 
Elapsed: 00:00:00.06
SQL> 
 
--檢視執行計劃
 
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                       
-------------------------------------                       
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 |               
-------------------------------------------------------------------------------------------------------------------               
|   0 | SELECT STATEMENT       |                     |      1 |        |          |    269 |00:00:00.03 |      61 |               
|   1 |  NESTED LOOPS          |                     |      1 |    313 | 00:00:01 |    269 |00:00:00.03 |      61 |               
|*  2 |   HASH JOIN            |                     |      1 |    313 | 00:00:01 |    269 |00:00:00.02 |      40 |               
|*  3 |    TABLE ACCESS FULL   | PRODUCT_INFORMATION |      1 |     87 | 00:00:01 |     87 |00:00:00.01 |      15 |               
|   4 |    INDEX FAST FULL SCAN| ORDER_ITEMS_UK      |      1 |    665 | 00:00:01 |    665 |00:00:00.01 |      25 |               
|*  5 |   INDEX UNIQUE SCAN    | ORDER_PK            |    269 |      1 |          |    269 |00:00:00.01 |      21 |               
-------------------------------------------------------------------------------------------------------------------               
 
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: 00:00:00.07
 
--檢視動態檢視
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          2         122        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
--第四次執行
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     
...
      2401 SPNIX3.3 AU      
 
269 rows selected.
 
Elapsed: 00:00:00.05
SQL> 
 --檢視執行計劃
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                       
-------------------------------------                       
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 |               
-------------------------------------------------------------------------------------------------------------------               
|   0 | SELECT STATEMENT       |                     |      1 |        |          |    269 |00:00:00.02 |      61 |               
|   1 |  NESTED LOOPS          |                     |      1 |    313 | 00:00:01 |    269 |00:00:00.02 |      61 |               
|*  2 |   HASH JOIN            |                     |      1 |    313 | 00:00:01 |    269 |00:00:00.01 |      40 |               
|*  3 |    TABLE ACCESS FULL   | PRODUCT_INFORMATION |      1 |     87 | 00:00:01 |     87 |00:00:00.01 |      15 |               
|   4 |    INDEX FAST FULL SCAN| ORDER_ITEMS_UK      |      1 |    665 | 00:00:01 |    665 |00:00:00.01 |      25 |               
|*  5 |   INDEX UNIQUE SCAN    | ORDER_PK            |    269 |      1 |          |    269 |00:00:00.01 |      21 |               
-------------------------------------------------------------------------------------------------------------------               
 
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: 00:00:00.02
SQL> 
--檢視動態檢視
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

我們發現以後的執行都會變成軟解析，使用第二次產生的執行計劃。
通過CFB功能使優化器能夠在以後的執行中選擇更優的執行計劃，從得到更好的執行效率。

CFB的處理流程

下面通過以下流程圖來總體的回顧一下CFB的處理過程。

在下列情況CBO可能無法估算出準確的Cardinality，Oracle會啟用CFB功能：

?沒有收集表的統計資訊，並且dynamic sampling 也沒有開啟；

? 一個表的查詢條件涉及多列，但卻沒有收集擴充套件的統計資訊(extended statistics)

? 查詢條件複雜（比如條件有函式）

針對上述情況，Oracle會採取如下的CFB流程處理：

1. SQL文第一次執行時，Oracle會監控操作的實際行數（A-Row），然後對比CBO估算的行數（E-Row）。

2. 如果兩個值相差很大，就記錄實際行數（A-Row），做上標記。

下次執行時再次進行硬解析，根據實際行數來重新生成執行計劃。

3. 如果兩個值相差不大，CBO就不再監控這條SQL語句。

Oracle優化器之基數反饋（CardinalityFeedback）功能

概述

例子1（CFB無效）

例子2（CFB有效）

CFB的處理流程

Oracle優化器之基數反饋（CardinalityFeedback）功能

優化演算法之牛頓法（轉）

Python黑帽程式設計 4.1 Sniffer(嗅探器)之資料捕獲（上）

Oracle資料庫 —— SQL之select語句（2） where子句和資料排序

Oracle優化——星型聯結轉換（star_transformation_enabled）

Oracle優化器、優化模式、表的連線方式（Hash Join、Nested Loop、Sort Merge Join）

Oracle之SQL優化-索引的基本原理（二）

Oracle優化器：星型轉換（Star Query Transformation ）

安卓MP3播放器開發實例（3）之進度條和歌詞更新的實現

Java之集合初探（二）Iterator（叠代器），collections，打包/解包（裝箱拆箱），泛型(Generic)，comparable接口

百度編輯器（ueditor）@功能之獲取坐標

器學習數學基礎之矩陣理論（二）

racadm環境之服務器BIOS設置（三）

Oracle 數據庫體系結構（二）：服務器結構

UVM暫存器篇之三：暫存器模型的整合（上）

UVM暫存器篇之五：暫存器模型的整合（下）

UVM暫存器篇之四：暫存器模型的整合（中）

Oracle EBS 清除併發請求和（或）管理器資料請求

Java設計模式之外觀模式和裝飾器模式的設計（精選）

跟我學 Java 8 新特性之 Stream 流（七）流與迭代器，流系列大結局

Oracle優化器之基數反饋（CardinalityFeedback）功能

概述

例子1（CFB無效）

例子2（CFB有效）

CFB的處理流程

相關推薦