恆等查詢條件改變執行計劃——SQL優化之Everything is Possible

有的時候開發人員為了方便會在WHERE語句後面新增一個1=1，這樣在處理頁面傳入的條件是就可用不用判斷直接新增AND 條件。

一直認為新增一個恆等條件，不會對Oracle的查詢造成什麼影響，最多不過影響一下Oracle的效能，但是今天突然發現，這個恆等的查詢條件居然可以影響Oracle的執行計劃。

首先看一個簡單的例子：

SQL> CREATE TABLE T1 AS SELECT * FROM DBA_OBJECTS
2 WHERE OWNER = 'SYS'
3 AND OBJECT_TYPE NOT LIKE '%BODY'
4 AND OBJECT_TYPE NOT LIKE 'JAVA%';

Table created.

SQL> CREATE TABLE T2 AS SELECT * FROM DBA_SEGMENTS WHERE OWNER = 'SYS';

Table created.

SQL> CREATE TABLE T3 AS SELECT * FROM DBA_INDEXES WHERE OWNER = 'SYS';

Table created.

SQL> ALTER TABLE T1 ADD CONSTRAINT PK_T1 PRIMARY KEY (OBJECT_NAME);

Table altered.

SQL> CREATE INDEX IND_T2_SEGNAME ON T2(SEGMENT_NAME);

Index created.

SQL> CREATE INDEX IND_T3_TABNAME ON T3(TABLE_NAME);

Index created.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'T1', METHOD_OPT => 'FOR ALL INDEXED COLUMNS SIZE 100', CASCADE => TRUE)

PL/SQL procedure successfully completed.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'T2', METHOD_OPT => 'FOR ALL INDEXED COLUMNS SIZE 100', CASCADE => TRUE)

PL/SQL procedure successfully completed.

SQL> EXEC DBMS_STATS.GATHER_TABLE_STATS(USER, 'T3', METHOD_OPT => 'FOR ALL INDEXED COLUMNS SIZE 100', CASCADE => TRUE)

PL/SQL procedure successfully completed.

SQL> SET AUTOT TRACE EXP
SQL> SELECT T1.OBJECT_NAME, T1.OBJECT_TYPE, T2.TABLESPACE_NAME
2 FROM T1, T2
3 WHERE T1.OBJECT_NAME = T2.SEGMENT_NAME
4 AND NOT EXISTS
5 (
6 SELECT 1 FROM T3
7 WHERE T3.INDEX_NAME = T1.OBJECT_NAME
8 AND T3.TABLE_NAME = 'OBJ$'
9 )
10 ;

Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=12 Card=668 Bytes=58784)
1 0 HASH JOIN (ANTI) (Cost=12 Card=668 Bytes=58784)
2 1 HASH JOIN (Cost=9 Card=668 Bytes=39412)
3 2 TABLE ACCESS (FULL) OF 'T2' (Cost=2 Card=668 Bytes=21376)
4 2 TABLE ACCESS (FULL) OF 'T1' (Cost=6 Card=3806 Bytes=102762)
5 1 TABLE ACCESS (BY INDEX ROWID) OF 'T3' (Cost=2 Card=2 Bytes=58)
6 5 INDEX (RANGE SCAN) OF 'IND_T3_TABNAME' (NON-UNIQUE) (Cost=1 Card=2)

SQL> SELECT T1.OBJECT_NAME, T1.OBJECT_TYPE, T2.TABLESPACE_NAME
2 FROM T1, T2
3 WHERE T1.OBJECT_NAME = T2.SEGMENT_NAME
4 AND NOT EXISTS
5 (
6 SELECT 1 FROM T3
7 WHERE 1 = 1
8 AND T3.INDEX_NAME = T1.OBJECT_NAME
9 AND T3.TABLE_NAME = 'OBJ$'
10 )
11 ;

Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=12 Card=668 Bytes=50768)
1 0 HASH JOIN (ANTI) (Cost=12 Card=668 Bytes=50768)
2 1 HASH JOIN (Cost=9 Card=668 Bytes=39412)
3 2 TABLE ACCESS (FULL) OF 'T2' (Cost=2 Card=668 Bytes=21376)
4 2 TABLE ACCESS (FULL) OF 'T1' (Cost=6 Card=3806 Bytes=102762)
5 1 VIEW OF 'VW_SQ_1' (Cost=2 Card=2 Bytes=34)
6 5 TABLE ACCESS (BY INDEX ROWID) OF 'T3' (Cost=2 Card=2 Bytes=58)
7 6 INDEX (RANGE SCAN) OF 'IND_T3_TABNAME' (NON-UNIQUE) (Cost=1 Card=2)

觀察兩個查詢語句會發現，兩個查詢語句唯一的區別就是第二個查詢語句中的NOT EXISTS子查詢中包含了一個恆等查詢條件1 = 1。

僅僅是這一點的區別，卻造成了兩個SQL語句的執行計劃出現了差異。對於第二個查詢Oracle居然多生成了一個臨時的VIEW。也許有人認為這個執行計劃沒有太大的區別，基本上可以認為等價。但是由於多生成了一個VIEW的步驟，必然會造成效能的下降，更為關鍵的是：這說明Oracle認為兩個SQL語句是不同的，而且處理方式也是不同的。

如果一個簡單的VIEW步驟還不能說明什麼問題的話，那麼看看下面這個在實際執行中碰到的問題：

SQL> SET AUTOT TRACE EXP
SQL> SELECT A.ID,
2 B.NAME_CHN,
3 B.SPEC,
4 A.MANUFACTURE_ID
5 FROM CAT_PRODUCT A,
6 CAT_DRUG B
7 WHERE A.MEDICAL_ID = B.ID
8 AND A.CHECK_FLAG = 1
9 AND NOT EXISTS
10 (
11 SELECT 1
12 FROM PROJECT_SUPPLY_PRODUCT C
13 WHERE C.DATA_PRODUCT = A.ID
14 AND C.ENABLE_FLAG = '1'
15 AND C.PROJECT_ID = 'MRBR00000000000000709824'
16 )
17 ;
Elapsed: 00:00:00.01

Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=483 Card=1 Bytes=213)
1 0 NESTED LOOPS (ANTI) (Cost=483 Card=1 Bytes=213)
2 1 NESTED LOOPS (Cost=481 Card=1 Bytes=160)
3 2 VIEW OF 'index$_join$_001' (Cost=480 Card=1 Bytes=77)
4 3 HASH JOIN
5 4 HASH JOIN
6 5 INDEX (FAST FULL SCAN) OF 'PK_CAT_PRODUCT' (UNIQUE) (Cost=33 Card=1 Bytes=77)
7 5 INDEX (FAST FULL SCAN) OF 'TU_CAT_PRODUCT_MED_CHECK' (NON-UNIQUE) (Cost=33 Card=1 Bytes=77)
8 4 INDEX (FAST FULL SCAN) OF 'TU_CAT_PRODUCT_MANUFACT_ID' (NON-UNIQUE) (Cost=33 Card=1 Bytes=77)
9 2 TABLE ACCESS (BY INDEX ROWID) OF 'CAT_DRUG' (Cost=1 Card=1 Bytes=83)
10 9 INDEX (UNIQUE SCAN) OF 'PK_CAT_DRUG' (UNIQUE)
11 1 TABLE ACCESS (BY INDEX ROWID) OF 'PROJECT_SUPPLY_PRODUCT' (Cost=2 Card=1 Bytes=53)
12 11 INDEX (RANGE SCAN) OF 'IDX_PROJECT_SUPPLY_PRODUCT_PJ' (NON-UNIQUE) (Cost=1 Card=1)

SQL> SELECT A.ID,
2 B.NAME_CHN,
3 B.SPEC,
4 A.MANUFACTURE_ID
5 FROM CAT_PRODUCT A,
6 CAT_DRUG B
7 WHERE A.MEDICAL_ID = B.ID
8 AND A.CHECK_FLAG = 1
9 AND NOT EXISTS
10 (
11 SELECT 1
12 FROM PROJECT_SUPPLY_PRODUCT C
13 WHERE 1 = 1
14 AND C.DATA_PRODUCT = A.ID
15 AND C.ENABLE_FLAG = '1'
16 AND C.PROJECT_ID = 'MRBR00000000000000709824'
17 )
18 ;
Elapsed: 00:00:00.00

Execution Plan
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=CHOOSE (Cost=484 Card=1 Bytes=186)
1 0 HASH JOIN (ANTI) (Cost=484 Card=1 Bytes=186)
2 1 NESTED LOOPS (Cost=481 Card=1 Bytes=160)
3 2 VIEW OF 'index$_join$_001' (Cost=480 Card=1 Bytes=77)
4 3 HASH JOIN
5 4 HASH JOIN
6 5 INDEX (FAST FULL SCAN) OF 'PK_CAT_PRODUCT' (UNIQUE) (Cost=33 Card=1 Bytes=77)
7 5 INDEX (FAST FULL SCAN) OF 'TU_CAT_PRODUCT_MED_CHECK' (NON-UNIQUE) (Cost=33 Card=1 Bytes=77)
8 4 INDEX (FAST FULL SCAN) OF 'TU_CAT_PRODUCT_MANUFACT_ID' (NON-UNIQUE) (Cost=33 Card=1 Bytes=77)
9 2 TABLE ACCESS (BY INDEX ROWID) OF 'CAT_DRUG' (Cost=1 Card=1 Bytes=83)
10 9 INDEX (UNIQUE SCAN) OF 'PK_CAT_DRUG' (UNIQUE)
11 1 VIEW OF 'VW_SQ_1' (Cost=2 Card=1 Bytes=26)
12 11 TABLE ACCESS (BY INDEX ROWID) OF 'PROJECT_SUPPLY_PRODUCT' (Cost=2 Card=1 Bytes=53)
13 12 INDEX (RANGE SCAN) OF 'IDX_PROJECT_SUPPLY_PRODUCT_PJ' (NON-UNIQUE) (Cost=1 Card=1)

SQL語句仍然是隻相差了NOT EXISTS語句中的一個1 = 1的恆等條件，但是如果不加這個恆等條件，Oracle用的是NESTED LOOP ANTI，而加上了這個恆等條件，則Oracle選擇了HASH JOIN ANTI。

這次執行計劃的改變足以使SQL的執行時間發生成百上千倍的變化。

一個小小的恆等查詢條件居然可以使查詢計劃發生變化，這是我從來沒有想到的。

通過這個問題，我總結了如下三點：

要敢於懷疑任何事情，沒有什麼是不可能的，一切以實踐的結果為準；

不要在SQL上新增一些沒有必要的小零碎，一些看似無害的東西在一些情況下可能會引發嚴重的問題。

建立良好的編碼風格，以及嚴格SQL語句的編寫制度的必要性。

來自 “ ITPUB部落格 ” ，連結：http://blog.itpub.net/4227/viewspace-69130/，如需轉載，請註明出處，否則將追究法律責任。