在ORACLE裡，一般認為”與NULL是等價的，除了一些特別的語法，比如有id is null,但是沒有id is ”，以及”作為字元型別的特殊性，在decode等函式要求型別匹配的時候，與NULL可能不等價。本篇文章主要說明下”與NULL在SQL最佳化器中也有很大的區別，如果不瞭解這個區別，很可能SQL語句的效能，將要陷入災難境地。此問題來源於：http://www.itpub.net/thread-1838241-1-1.html 討論

1.示例分析

create table t1 as select * from dba_objects ;
create table t2 as select * from dba_objects ;
create table t3 as select * from dba_objects ;
/
begin
for xx in 1 .. 5 loop
insert into t1 select * from t1;
insert into t2 select * from t2;
insert into t3 select * from t3;
commit;
end loop;
end;
/
create index i1 on t1(object_name);
create index i2 on t2(object_name);
create index i3 on t3(object_name);
create index i4 on t1(subobject_name);
create index i5 on t2(subobject_name);
create index i6 on t3(subobject_name);

–收集統計資訊省略

針對下列語句，ORACLE對t2,t3走了FULL TABLE SCAN:

dingjun123@ORADB> show rel
release 1102000100

dingjun123@ORADB> set autotrace traceonly exp
dingjun123@ORADB> select *
2 from (select t1.subobject_name, t1.object_name, t1.object_type
3 from t1
4 union all
5 select ”, t2.object_name, t2.object_type
6 from t2
7 union all
8 select t3.subobject_name, ”, t3.object_type from t3) t
9 where (t.subobject_name = ‘T’ or t.object_name = ‘T’);
Elapsed: 00:00:00.05

Execution Plan
———————————————————-
Plan hash value: 3458763867
——————————————————————————————-
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
——————————————————————————————-
| 0 | SELECT STATEMENT | | 62890 | 5773K| 18934 (1)| 00:03:48 |
| 1 | VIEW | | 62890 | 5773K| 18934 (1)| 00:03:48 |
| 2 | UNION-ALL | | | | | |
| 3 | TABLE ACCESS BY INDEX ROWID | T1 | 12678 | 433K| 38 (0)| 00:00:01 |
| 4 | BITMAP CONVERSION TO ROWIDS | | | | | |
| 5 | BITMAP OR | | | | | |
| 6 | BITMAP CONVERSION FROM ROWIDS| | | | | |
|* 7 | INDEX RANGE SCAN | I4 | | | 1 (0)| 00:00:01 |
| 8 | BITMAP CONVERSION FROM ROWIDS| | | | | |
|* 9 | INDEX RANGE SCAN | I1 | | | 3 (0)| 00:00:01 |
|* 10 | TABLE ACCESS FULL | T2 | 24167 | 778K| 9429 (1)| 00:01:54 |
|* 11 | TABLE ACCESS FULL | T3 | 26045 | 279K| 9466 (1)| 00:01:54 |
——————————————————————————————-
Predicate Information (identified by operation id):
—————————————————
7 – access(“T1″.”SUBOBJECT_NAME”=’T')
9 – access(“T1″.”OBJECT_NAME”=’T')
10 – filter(”=’T’ OR “T2″.”OBJECT_NAME”=’T')
11 – filter(”=’T’ OR “T3″.”SUBOBJECT_NAME”=’T')

看錶t2、t3對應謂詞的實際基數情況:

dingjun123@ORADB> SELECT COUNT(*) FROM t2 WHERE ”=’T’ OR “T2″.”OBJECT_NAME”=’T';
COUNT(*)
———-
128
1 row selected.
Elapsed: 00:00:11.50

dingjun123@ORADB> SELECT COUNT(*) FROM t3 WHERE ”=’T’ OR “T3″.”SUBOBJECT_NAME”=’T';
COUNT(*)
———-
0
1 row selected.
Elapsed: 00:00:14.85

CBO最佳化器估算的對應謂詞的基數與實際的技術差別如下：

表名	估算基數	實際基數	差別倍數
T2	24167	128	188.80
T3	26045	0	26045

很顯然，實際基數與估算的基數差別太大，從而最佳化器選擇了錯誤的執行路徑，正確的執行路徑應該是走索引的。究其原因，可以分析下，

dingjun123@ORADB> SELECT COUNT(*) FROM t2 WHERE ”T2″.”OBJECT_NAME”=’T';
Elapsed: 00:00:00.00
Execution Plan
———————————————————-
Plan hash value: 2583336616
————————————————————————–
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
————————————————————————–
| 0 | SELECT STATEMENT | | 1 | 24 | 3 (0)| 00:00:01 |
| 1 | SORT AGGREGATE | | 1 | 24 | | |
|* 2 | INDEX RANGE SCAN| I2 | 52 | 1248 | 3 (0)| 00:00:01 |
————————————————————————–
Predicate Information (identified by operation id):
—————————————————
2 – access(“T2″.”OBJECT_NAME”=’T')

dingjun123@ORADB> SELECT COUNT(*) FROM t2 WHERE ”=’T';
Elapsed: 00:00:00.00
Execution Plan
———————————————————-
Plan hash value: 402395414
——————————————————————–
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
——————————————————————–
| 0 | SELECT STATEMENT | | 1 | 9411 (1)| 00:01:53 |
| 1 | SORT AGGREGATE | | 1 | | |
|* 2 | FILTER | | | | |
| 3 | TABLE ACCESS FULL| T2 | 2411K| 9411 (1)| 00:01:53 |
——————————————————————–
Predicate Information (identified by operation id):
—————————————————
2 – filter(”=’T')

很顯然，是由於謂詞”=’T'造成了最佳化器的估算錯誤，”=’T'最佳化器估算100%的選擇性，透過OR一合併，針對表T2的最終基數是24167，因此，最佳化器選擇全表掃描。這很顯然是錯誤的，”=’T',走FILTER應該轉為NULL IS NOT NULL的形式，最終此分支根本無需計算才對。下面換成NULL測試：

dingjun123@ORADB> SELECT COUNT(*) FROM t2 WHERE NULL=’T';
Elapsed: 00:00:00.00
Execution Plan
———————————————————-
Plan hash value: 402395414
——————————————————————–
| Id | Operation | Name | Rows | Cost (%CPU)| Time |
——————————————————————–
| 0 | SELECT STATEMENT | | 1 | 0 (0)| |
| 1 | SORT AGGREGATE | | 1 | | |
|* 2 | FILTER | | | | |
| 3 | TABLE ACCESS FULL| T2 | 2411K| 9411 (1)| 00:01:53 |
——————————————————————–
Predicate Information (identified by operation id):
—————————————————
2 – filter(NULL IS NOT NULL)

換成NULL=’T'後，FILTER自動轉為NULL IS NOT NULL，顯然，這是個永遠不成立的條件，所以，根本無須執行ID=3的操作，最終ID=0的結果COST=0，針對這種FILTER單分支的執行計劃注意，雖然子步驟3COST=9411，但是實際可能沒有執行，詳細的計劃可以用DBMS_XPLAN.DISPLAY_CURSOR檢視：

dingjun123@ORADB> alter session set statistics_level=all;
Session altered.
Elapsed: 00:00:00.04

dingjun123@ORADB> SELECT COUNT(*) FROM t2 WHERE NULL=’T';
COUNT(*)
———-
0
1 row selected.
Elapsed: 00:00:00.05

dingjun123@ORADB> @display_cursor
PLAN_TABLE_OUTPUT
————————————————————————————————–
SQL_ID f33mr91wyuq03, child number 0
————————————-
SELECT COUNT(*) FROM t2 WHERE NULL=’T’
Plan hash value: 402395414
—————————————————————————-
| Id | Operation | Name | Starts | E-Rows | A-Rows | A-Time |
—————————————————————————-
| 0 | SELECT STATEMENT | | 1 | | 1 |00:00:00.01 |
| 1 | SORT AGGREGATE | | 1 | 1 | 1 |00:00:00.01 |
|* 2 | FILTER | | 1 | | 0 |00:00:00.01 |
| 3 | TABLE ACCESS FULL| T2 | 0 | 2411K| 0 |00:00:00.01 |
—————————————————————————-
Predicate Information (identified by operation id):
————————————————–
2 – filter(NULL IS NOT NULL)

20 rows selected.
Elapsed: 00:00:00.74

透過ID=3步驟的Starts=0,很清晰地看出，最終未執行ID=3的步驟。從這點上說，最佳化器針對”與NULL的相關計算方式還是有很大區別。

2.問題解決
透過以上分析，將”改為NULL，問題得到有效解決或者把謂詞寫到每個子SQL裡面。當然，只要是非”的都是可以的，比如’ ‘(空格）。

dingjun123@ORADB> SELECT *
2 FROM (SELECT t1.subobject_name, t1.object_name, t1.object_type
3 FROM t1
4 UNION ALL
5 SELECT NULL, t2.object_name, t2.object_type
6 FROM t2
7 UNION ALL
8 SELECT t3.subobject_name, NULL, t3.object_type FROM t3) t
9 WHERE (t.subobject_name = ‘T’ OR t.object_name = ‘T’);

256 rows selected.
Elapsed: 00:00:00.72

Execution Plan
———————————————————-
Plan hash value: 3324794093
——————————————————————————————-
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
——————————————————————————————-
| 0 | SELECT STATEMENT | | 12746 | 1170K| 92 (0)| 00:00:02 |
| 1 | VIEW | | 12746 | 1170K| 92 (0)| 00:00:02 |
| 2 | UNION-ALL | | | | | |
| 3 | TABLE ACCESS BY INDEX ROWID | T1 | 12678 | 433K| 38 (0)| 00:00:01 |
| 4 | BITMAP CONVERSION TO ROWIDS | | | | | |
| 5 | BITMAP OR | | | | | |
| 6 | BITMAP CONVERSION FROM ROWIDS| | | | | |
|* 7 | INDEX RANGE SCAN | I4 | | | 1 (0)| 00:00:01 |
| 8 | BITMAP CONVERSION FROM ROWIDS| | | | | |
|* 9 | INDEX RANGE SCAN | I1 | | | 3 (0)| 00:00:01 |
| 10 | TABLE ACCESS BY INDEX ROWID | T2 | 52 | 1716 | 41 (0)| 00:00:01 |
|* 11 | INDEX RANGE SCAN | I2 | 52 | | 3 (0)| 00:00:01 |
| 12 | TABLE ACCESS BY INDEX ROWID | T3 | 16 | 176 | 13 (0)| 00:00:01 |
|* 13 | INDEX RANGE SCAN | I6 | 16 | | 1 (0)| 00:00:01 |
——————————————————————————————-

Predicate Information (identified by operation id):
—————————————————
7 – access(“T1″.”SUBOBJECT_NAME”=’T')
9 – access(“T1″.”OBJECT_NAME”=’T')
11 – access(“T2″.”OBJECT_NAME”=’T')
13 – access(“T3″.”SUBOBJECT_NAME”=’T')

Statistics
———————————————————-
1 recursive calls
0 db block gets
241 consistent gets
99 physical reads
0 redo size
4006 bytes sent via SQL*Net to client
602 bytes received via SQL*Net from client
19 SQL*Net roundtrips to/from client
0 sorts (memory)
0 sorts (disk)
256 rows processed

ORACLE最佳化器雖然強大，但是最佳化器要判斷的東西實在太多太複雜，導致ORACLE強大的最佳化器也不一定是很完善的，有各種各樣的BUG，或未完善的地方，值得我們注意，但是，相信各種新的版本上，會有更多驚喜的特性。
注：以上問題，在RBO中同樣存在。

【新炬網路名師大講堂】CBO中”與NULL在cardinality計算上的差別

相關文章