CBO與RBO下的IN/EXISTS
晚上抽空看了看ask tom的RSS,發現兩篇應該說很入門的關於IN/EXISTS的文章:http://asktom.oracle.com/pls/ask/f?p=4950:8:::::F4950_P8_DISPLAYID:953229842074,http://asktom.oracle.com/pls/ask/f?p=4950:8:::::F4950_P8_DISPLAYID:442029737684
文章很長,也沒有仔細看完,不過有些東西還是很有意思的,動手實驗了一下。大家隨便看看咯,Tom當初回答問題的環境我已經沒辦法測試了,在10.1.0.4下做了些測試,CBO和RBO(使用hints)下還是很大區別的。
SQL> select * from scott.emp;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7499 ALLEN SALESMAN 7698 20-2月 -81 1600 300 30
7521 WARD SALESMAN 7698 22-2月 -81 1250 500 30
7566 JONES MANAGER 7839 02-4月 -81 2975 20
7654 MARTIN SALESMAN 7698 28-9月 -81 1250 1400 30
7698 BLAKE MANAGER 7839 01-5月 -81 2850 30
7782 CLARK MANAGER 7839 09-6月 -81 2450 10
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7839 KING PRESIDENT 17-11月-81 5000 10
7844 TURNER SALESMAN 7698 08-9月 -81 1500 0 30
7876 ADAMS CLERK 7788 23-5月 -87 1100 20
7900 JAMES CLERK 7698 03-12月-81 950 30
7902 FORD ANALYST 7566 03-12月-81 3000 20
7934 MILLER CLERK 7782 23-1月 -82 1300 10
已選擇14行。
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=3 Card=14 Bytes=1218)
1 0 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=1218)
SQL> create table tmp_emp as select * from scott.emp where ename like 'S%';
表已創建。
SQL> select * from tmp_emp;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=3 Card=2 Bytes=174)
1 0 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=174)
先測試一下IN:
SQL> select * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=7 Card=2 Bytes=188)
1 0 HASH JOIN (SEMI) (Cost=7 Card=2 Bytes=188)
2 1 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=174)
3 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=98)
表沒有做過分析,使用了HASH JOIN (SEMI)。
Tom在回答問題的時候提到:
Select * from T1 where x in ( select y from T2 )
is typically processed as:
select *
from t1, ( select distinct y from t2 ) t2
where t1.x = t2.y;
The subquery is evaluated, distincted, indexed (or hashed or sorted) and then
joined to the original table -- typically.
顯然第二個查詢是很奇怪的。也正是這個促使我打開了數據庫測試,難道一個IN還需要先DISTINCT一下?沒見過IN產生排序操作啊。
SQL> ed
已寫入 file afiedt.buf
1 select * from tmp_emp,(select distinct ename from scott.emp) t
2* where tmp_emp.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=8 Card=2 Bytes=188)
1 0 VIEW (Cost=8 Card=2 Bytes=188)
2 1 SORT (UNIQUE) (Cost=8 Card=2 Bytes=202)
3 2 HASH JOIN (Cost=7 Card=2 Bytes=202)
4 3 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=188)
5 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=98)
顯然不一樣了。那麼改用RULE模式:
SQL> ed
已寫入 file afiedt.buf
1 select /*+ rule */ * from tmp_emp,(select distinct ename from scott.emp) t
2* where tmp_emp.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 MERGE JOIN
2 1 VIEW
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 SORT (JOIN)
6 5 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE)
再試試加上索引:
SQL> alter table tmp_emp add constraints pk_tmpemp primary key(ename);
表已更改。
分析一下:
SQL> analyze table tmp_emp compute statistics for table for all indexed columns;
表已分析。
SQL> analyze table scott.emp compute statistics for table for all columns;
表已分析。
SQL> select *
2 from tmp_emp e,(select distinct ename from scott.emp) t
3 where e.ename=t.ename;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=6 Card=2 Bytes=188)
1 0 VIEW (Cost=6 Card=2 Bytes=188)
2 1 SORT (UNIQUE) (Cost=6 Card=2 Bytes=88)
3 2 NESTED LOOPS (Cost=5 Card=2 Bytes=88)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
5 3 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
SQL> ed
已寫入 file afiedt.buf
1 select /*+rule*/ *
2 from tmp_emp e,(select distinct ename from scott.emp) t
3* where e.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 NESTED LOOPS
2 1 VIEW
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE))
SQL> select * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=1 Bytes=44)
1 0 NESTED LOOPS (Cost=5 Card=1 Bytes=44)
2 1 SORT (UNIQUE) (Cost=3 Card=14 Bytes=70)
3 2 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
4 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
5 4 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
SQL> select /*+rule*/ * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 NESTED LOOPS
2 1 VIEW OF 'VW_NSO_1' (VIEW)
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE))
這次確實接近了很多,總體上看和Tom說的情況差不多。不過RBO這種模式對後邊的表對應列選擇性低時應該很好,而其他情況恐怕不見得是優化的。下面看看EXISTS:
SQL> select * from tmp_emp t where exists(select null from scott.emp e where t.ename=e.ename);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=1 Bytes=44)
1 0 NESTED LOOPS (Cost=5 Card=1 Bytes=44)
2 1 SORT (UNIQUE) (Cost=3 Card=14 Bytes=70)
3 2 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
4 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
5 4 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
看着和用IN一樣哦。
SQL> select /*+rule*/ * from tmp_emp t where exists(select null from scott.emp e where t.ename=e.ename);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 FILTER
2 1 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE)
3 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
用RBO就不同咯!
SQL> select t.* from tmp_emp t,scott.emp e where t.ename=e.ename;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=2 Bytes=88)
1 0 NESTED LOOPS (Cost=5 Card=2 Bytes=88)
2 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
3 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
4 3 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
這樣看來在CBO下,使用內關聯、IN、EXISTS很可能得到同一個執行計劃(更多的情況就不測試了),優化器會發現三者的語義是相同的;而在較早的RBO下EXISTS採用FILTER而IN相當於對子查詢先DISTINCT後關聯,內關聯則是直接關聯就行了。
Tom在他最新的回覆中這樣說:
Use the RBO and see what you see. way back when I wrote this, that was the "more popular" of the two perhaps
today in 2005, what I said years ago using the RBO does not apply to the CBO. the cbo is smart enough to recognize these two things are effectively the same.
IN相當於對子查詢先DISTINCT後關聯這一條真的沒有想通,RBO爲什麼做這樣的事呢?實在是沒有普適性,我個人的理解就是設計的時候認爲IN後面是跟一個值列表的情況居多,當然先把值算出來,然後NL就可以了,如果後面是一個表裏的值那也就一樣處理;而且使用IN的時候也許大多是子查詢對外層查詢的篩選性高,即外層的表較大,而子查詢的返回值較少。看看不同情況的不同執行計劃,Oracle在CBO上確實還是花了點心思的,赫赫。
結論:在RBO下,使用IN還是EXISTS需要視情況而定,只要記住使用IN存在排序和DISTINCT這一步驟應該就不難判斷;CBO下優化器會爲你選擇,怎麼寫就只是習慣問題了。
文章很長,也沒有仔細看完,不過有些東西還是很有意思的,動手實驗了一下。大家隨便看看咯,Tom當初回答問題的環境我已經沒辦法測試了,在10.1.0.4下做了些測試,CBO和RBO(使用hints)下還是很大區別的。
SQL> select * from scott.emp;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7499 ALLEN SALESMAN 7698 20-2月 -81 1600 300 30
7521 WARD SALESMAN 7698 22-2月 -81 1250 500 30
7566 JONES MANAGER 7839 02-4月 -81 2975 20
7654 MARTIN SALESMAN 7698 28-9月 -81 1250 1400 30
7698 BLAKE MANAGER 7839 01-5月 -81 2850 30
7782 CLARK MANAGER 7839 09-6月 -81 2450 10
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7839 KING PRESIDENT 17-11月-81 5000 10
7844 TURNER SALESMAN 7698 08-9月 -81 1500 0 30
7876 ADAMS CLERK 7788 23-5月 -87 1100 20
7900 JAMES CLERK 7698 03-12月-81 950 30
7902 FORD ANALYST 7566 03-12月-81 3000 20
7934 MILLER CLERK 7782 23-1月 -82 1300 10
已選擇14行。
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=3 Card=14 Bytes=1218)
1 0 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=1218)
SQL> create table tmp_emp as select * from scott.emp where ename like 'S%';
表已創建。
SQL> select * from tmp_emp;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=3 Card=2 Bytes=174)
1 0 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=174)
先測試一下IN:
SQL> select * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=7 Card=2 Bytes=188)
1 0 HASH JOIN (SEMI) (Cost=7 Card=2 Bytes=188)
2 1 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=174)
3 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=98)
表沒有做過分析,使用了HASH JOIN (SEMI)。
Tom在回答問題的時候提到:
Select * from T1 where x in ( select y from T2 )
is typically processed as:
select *
from t1, ( select distinct y from t2 ) t2
where t1.x = t2.y;
The subquery is evaluated, distincted, indexed (or hashed or sorted) and then
joined to the original table -- typically.
顯然第二個查詢是很奇怪的。也正是這個促使我打開了數據庫測試,難道一個IN還需要先DISTINCT一下?沒見過IN產生排序操作啊。
SQL> ed
已寫入 file afiedt.buf
1 select * from tmp_emp,(select distinct ename from scott.emp) t
2* where tmp_emp.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=8 Card=2 Bytes=188)
1 0 VIEW (Cost=8 Card=2 Bytes=188)
2 1 SORT (UNIQUE) (Cost=8 Card=2 Bytes=202)
3 2 HASH JOIN (Cost=7 Card=2 Bytes=202)
4 3 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE) (Cost=3 Card=2 Bytes=188)
5 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=98)
顯然不一樣了。那麼改用RULE模式:
SQL> ed
已寫入 file afiedt.buf
1 select /*+ rule */ * from tmp_emp,(select distinct ename from scott.emp) t
2* where tmp_emp.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 MERGE JOIN
2 1 VIEW
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 SORT (JOIN)
6 5 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE)
再試試加上索引:
SQL> alter table tmp_emp add constraints pk_tmpemp primary key(ename);
表已更改。
分析一下:
SQL> analyze table tmp_emp compute statistics for table for all indexed columns;
表已分析。
SQL> analyze table scott.emp compute statistics for table for all columns;
表已分析。
SQL> select *
2 from tmp_emp e,(select distinct ename from scott.emp) t
3 where e.ename=t.ename;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=6 Card=2 Bytes=188)
1 0 VIEW (Cost=6 Card=2 Bytes=188)
2 1 SORT (UNIQUE) (Cost=6 Card=2 Bytes=88)
3 2 NESTED LOOPS (Cost=5 Card=2 Bytes=88)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
5 3 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
SQL> ed
已寫入 file afiedt.buf
1 select /*+rule*/ *
2 from tmp_emp e,(select distinct ename from scott.emp) t
3* where e.ename=t.ename
SQL> /
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO ENAME
---------- ---------- --------- ---------- -------------- ---------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20 SCOTT
7369 SMITH CLERK 7902 17-12月-80 800 20 SMITH
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 NESTED LOOPS
2 1 VIEW
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE))
SQL> select * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=1 Bytes=44)
1 0 NESTED LOOPS (Cost=5 Card=1 Bytes=44)
2 1 SORT (UNIQUE) (Cost=3 Card=14 Bytes=70)
3 2 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
4 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
5 4 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
SQL> select /*+rule*/ * from tmp_emp where ename in (select ename from scott.emp);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 NESTED LOOPS
2 1 VIEW OF 'VW_NSO_1' (VIEW)
3 2 SORT (UNIQUE)
4 3 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
5 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE)
6 5 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE))
這次確實接近了很多,總體上看和Tom說的情況差不多。不過RBO這種模式對後邊的表對應列選擇性低時應該很好,而其他情況恐怕不見得是優化的。下面看看EXISTS:
SQL> select * from tmp_emp t where exists(select null from scott.emp e where t.ename=e.ename);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
7369 SMITH CLERK 7902 17-12月-80 800 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=1 Bytes=44)
1 0 NESTED LOOPS (Cost=5 Card=1 Bytes=44)
2 1 SORT (UNIQUE) (Cost=3 Card=14 Bytes=70)
3 2 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
4 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
5 4 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
看着和用IN一樣哦。
SQL> select /*+rule*/ * from tmp_emp t where exists(select null from scott.emp e where t.ename=e.ename);
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=HINT: RULE
1 0 FILTER
2 1 TABLE ACCESS (FULL) OF 'TMP_EMP' (TABLE)
3 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE)
用RBO就不同咯!
SQL> select t.* from tmp_emp t,scott.emp e where t.ename=e.ename;
EMPNO ENAME JOB MGR HIREDATE SAL COMM DEPTNO
---------- ---------- --------- ---------- -------------- ---------- ---------- ----------
7369 SMITH CLERK 7902 17-12月-80 800 20
7788 SCOTT ANALYST 7566 19-4月 -87 3000 20
執行計劃
----------------------------------------------------------
0 SELECT STATEMENT Optimizer=ALL_ROWS (Cost=5 Card=2 Bytes=88)
1 0 NESTED LOOPS (Cost=5 Card=2 Bytes=88)
2 1 TABLE ACCESS (FULL) OF 'EMP' (TABLE) (Cost=3 Card=14 Bytes=70)
3 1 TABLE ACCESS (BY INDEX ROWID) OF 'TMP_EMP' (TABLE) (Cost=1 Card=1 Bytes=39)
4 3 INDEX (UNIQUE SCAN) OF 'PK_TMPEMP' (INDEX (UNIQUE)) (Cost=0 Card=1)
這樣看來在CBO下,使用內關聯、IN、EXISTS很可能得到同一個執行計劃(更多的情況就不測試了),優化器會發現三者的語義是相同的;而在較早的RBO下EXISTS採用FILTER而IN相當於對子查詢先DISTINCT後關聯,內關聯則是直接關聯就行了。
Tom在他最新的回覆中這樣說:
Use the RBO and see what you see. way back when I wrote this, that was the "more popular" of the two perhaps
today in 2005, what I said years ago using the RBO does not apply to the CBO. the cbo is smart enough to recognize these two things are effectively the same.
IN相當於對子查詢先DISTINCT後關聯這一條真的沒有想通,RBO爲什麼做這樣的事呢?實在是沒有普適性,我個人的理解就是設計的時候認爲IN後面是跟一個值列表的情況居多,當然先把值算出來,然後NL就可以了,如果後面是一個表裏的值那也就一樣處理;而且使用IN的時候也許大多是子查詢對外層查詢的篩選性高,即外層的表較大,而子查詢的返回值較少。看看不同情況的不同執行計劃,Oracle在CBO上確實還是花了點心思的,赫赫。
結論:在RBO下,使用IN還是EXISTS需要視情況而定,只要記住使用IN存在排序和DISTINCT這一步驟應該就不難判斷;CBO下優化器會爲你選擇,怎麼寫就只是習慣問題了。
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.