【sql调优之执行计划】merge semi join and merge anti join

版本：10.2.0.4 Semi join（也有叫半连接的）多在子查询in或者exists等中使用，对于外部行集，查找内部（即子查询）行集，匹配第一行之后就返回，不再往下查找例如： SQL> select b.* 2 from scott.dept b 3 where b.deptno in (select deptno from scott.emp a) 4 ; DEPTNO DNAME LOC ---------- -------------- ------------- 10 ACCOUNTING NEW YORK 20 RESEARCH DALLAS 30 SALES CHICAGO Execution Plan ---------------------------------------------------------- Plan hash value: 1090737117 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 69 | 6 (17)| 00:00:01 | | 1 | MERGE JOIN SEMI | | 3 | 69 | 6 (17)| 00:00:01 | | 2 | TABLE ACCESS BY INDEX ROWID| DEPT | 4 | 80 | 2 (0)| 00:00:01 | | 3 | INDEX FULL SCAN | PK_DEPT | 4 | | 1 (0)| 00:00:01 | |* 4 | SORT UNIQUE | | 14 | 42 | 4 (25)| 00:00:01 | | 5 | TABLE ACCESS FULL | EMP | 14 | 42 | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("B"."DEPTNO"="DEPTNO") filter("B"."DEPTNO"="DEPTNO") Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 7 consistent gets 0 physical reads 0 redo size 614 bytes sent via SQL*Net to client 400 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 3 rows processed SQL> 或者： SQL> select b.* 2 from scott.dept b 3 where exists (select 1 from scott.emp a where a.deptno = b.deptno) 4 ; DEPTNO DNAME LOC ---------- -------------- ------------- 10 ACCOUNTING NEW YORK 20 RESEARCH DALLAS 30 SALES CHICAGO Execution Plan ---------------------------------------------------------- Plan hash value: 1090737117 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 69 | 6 (17)| 00:00:01 | | 1 | MERGE JOIN SEMI | | 3 | 69 | 6 (17)| 00:00:01 | | 2 | TABLE ACCESS BY INDEX ROWID| DEPT | 4 | 80 | 2 (0)| 00:00:01 | | 3 | INDEX FULL SCAN | PK_DEPT | 4 | | 1 (0)| 00:00:01 | |* 4 | SORT UNIQUE | | 14 | 42 | 4 (25)| 00:00:01 | | 5 | TABLE ACCESS FULL | EMP | 14 | 42 | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("A"."DEPTNO"="B"."DEPTNO") filter("A"."DEPTNO"="B"."DEPTNO") Statistics ---------------------------------------------------------- 1 recursive calls 0 db block gets 7 consistent gets 0 physical reads 0 redo size 614 bytes sent via SQL*Net to client 400 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 3 rows processed SQL> 可以看到这种情况下，in和exsits的执行计划完全相同，且都使用了merge join semi的oporation 而not in或者not exists则不同，Oracle7.3版本之前not exists和not in还使用的tilter，merge anti join和hash anti join访问路径是后来增加的。 例子： SQL> select b.* 2 from scott.dept b 3 where not exists (select 1 from scott.emp a where a.deptno = b.deptno) 4 ; DEPTNO DNAME LOC ---------- -------------- ------------- 40 OPERATIONS BOSTON Execution Plan ---------------------------------------------------------- Plan hash value: 1353548327 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 23 | 6 (17)| 00:00:01 | | 1 | MERGE JOIN ANTI | | 1 | 23 | 6 (17)| 00:00:01 | | 2 | TABLE ACCESS BY INDEX ROWID| DEPT | 4 | 80 | 2 (0)| 00:00:01 | | 3 | INDEX FULL SCAN | PK_DEPT | 4 | | 1 (0)| 00:00:01 | |* 4 | SORT UNIQUE | | 14 | 42 | 4 (25)| 00:00:01 | | 5 | TABLE ACCESS FULL | EMP | 14 | 42 | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("A"."DEPTNO"="B"."DEPTNO") filter("A"."DEPTNO"="B"."DEPTNO") Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 5 consistent gets 0 physical reads 0 redo size 535 bytes sent via SQL*Net to client 400 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 1 rows processed 这里是merge join anti（也叫反连接），和semi相反，只有外部行在内部不能匹配的时候才返回。 而，not in则和not exsits不同，执行计划显示的是filter： SQL> select b.* 2 from scott.dept b 3 where b.deptno not in (select deptno from scott.emp a) 4 ; DEPTNO DNAME LOC ---------- -------------- ------------- 40 OPERATIONS BOSTON Execution Plan ---------------------------------------------------------- Plan hash value: 3547749009 --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 60 | 7 (0)| 00:00:01 | |* 1 | FILTER | | | | | | | 2 | TABLE ACCESS FULL| DEPT | 4 | 80 | 3 (0)| 00:00:01 | |* 3 | TABLE ACCESS FULL| EMP | 2 | 6 | 2 (0)| 00:00:01 | --------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - filter( NOT EXISTS (SELECT /*+ */ 0 FROM "SCOTT"."EMP" "A" WHERE LNNVL("DEPTNO"<>:B1))) 3 - filter(LNNVL("DEPTNO"<>:B1)) Statistics ---------------------------------------------------------- 1 recursive calls 0 db block gets 19 consistent gets 5 physical reads 0 redo size 535 bytes sent via SQL*Net to client 400 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 1 rows processed SQL> 我们知道not in与not exsits并不能等同，从执行计划上来看，not in的执行计划的operation是filter，而且内表和外表都是全表，没有使用索引，而从谓词信息中来看，operation 1为： 1 - filter( NOT EXISTS (SELECT /*+ */ 0 FROM "SCOTT"."EMP" "A" WHERE LNNVL("DEPTNO"<>:B1))) Null值对not in影响较大，如果稍稍修改一下这个查询，则又有不同了： SQL> select b.* 2 from scott.dept b 3 where b.deptno not in (select nvl(deptno,0) from scott.emp a); DEPTNO DNAME LOC ---------- -------------- ------------- 40 OPERATIONS BOSTON Execution Plan ---------------------------------------------------------- Plan hash value: 1353548327 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 23 | 6 (17)| 00:00:01 | | 1 | MERGE JOIN ANTI | | 1 | 23 | 6 (17)| 00:00:01 | | 2 | TABLE ACCESS BY INDEX ROWID| DEPT | 4 | 80 | 2 (0)| 00:00:01 | | 3 | INDEX FULL SCAN | PK_DEPT | 4 | | 1 (0)| 00:00:01 | |* 4 | SORT UNIQUE | | 14 | 42 | 4 (25)| 00:00:01 | | 5 | TABLE ACCESS FULL | EMP | 14 | 42 | 3 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 4 - access("B"."DEPTNO"=NVL("DEPTNO",0)) filter("B"."DEPTNO"=NVL("DEPTNO",0)) Statistics ---------------------------------------------------------- 1 recursive calls 0 db block gets 5 consistent gets 0 physical reads 0 redo size 535 bytes sent via SQL*Net to client 400 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 1 rows processed SQL> 奇怪的是，这里使用了索引，因为告诉了oracle不会有null值，而且谓词信息也发生了改变： 4 - access("B"."DEPTNO"=NVL("DEPTNO",0)) 这也是使用not in需要注意的地方。

http://www.itpub.net/viewthread.php?tid=438650&extra=&page=1

转自：http://space.itpub.net/16179598/viewspace-671211