???? ??在SQL tuning中，不良寫法導致SQL執行效率比比皆是。最近的SQL tuning中一個外部連接寫法不當導致過SQL執行時間超過15分鐘左右此即
為一例。通過修正該SQL的寫法之后，過高的邏輯讀呈數量級下降以及SQL語句執行時間也大幅下降。下面給出一個列子來演示該情形。

一、創建演示環境

-->當前數據庫版本 SQL> select * from v$version where rownum<2; BANNER ---------------------------------------------------------------- Oracle Database 10g Enterprise Edition Release 10.2.0.1.0 - Prod -->創建演示表并插入記錄 SQL> create table t as select empno,ename,job,sal,deptno 2 from emp where 1=2; SQL> insert into t select empno,ename,job,sal,deptno 2 from emp e where empno=(select max(empno) from emp where deptno=e.deptno); SQL> insert into t(empno,ename,job,sal) values(8888,'ROBINSON','DBA',2000); SQL> insert into t(empno,ename,job,sal) values(9999,'JACKSON','CLERK',2500); SQL> commit; SQL> analyze table t compute statistics; SQL> select * from t; EMPNO ENAME JOB SAL DEPTNO ---------- ---------- --------- ---------- ---------- 7934 MILLER CLERK 1300 10 7902 FORD ANALYST 3000 20 7900 JAMES CLERK 950 30 8888 ROBINSON DBA 2000 9999 JACKSON CLERK 2500 -->使用left join連接查看數據，此時表t中所有記錄被返回 SQL> select empno,ename,sal,dname from t left join dept d on t.deptno=d.deptno; EMPNO ENAME SAL DNAME ---------- ---------- ---------- -------------- 7934 MILLER 1300 ACCOUNTING 7902 FORD 3000 RESEARCH 7900 JAMES 950 SALES 8888 ROBINSON 2000 9999 JACKSON 2500 -->下面同樣是使用left join連接，但在on子句中增加了過濾條件t.sal>=2000 -->從下面的返回結果可知，t.sal>=2000子句并沒有過濾掉sal小于2000的記錄 SQL> select empno,ename,sal,dname from t left join dept d -->簡稱語句A 2 on(t.deptno=d.deptno and t.sal>=2000); EMPNO ENAME SAL DNAME ---------- ---------- ---------- -------------- 7934 MILLER 1300 7902 FORD 3000 RESEARCH 7900 JAMES 950 8888 ROBINSON 2000 9999 JACKSON 2500 -->使用left join連接，將過濾條件放到where 子句中 -->此時僅僅t.sal>=2000且符合t.deptno=d.deptno的記錄被返回(結果與所期望一致) SQL> select empno,ename,sal,dname from t left join dept d -->簡稱語句B 2 on t.deptno=d.deptno where t.sal>=2000; EMPNO ENAME SAL DNAME ---------- ---------- ---------- -------------- 7902 FORD 3000 RESEARCH 8888 ROBINSON 2000 9999 JACKSON 2500 -->查看執行計劃 SQL> set autotrace traceonly exp; -->語句A(過濾條件位于on 子句中的情形)的執行計劃 SQL> select empno,ename,sal,dname from t left join dept d 2 (on t.deptno=d.deptno and t.sal>=2000); Execution Plan ---------------------------------------------------------- Plan hash value: 2195752858 ----------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ----------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 5 | 120 | 6 (0)| 00:00:01 | | 1 | NESTED LOOPS OUTER | | 5 | 120 | 6 (0)| 00:00:01 | | 2 | TABLE ACCESS FULL | T | 5 | 70 | 3 (0)| 00:00:01 | | 3 | VIEW | | 1 | 10 | 1 (0)| 00:00:01 | | 4 | TABLE ACCESS BY INDEX ROWID| DEPT | 1 | 13 | 1 (0)| 00:00:01 | |* 5 | INDEX UNIQUE SCAN | PK_DEPT | 1 | | 0 (0)| 00:00:01 | ----------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 5 - access("T"."DEPTNO"="D"."DEPTNO") -->重點關注這里的謂詞信息,兩個過濾條件合在一起 filter("T"."DEPTNO" IS NOT NULL AND "T"."SAL">=2000) -->從執行計劃來看位于第5步為INDEX UNIQUE SCAN Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 11 consistent gets -->此時的邏輯讀為11 0 physical reads 0 redo size 696 bytes sent via SQL*Net to client 385 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 5 rows processed -->語句B(將謂詞信息置于到where子句中的情形)的執行計劃 SQL> select empno,ename,sal,dname from t left join dept d 2 on t.deptno=d.deptno where t.sal>=2000; Execution Plan ---------------------------------------------------------- Plan hash value: 832694258 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 81 | 4 (0)| 00:00:01 | | 1 | NESTED LOOPS OUTER | | 3 | 81 | 4 (0)| 00:00:01 | |* 2 | TABLE ACCESS FULL | T | 3 | 42 | 3 (0)| 00:00:01 | | 3 | TABLE ACCESS BY INDEX ROWID| DEPT | 1 | 13 | 1 (0)| 00:00:01 | |* 4 | INDEX UNIQUE SCAN | PK_DEPT | 1 | | 0 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 2 - filter("T"."SAL">=2000) -->此時的謂詞信息分為兩部分，"T"."SAL">=2000位于第二步 4 - access("T"."DEPTNO"="D"."DEPTNO"(+)) -->此條謂詞信息用于實現表連接 Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 10 consistent gets -->此時的邏輯讀為10，由于2 - filter("T"."SAL">=2000)過濾后，內部循環少執行了一次 0 physical reads 0 redo size 658 bytes sent via SQL*Net to client 385 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 3 rows processed /**************************************************/ /* Author: Robinson Cheng */ /* Blog: http://blog.csdn.net/robinson_0612 */ /* MSN: robinson_0612@hotmail.com */ /* QQ: 645746311 */ /**************************************************/ -->從上面的觀察中發現上述兩條SQL語句執行計劃并非最佳，存在改良的余地 -->由于是nested loops outer,因此考慮在表t的謂詞列增加索引以快速過濾記錄 SQL> create index i_t_sal on t(sal); SQL> exec dbms_stats.gather_table_stats('SCOTT','T',cascade=>true); -->增加索引后兩個語句的執行情況 -->語句A的執行計劃以及統計信息沒有發生任何變化 SQL> select empno,ename,sal,dname from t left join dept d 2 on (t.deptno=d.deptno and t.sal>=2000); Execution Plan ---------------------------------------------------------- Plan hash value: 2195752858 ----------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ----------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 5 | 140 | 6 (0)| 00:00:01 | | 1 | NESTED LOOPS OUTER | | 5 | 140 | 6 (0)| 00:00:01 | | 2 | TABLE ACCESS FULL | T | 5 | 90 | 3 (0)| 00:00:01 | | 3 | VIEW | | 1 | 10 | 1 (0)| 00:00:01 | | 4 | TABLE ACCESS BY INDEX ROWID| DEPT | 1 | 13 | 1 (0)| 00:00:01 | |* 5 | INDEX UNIQUE SCAN | PK_DEPT | 1 | | 0 (0)| 00:00:01 | ----------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 5 - access("T"."DEPTNO"="D"."DEPTNO") filter("T"."DEPTNO" IS NOT NULL AND "T"."SAL">=2000) Statistics ---------------------------------------------------------- 1 recursive calls 0 db block gets 11 consistent gets 0 physical reads 0 redo size 696 bytes sent via SQL*Net to client 385 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 5 rows processed -->語句B的執行計劃發生變化，原來的全表掃描變為索引掃描 SQL> select empno,ename,sal,dname from t left join dept d 2 on t.deptno=d.deptno where t.sal>=2000; Execution Plan ---------------------------------------------------------- Plan hash value: 2452308905 ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 93 | 3 (0)| 00:00:01 | | 1 | NESTED LOOPS OUTER | | 3 | 93 | 3 (0)| 00:00:01 | | 2 | TABLE ACCESS BY INDEX ROWID| T | 3 | 54 | 2 (0)| 00:00:01 | |* 3 | INDEX RANGE SCAN | I_T_SAL | 3 | | 1 (0)| 00:00:01 | | 4 | TABLE ACCESS BY INDEX ROWID| DEPT | 1 | 13 | 1 (0)| 00:00:01 | |* 5 | INDEX UNIQUE SCAN | PK_DEPT | 1 | | 0 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 3 - access("T"."SAL">=2000) 5 - access("T"."DEPTNO"="D"."DEPTNO"(+)) Statistics ---------------------------------------------------------- 1 recursive calls 0 db block gets 6 consistent gets -->邏輯讀也由10下降到6 0 physical reads 0 redo size 658 bytes sent via SQL*Net to client 385 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 3 rows processed

二、總結
??1、盡可能避免SQL不良寫法導致的不良后果
??2、此例中由于將謂詞信息放到ON子句中，在數據量龐大的表(百萬行)連接中，則該寫法導致過多的物理和邏輯I/O,使得中間結果集龐大
??3、謂詞信息放到ON子句中同時也導致索引失效
??4、盡可能的在滿足需求的情況下減小中間結果集

三、更多參考

父游標、子游標及共享游標

綁定變量及其優缺點

dbms_xplan之display_cursor函數的使用

dbms_xplan之display函數的使用

執行計劃中各字段各模塊描述

使用 EXPLAIN PLAN 獲取SQL語句執行計劃

啟用 AUTOTRACE 功能

函數使得索引列失效

Oracle 綁定變量窺探

Oracle 自適應共享游標 ?

Oracle ROWID ?

收縮表段(shrink space)

臨時表空間的管理與受損恢復 ?

O racle 徹底 kill session ?

Oracle 硬解析與軟解析

共享池的調整與優化(Shared pool Tuning)

Buffer cache 的調整與優化(一)

Buffer cache 的調整與優化(二) ?

Oracle 表緩存(caching table)的使用

PGA的設置與調整

?

當心外部連接中的ON子句