1. 选用适合的ORACLE优化?
(tng) (tng) (tng) ORACLE的优化器共有3U?
(tng) (tng) a. (tng) RULE (Z规则) (tng) (tng) b. COST (Z成本) (tng) c. CHOOSE (选择?
(tng) (tng) (tng) 讄~省的优化器,可以通过对init.ora文g中OPTIMIZER_MODE参数的各U声?如RULE,COST,CHOOSE,ALL_ROWS,FIRST_ROWS . 你当然也在SQL句或是?x)?session)U对其进行覆?
(tng) (tng) Z(jin)使用Z成本的优化器(CBO, Cost-Based Optimizer) , 你必ȝ常运行analyze 命o(h),以增加数据库中的对象l计信息(object statistics)的准?
(tng) (tng) 如果数据库的优化器模式设|ؓ(f)选择?CHOOSE),那么实际的优化器模式和是否q行qanalyze命o(h)有关. 如果table已经被analyzeq? 优化器模式将自动成ؓ(f)CBO , 反之,数据库将采用RULE形式的优化器.
(tng) (tng) 在缺省情况下,ORACLE采用CHOOSE优化? Z(jin)避免那些不必要的全表扫描(full table scan) , 你必d量避免用CHOOSE优化?而直接采用基于规则或者基于成本的优化?
(tng) 2. (tng) (tng) (tng) (tng) (tng) (tng) 讉KTable的方?
(tng) ORACLE 采用两种讉K表中记录的方?
a. (tng) (tng) (tng) (tng) (tng) (tng) 全表扫描 (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 全表扫描是序地访问表中每条记? ORACLE采用一ơ读入多个数据块(database block)的方式优化全表扫?
(tng) (tng) (tng) b. (tng) (tng) (tng) (tng) (tng) (tng) 通过ROWID讉K?
(tng) (tng) (tng) (tng) (tng) (tng) 你可以采用基于ROWID的访问方式情?提高讉K表的效率, , ROWID包含?jin)表中记录的物理位置信?.ORACLE采用索引(INDEX)实现?jin)数据和存放数据的物理位|?ROWID)之间的联p? 通常索引提供?jin)快速访问ROWID的方?因此那些Z索引列的查询可以得到性能上的提高.
(tng)3. (tng) (tng) (tng) (tng) (tng) (tng) ׃nSQL语句
Z(jin)不重复解析相同的SQL语句,在第一ơ解析之? ORACLESQL语句存放在内存(sh).q块位于pȝ全局区域SGA(system global area)的共享池(shared buffer pool)中的内存可以被所有的数据库用户共? 因此,当你执行一个SQL语句(有时被称Z个游??如果?
和之前的执行q的语句完全相同, ORACLEp很快获得已经被解析的语句以及(qing)最好的执行路径. ORACLE的这个功能大大地提高?sh)(jin)SQL的执行性能q节省了(jin)内存的?
(tng) (tng) (tng) (tng) 可惜的是ORACLE只对单的表提供高速缓?cache buffering) ,q个功能q不适用于多表连接查?
数据库管理员必须在init.ora中ؓ(f)q个区域讄合适的参数,当这个内存区域越?可以保留更多的语句,当然被共享的可能性也p大了(jin).
当你向ORACLE 提交一个SQL语句,ORACLE?x)首先在q块内存?sh)查扄同的语?
(tng)q里需要注明的?ORACLE对两者采取的是一U严格匹?要达成共?SQL语句必须
完全相同(包括I格,换行{?.
(tng) (tng) (tng) (tng) ׃n的语句必L三个条?
(tng)A. (tng) (tng) (tng) (tng) (tng) 字符U的比较:
当前被执行的语句和共享池中的语句必须完全相同.
(tng) (tng) (tng) (tng) (tng) 例如:
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT * FROM EMP;
(tng) (tng) (tng) (tng) (tng) 和下列每一个都不同
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT * from EMP;
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) Select * From Emp;
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT (tng) (tng) (tng) (tng) (tng) * (tng) (tng) (tng) (tng) FROM EMP;
B. (tng) (tng) (tng) (tng) (tng) 两个语句所指的对象必须完全相同:
例如:
(tng) (tng) 用户 对象?如何讉K
Jack sal_limit private synonym
Work_city public synonym
Plant_detail public synonym
Jill sal_limit private synonym
Work_city public synonym
Plant_detail table owner
(tng) (tng) (tng) (tng) 考虑一下下列SQL语句能否在这两个用户之间׃n.
(tng)SQL 能否׃n 原因
select max(sal_cap) from sal_limit; 不能 每个用户都有一个private synonym - sal_limit , 它们是不同的对象
select count(*0 from work_city where sdesc like 'NEW%'; ?两个用户讉K相同的对象public synonym - work_city (tng)
select a.sdesc,b.location from work_city a , plant_detail b where a.city_id = b.city_id 不能 用户jack 通过private synonym讉Kplant_detail 而jill 是表的所有?对象不同.
C. (tng) (tng) (tng) (tng) (tng) 两个SQL语句中必M用相同的名字的绑定变?bind variables)
例如Q第一l的两个SQL语句是相同的(可以׃n),而第二组中的两个语句是不同的(即在运行时,赋于不同的绑定变量相同的?
a.
select pin , name from people where pin = :blk1.pin;
select pin , name from people where pin = :blk1.pin;
b.
select pin , name from people where pin = :blk1.ot_ind;
select pin , name from people where pin = :blk1.ov_ind;
4. 选择最有效率的表名序(只在Z规则的优化器中有?
ORACLE的解析器按照从右到左的顺序处理FROM子句中的表名,因此FROM子句中写在最后的?基础?driving table)被最先处? 在FROM子句中包含多个表的情况下,你必选择记录条数最的表作为基?当ORACLE处理多个表时, ?x)运用排序?qing)合ƈ的方式连接它?首先,扫描W一个表(FROM子句中最后的那个?q对记录q行zֺ,然后扫描W二个表(FROM子句中最后第二个?,最后将所有从W二个表中检索出的记录与W一个表中合适记录进行合q?
例如: (tng) (tng) (tng) (tng) (tng) ?TAB1 16,384 条记?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) ?TAB2 1 (tng) (tng) (tng) (tng) (tng) 条记?
(tng) (tng) (tng) (tng) 选择TAB2作ؓ(f)基础?(最好的Ҏ(gu))
(tng) (tng) (tng) (tng) select count(*) from tab1,tab2 (tng) (tng) 执行旉0.96U?
(tng) (tng) (tng) (tng) 选择TAB2作ؓ(f)基础?(不佳的方?
select count(*) from tab2,tab1 (tng) (tng) 执行旉26.09U?
如果?个以上的表连接查? 那就需要选择交叉?intersection table)作ؓ(f)基础? 交叉表是指那个被其他表所引用的表.
例如: (tng) (tng) EMP表描qC(jin)LOCATION表和CATEGORY表的交集.
SELECT * (tng)
FROM LOCATION L , (tng)
(tng) (tng) (tng) (tng) (tng) CATEGORY C,
(tng) (tng) (tng) (tng) (tng) EMP E (tng)
WHERE E.EMP_NO BETWEEN 1000 AND 2000
AND E.CAT_NO = C.CAT_NO
AND E.LOCN = L.LOCN
(tng)比下列SQL更有效率
SELECT * (tng)
FROM EMP E ,
LOCATION L , (tng)
(tng) (tng) (tng) (tng) (tng) CATEGORY C
WHERE (tng) E.CAT_NO = C.CAT_NO
AND E.LOCN = L.LOCN
AND E.EMP_NO BETWEEN 1000 AND 2000
5. (tng) (tng) (tng) (tng) (tng) (tng) WHERE子句中的q接序Q?
(tng) (tng) ORACLE采用自下而上的顺序解析WHERE子句,Ҏ(gu)q个原理,表之间的q接必须写在其他WHERE条g之前, 那些可以qo(h)掉最大数量记录的条g必须写在WHERE子句的末?
(tng)例如:
(tng)(低效,执行旉156.3U?
SELECT …?
FROM EMP E
WHERE (tng) SAL >; 50000
AND (tng) (tng) (tng) JOB = ‘MANAGER?
AND (tng) (tng) (tng) 25 < (SELECT COUNT(*) FROM EMP
WHERE MGR=E.EMPNO);
(tng)(高效,执行旉10.6U?
SELECT …?
FROM EMP E
WHERE 25 < (SELECT COUNT(*) FROM EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE MGR=E.EMPNO)
AND (tng) (tng) (tng) SAL >; 50000
AND (tng) (tng) (tng) JOB = ‘MANAGER?
6. (tng) (tng) (tng) (tng) SELECT子句中避免??* ?
当你惛_SELECT子句中列出所有的COLUMN?使用动态SQL列引???是一个方便的Ҏ(gu).不幸的是,q是一个非怽效的Ҏ(gu). 实际?ORACLE在解析的q程? ?x)将??依次转换成所有的列名, q个工作是通过查询数据字典完成? q意味着耗费更多的时? (tng)
7. (tng) (tng) (tng) (tng) 减少讉K数据库的ơ数
当执行每条SQL语句? ORACLE在内部执行了(jin)许多工作: 解析SQL语句, 估算索引的利用率, l定变量 , L据块{等. 由此可见, 减少讉K数据库的ơ数 , p实际上减ORACLE的工作量.
(tng)例如,
(tng) (tng) (tng) 以下有三U方法可以检索出雇员L(fng)?342?291的职?
(tng)Ҏ(gu)1 (最低效)
(tng) (tng) (tng) SELECT EMP_NAME , SALARY , GRADE
(tng) (tng) (tng) FROM EMP (tng)
(tng) (tng) (tng) WHERE EMP_NO = 342;
(tng) (tng) (tng) (tng) SELECT EMP_NAME , SALARY , GRADE
(tng) (tng) (tng) FROM EMP (tng)
(tng) (tng) (tng) WHERE EMP_NO = 291;
Ҏ(gu)2 (ơ低?
(tng) (tng) (tng) (tng) (tng) (tng) DECLARE (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) CURSOR C1 (E_NO NUMBER) IS (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT EMP_NAME,SALARY,GRADE
FROM EMP (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE EMP_NO = E_NO;
(tng) (tng) (tng) BEGIN (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) OPEN C1(342);
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FETCH C1 INTO ?..,.. ;
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) OPEN C1(291);
(tng) (tng) (tng) (tng) (tng) (tng) FETCH C1 INTO ?..,.. ;
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) CLOSE C1;
(tng) (tng) (tng) (tng) (tng) END;
Ҏ(gu)3 (高效)
(tng) (tng) (tng) SELECT A.EMP_NAME , A.SALARY , A.GRADE,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) B.EMP_NAME , B.SALARY , B.GRADE
(tng) (tng) (tng) FROM EMP A,EMP B
(tng) (tng) (tng) WHERE A.EMP_NO = 342
(tng) (tng) (tng) AND (tng) (tng) B.EMP_NO = 291;
(tng)注意:
在SQL*Plus , SQL*Forms和Pro*C中重新设|ARRAYSIZE参数, 可以增加每次数据库访问的(g)索数据量 ,gؓ(f)200.
8. (tng) (tng) (tng) (tng) (tng) (tng) 使用DECODE函数来减处理时?
使用DECODE函数可以避免重复扫描相同记录或重复连接相同的?
例如:
(tng) (tng) SELECT COUNT(*)QSUM(SAL)
(tng) (tng) FROM EMP
(tng) (tng) WHERE DEPT_NO = 0020
(tng) (tng) AND ENAME LIKE ‘SMITH%?
(tng) (tng) SELECT COUNT(*)QSUM(SAL)
(tng) (tng) FROM EMP
(tng) (tng) WHERE DEPT_NO = 0030
(tng) (tng) AND ENAME LIKE ‘SMITH%?
你可以用DECODE函数高效地得到相同结?
SELECT COUNT(DECODE(DEPT_NO,0020,’X?NULL)) D0020_COUNT,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) COUNT(DECODE(DEPT_NO,0030,’X?NULL)) D0030_COUNT,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SUM(DECODE(DEPT_NO,0020,SAL,NULL)) D0020_SAL,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SUM(DECODE(DEPT_NO,0030,SAL,NULL)) D0030_SAL
FROM EMP WHERE ENAME LIKE ‘SMITH%?
cM?DECODE函数也可以运用于GROUP BY 和ORDER BY子句?
9. (tng) (tng) (tng) (tng) (tng) (tng) 整合?无关联的数据库访?
如果你有几个单的数据库查询语?你可以把它们整合C个查询中(即它们之间没有关系)
例如:
(tng)SELECT NAME (tng)
FROM EMP (tng)
WHERE EMP_NO = 1234;
(tng)SELECT NAME (tng)
FROM DPT
WHERE DPT_NO = 10 ;
SELECT NAME (tng)
FROM CAT
WHERE CAT_TYPE = ‘RD?
上面?个查询可以被合ƈ成一?
(tng)SELECT E.NAME , D.NAME , C.NAME
FROM CAT C , DPT D , EMP E,DUAL X
WHERE NVL(‘X?X.DUMMY) = NVL(‘X?E.ROWID(+))
AND NVL(‘X?X.DUMMY) = NVL(‘X?D.ROWID(+))
AND NVL(‘X?X.DUMMY) = NVL(‘X?C.ROWID(+))
AND E.EMP_NO(+) = 1234
AND D.DEPT_NO(+) = 10
AND C.CAT_TYPE(+) = ‘RD?
(tng)(译者按: 虽然采取q种Ҏ(gu),效率得到提高,但是E序的可L大大降?所以读?q是要权衡之间的利弊)
10. (tng) (tng) (tng) (tng) (tng) (tng) 删除重复记录
最高效的删除重复记录方?( 因ؓ(f)使用?jin)ROWID)
DELETE FROM EMP E
WHERE E.ROWID >; (SELECT MIN(X.ROWID) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM EMP X
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE X.EMP_NO = E.EMP_NO);
11. (tng) (tng) (tng) (tng) (tng) (tng) 用TRUNCATE替代DELETE
当删除表中的记录?在通常情况? 回滚D?rollback segments ) 用来存放可以被恢复的信息. 如果你没有COMMIT事务,ORACLE?x)将数据恢复到删除之前的状?准确地说是恢复到执行删除命o(h)之前的状?
而当q用TRUNCATE? 回滚D不再存放Q何可被恢复的信息.当命令运行后,数据不能被恢?因此很少的资源被调用,执行旉也会(x)很短.
(tng)(译者按: TRUNCATE只在删除全表适用,TRUNCATE是DDL不是DML)
12. (tng) (tng) (tng) (tng) (tng) (tng) 量多用COMMIT
只要有可?在程序中量多用COMMIT, q样E序的性能得到提高,需求也?x)因为COMMIT所释放的资源而减?
(tng)COMMIT所释放的资?
a. (tng) (tng) (tng) (tng) (tng) (tng) 回滚D上用于恢复数据的信?
b. (tng) (tng) (tng) (tng) (tng) (tng) 被程序语句获得的?
c. (tng) (tng) (tng) (tng) (tng) (tng) redo log buffer 中的I间
d. (tng) (tng) (tng) (tng) (tng) (tng) ORACLE为管理上q?U资源中的内部花?
(tng)(译者按: 在用COMMIT时必要注意C务的完整?现实中效率和事务完整性往往是鱼和熊掌不可得?
13. (tng) (tng) (tng) (tng) (tng) (tng) 计算记录条数
(tng) (tng) (tng) (tng) 和一般的观点相反, count(*) 比count(1)E快 , 当然如果可以通过索引(g)?对烦(ch)引列的计C旧是最快的. 例如 COUNT(EMPNO)
(tng) (译者按: 在CSDN论坛?曄Ҏ(gu)有过相当热烈的讨? 作者的观点q不十分准确,通过实际的测?上述三种Ҏ(gu)q没有显著的性能差别)
(tng)14. (tng) (tng) (tng) (tng) (tng) (tng) 用Where子句替换HAVING子句
(tng) (tng) (tng) (tng) (tng) 避免使用HAVING子句, HAVING 只会(x)在检索出所有记录之后才对结果集q行qo(h). q个处理需要排?总计{操? 如果能通过WHERE子句限制记录的数?那就能减这斚w的开销.
(tng)例如:
(tng) (tng) (tng) (tng) 低效:
(tng) (tng) (tng) (tng) SELECT REGIONQAVG(LOG_SIZE)
(tng) (tng) (tng) (tng) FROM LOCATION
(tng) (tng) (tng) (tng) GROUP BY REGION
(tng) (tng) (tng) (tng) HAVING REGION REGION != ‘SYDNEY?
(tng) (tng) (tng) (tng) AND REGION != ‘PERTH?
(tng) (tng) (tng) (tng) (tng) 高效
(tng) (tng) (tng) (tng) SELECT REGIONQAVG(LOG_SIZE)
(tng) (tng) (tng) (tng) FROM LOCATION
(tng) (tng) (tng) (tng) WHERE REGION REGION != ‘SYDNEY?
(tng) (tng) (tng) (tng) AND REGION != ‘PERTH?
(tng) (tng) (tng) (tng) GROUP BY REGION
(译者按: HAVING 中的条g一般用于对一些集合函数的比较,如COUNT() {等. 除此而外,一般的条g应该写在WHERE子句?
15. (tng) (tng) (tng) (tng) (tng) (tng) 减少对表的查?
在含有子查询的SQL语句?要特别注意减对表的查询.
(tng) 例如: (tng)
(tng) (tng) (tng) (tng) 低效
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT TAB_NAME
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM TABLES
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE TAB_NAME = ( SELECT TAB_NAME (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM TAB_COLUMNS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE VERSION = 604)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) AND DB_VER= ( SELECT DB_VER (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM TAB_COLUMNS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE VERSION = 604)
(tng) (tng) (tng) (tng) 高效
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT TAB_NAME
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM TABLES
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE (tng) (TAB_NAME,DB_VER)
(tng)= ( SELECT TAB_NAME,DB_VER) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM TAB_COLUMNS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE VERSION = 604)
(tng) (tng) (tng) (tng) Update 多个Column 例子:
(tng) (tng) (tng) (tng) 低效:
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) UPDATE EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SET EMP_CAT = (SELECT MAX(CATEGORY) FROM EMP_CATEGORIES),
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SAL_RANGE = (SELECT MAX(SAL_RANGE) FROM EMP_CATEGORIES)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE EMP_DEPT = 0020;
(tng) (tng) (tng) (tng) 高效:
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) UPDATE EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SET (EMP_CAT, SAL_RANGE)
(tng)= (SELECT MAX(CATEGORY) , MAX(SAL_RANGE)
(tng)FROM EMP_CATEGORIES)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE EMP_DEPT = 0020;
(tng) (tng) 16. (tng) (tng) (tng) (tng) (tng) (tng) 通过内部函数提高SQL效率.
(tng) (tng) (tng) (tng) (tng) SELECT H.EMPNO,E.ENAME,H.HIST_TYPE,T.TYPE_DESC,COUNT(*)
(tng) (tng) (tng) (tng) FROM HISTORY_TYPE T,EMP E,EMP_HISTORY H
(tng) (tng) (tng) (tng) WHERE H.EMPNO = E.EMPNO
AND H.HIST_TYPE = T.HIST_TYPE
GROUP BY H.EMPNO,E.ENAME,H.HIST_TYPE,T.TYPE_DESC;
通过调用下面的函数可以提高效?
FUNCTION LOOKUP_HIST_TYPE(TYP IN NUMBER) RETURN VARCHAR2
AS
(tng) (tng) (tng) TDESC VARCHAR2(30);
(tng) (tng) (tng) CURSOR C1 IS (tng) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT TYPE_DESC (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM HISTORY_TYPE
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE HIST_TYPE = TYP;
BEGIN (tng)
(tng) (tng) (tng) OPEN C1;
(tng) (tng) (tng) FETCH C1 INTO TDESC;
(tng) (tng) (tng) CLOSE C1;
(tng) (tng) (tng) RETURN (NVL(TDESC,??);
END;
(tng)
FUNCTION LOOKUP_EMP(EMP IN NUMBER) RETURN VARCHAR2
AS
(tng) (tng) (tng) ENAME VARCHAR2(30);
(tng) (tng) (tng) CURSOR C1 IS (tng) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT ENAME
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE EMPNO=EMP;
BEGIN (tng)
(tng) (tng) (tng) OPEN C1;
(tng) (tng) (tng) FETCH C1 INTO ENAME;
(tng) (tng) (tng) CLOSE C1;
(tng) (tng) (tng) RETURN (NVL(ENAME,??);
END;
(tng)
SELECT H.EMPNO,LOOKUP_EMP(H.EMPNO),
H.HIST_TYPE,LOOKUP_HIST_TYPE(H.HIST_TYPE),COUNT(*)
FROM EMP_HISTORY H
GROUP BY H.EMPNO , H.HIST_TYPE;
(tng)(译者按: l常在论坛中看到?’能不能用一个SQL写出??的脓(chung)? D不知复杂的SQL往往牺牲?jin)执行效? 能够掌握上面的运用函数解决问题的Ҏ(gu)在实际工作中是非常有意义?
17. (tng) (tng) (tng) (tng) (tng) (tng) 使用表的别名(Alias)
当在SQL语句中连接多个表? 请用表的别名ƈ把别名前~于每个Column?q样一?可以减解析的旉q减那些由Column歧义引v的语法错?
(tng) (译者注: Column歧义指的是由于SQL中不同的表具有相同的Column?当SQL语句中出现这个Column?SQL解析器无法判断这个Column的归?
18. (tng) (tng) (tng) (tng) (tng) (tng) 用EXISTS替代IN
在许多基于基表的查询?Z(jin)满一个条?往往需要对另一个表q行联接.在这U情况下, 使用EXISTS(或NOT EXISTS)通常提高查询的效率.
(tng)低效:
SELECT * (tng)
FROM EMP (基础?
WHERE EMPNO >; 0
AND DEPTNO IN (SELECT DEPTNO (tng)
FROM DEPT (tng)
WHERE LOC = ‘MELB?
(tng) (tng) (tng) 高效:
SELECT * (tng)
FROM EMP (基础?
WHERE EMPNO >; 0
AND EXISTS (SELECT ‘X’?
FROM DEPT (tng)
WHERE DEPT.DEPTNO = EMP.DEPTNO
AND LOC = ‘MELB?
(译者按: 相对来说,用NOT EXISTS替换NOT IN 更显著地提高效?下一节中指?
19. (tng) (tng) (tng) (tng) (tng) (tng) 用NOT EXISTS替代NOT IN
在子查询?NOT IN子句执行一个内部的排序和合q? 无论在哪U情况下,NOT IN都是最低效?(因ؓ(f)它对子查询中的表执行?jin)一个全表遍?. (tng) Z(jin)避免使用NOT IN ,我们可以把它改写成外q接(Outer Joins)或NOT EXISTS.
(tng)例如:
SELECT ?
FROM EMP
WHERE DEPT_NO NOT IN (SELECT DEPT_NO (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEPT (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE DEPT_CAT=’A?;
Z(jin)提高效率.改写?
(tng)(Ҏ(gu)一: 高效)
SELECT ?
FROM EMP A,DEPT B
WHERE A.DEPT_NO = B.DEPT(+)
AND B.DEPT_NO IS NULL
AND B.DEPT_CAT(+) = ‘A?
(tng)(Ҏ(gu)? 最高效)
SELECT ?
FROM EMP E
WHERE NOT EXISTS (SELECT ‘X’?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEPT D
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE D.DEPT_NO = E.DEPT_NO
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) AND DEPT_CAT = ‘A?;
20. (tng) (tng) (tng) (tng) (tng) (tng) 用表q接替换EXISTS
(tng) (tng) (tng) (tng) 通常来说 , 采用表连接的方式比EXISTS更有效率
(tng) (tng) (tng) (tng) (tng) SELECT ENAME
(tng) (tng) (tng) (tng) (tng) FROM EMP E
(tng) (tng) (tng) (tng) (tng) WHERE EXISTS (SELECT ‘X’?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEPT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE DEPT_NO = E.DEPT_NO
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) AND DEPT_CAT = ‘A?;
(tng) (tng) (tng) (tng) (更高?
(tng) (tng) (tng) (tng) (tng) SELECT ENAME
(tng) (tng) (tng) (tng) (tng) FROM DEPT D,EMP E
(tng) (tng) (tng) (tng) (tng) WHERE E.DEPT_NO = D.DEPT_NO
(tng) (tng) (tng) (tng) (tng) AND DEPT_CAT = ‘A?;
(tng) (译者按: 在RBO的情况下,前者的执行路径包括F(xin)ILTER,后者用NESTED LOOP)
(tng)21. (tng) (tng) (tng) (tng) (tng) (tng) 用EXISTS替换DISTINCT
当提交一个包含一对多表信?比如部门表和雇员?的查询时,避免在SELECT子句中用DISTINCT. 一般可以考虑用EXIST替换
(tng)例如:
低效:
(tng) (tng) (tng) SELECT DISTINCT DEPT_NO,DEPT_NAME
(tng) (tng) (tng) FROM DEPT D,EMP E
(tng) (tng) (tng) WHERE D.DEPT_NO = E.DEPT_NO
高效:
(tng) (tng) (tng) SELECT DEPT_NO,DEPT_NAME
(tng) (tng) (tng) FROM DEPT D
(tng) (tng) (tng) WHERE EXISTS ( SELECT ‘X?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM EMP E
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE E.DEPT_NO = D.DEPT_NO);
(tng) EXISTS 使查询更?因ؓ(f)RDBMS核心(j)模块在子查询的条g一旦满_,立刻q回l果.
(tng)22. (tng) (tng) (tng) (tng) (tng) (tng) 识别’低效执行’的SQL语句
用下列SQL工具扑և低效SQL:
(tng)SELECT EXECUTIONS , DISK_READS, BUFFER_GETS,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) ROUND((BUFFER_GETS-DISK_READS)/BUFFER_GETS,2) Hit_radio,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) ROUND(DISK_READS/EXECUTIONS,2) Reads_per_run,
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SQL_TEXT
FROM (tng) (tng) V$SQLAREA
WHERE (tng) EXECUTIONS>;0
AND (tng) (tng) (tng) (tng) BUFFER_GETS >; 0 (tng)
AND (BUFFER_GETS-DISK_READS)/BUFFER_GETS < 0.8 (tng)
ORDER BY 4 DESC;
(tng) (tng) (tng) (tng) (译者按: 虽然目前各种关于SQL优化的图形化工具层出不穷,但是写出自己的SQL工具来解决问题始l是一个最好的Ҏ(gu))
23. (tng) (tng) (tng) (tng) (tng) (tng) 使用TKPROF 工具来查询SQL性能状?
(tng)SQL trace 工具攉正在执行的SQL的性能状态数据ƈ记录C个跟t文件中. q个跟踪文g提供?jin)许多有用的信?例如解析ơ数.执行ơ数,CPU使用旉{?q些数据可以用来优化你的系l?
讄SQL TRACE在会(x)话? 有效
(tng) (tng) ALTER SESSION SET SQL_TRACE TRUE
讄SQL TRACE 在整个数据库有效? 你必dSQL_TRACE参数在init.ora中设为TRUE, USER_DUMP_DEST参数说明?jin)生成跟t文件的目录
(tng) (译者按: q一节中,作者ƈ没有提到TKPROF的用? 对SQL TRACE的用法也不够准确, 讄SQL TRACE首先要在init.ora中设定TIMED_STATISTICS, q样才能得到那些重要的时间状? 生成的trace文g是不可读?所以要用TKPROF工具对其q行转换,TKPROF有许多执行参? 大家可以参考ORACLE手册来了(jin)解具体的配置. )
24. (tng) (tng) (tng) (tng) (tng) (tng) 用EXPLAIN PLAN 分析SQL语句
(tng)EXPLAIN PLAN 是一个很好的分析SQL语句的工?它甚臛_以在不执行SQL的情况下分析语句. 通过分析,我们可以知道ORACLE是怎么栯接表,使用什么方式扫描表(索引扫描或全表扫?以及(qing)使用到的索引名称.
你需要按照从里到?从上C的次序解d析的l果. EXPLAIN PLAN分析的结果是用羃q的格式排列? 最内部的操作将被最先解? 如果两个操作处于同一层中,带有最操作号的将被首先执?
NESTED LOOP是少C按照上述规则处理的操? 正确的执行\径是(g)查对NESTED LOOP提供数据的操?其中操作h的被最先处?
译者按: (tng)
(tng)通过实践, 感到q是用SQLPLUS中的SET TRACE 功能比较方便.
举例:
(tng)SQL>; list
(tng) 1 (tng) SELECT *
(tng) 2 (tng) FROM dept, emp
(tng) 3* WHERE emp.deptno = dept.deptno
SQL>; set autotrace traceonly /*traceonly 可以不显C执行结?/
SQL>; /
14 rows selected.
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) NESTED LOOPS
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (FULL) OF 'EMP' (tng)
(tng) (tng) 3 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'DEPT'
(tng) (tng) 4 (tng) (tng) (tng) 3 (tng) (tng) (tng) (tng) (tng) (tng) INDEX (UNIQUE SCAN) OF 'PK_DEPT' (UNIQUE)
(tng)
Statistics
----------------------------------------------------------
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) recursive calls
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 2 (tng) db block gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 30 (tng) consistent gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) physical reads
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) redo size
(tng) (tng) (tng) (tng) (tng) (tng) 2598 (tng) bytes sent via SQL*Net to client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) 503 (tng) bytes received via SQL*Net from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 2 (tng) SQL*Net roundtrips to/from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (memory)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (disk)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 14 (tng) rows processed
通过以上分析,可以得出实际的执行步骤是:
1. (tng) (tng) (tng) (tng) (tng) (tng) TABLE ACCESS (FULL) OF 'EMP' (tng)
2. (tng) (tng) (tng) (tng) (tng) (tng) INDEX (UNIQUE SCAN) OF 'PK_DEPT' (UNIQUE)
3. (tng) (tng) (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'DEPT'
4. (tng) (tng) (tng) (tng) (tng) (tng) NESTED LOOPS (JOINING 1 AND 3)
? 目前许多W三方的工具如TOAD和ORACLE本n提供的工具如OMS的SQL Analyze都提供了(jin)极其方便的EXPLAIN PLAN工具.也许喜欢囑Ş化界面的朋友们可以选用它们.
25. (tng) (tng) (tng) (tng) (tng) (tng) 用烦(ch)引提高效?
索引是表的一个概念部?用来提高?gu)(g)索数据的效率. 实际?ORACLE使用?jin)一个复杂的自^衡B-treel构. 通常,通过索引查询数据比全表扫描要? 当ORACLE扑և执行查询和Update语句的最佌\径时, ORACLE优化器将使用索引. 同样在联l多个表时用烦(ch)引也可以提高效率. 另一个用烦(ch)引的好处?它提供了(jin)主键(primary key)的唯一性验?
除了(jin)那些LONG或LONG RAW数据cd, 你可以烦(ch)引几乎所有的? 通常, 在大型表中用烦(ch)引特别有? 当然,你也?x)发? 在扫描小表时,使用索引同样能提高效?
虽然使用索引能得到查询效率的提高,但是我们也必L意到它的代h(hun). 索引需要空间来
存储,也需要定期维? 每当有记录在表中增减或烦(ch)引列被修Ҏ(gu), 索引本n也会(x)被修? q意味着每条记录的INSERT , DELETE , UPDATEؓ(f)此多付出4 , 5 ơ的盘I(y)/O . 因ؓ(f)索引需要额外的存储I间和处?那些不必要的索引反而会(x)使查询反应时间变?
译者按: (tng)
定期的重构烦(ch)引是有必要的. (tng)
ALTER INDEX <INDEXNAME>; REBUILD <TABLESPACENAME>;
26. (tng) (tng) (tng) (tng) (tng) (tng) 索引的操?
ORACLE对烦(ch)引有两种讉K模式.
索引唯一扫描 ( INDEX UNIQUE SCAN)
大多数情况下, 优化器通过WHERE子句讉KINDEX.
例如:
表LODGING有两个烦(ch)?: 建立在LODGING列上的唯一性烦(ch)引LODGING_PK和徏立在MANAGER列上的非唯一性烦(ch)引LODGING$MANAGER. (tng)
(tng)
SELECT * (tng)
FROM LODGING
WHERE LODGING = ‘ROSE HILL?
(tng) (tng) 在内?, 上述SQL被分成两步执行, 首先 , LODGING_PK 索引通过索引唯一扫描的方式被讉K , 获得相对应的ROWID, 通过ROWID讉K表的方式 执行下一步检?
(tng) (tng) 如果被检索返回的列包括在INDEX列中,ORACLE不执行W二步的处理(通过ROWID讉K?. 因ؓ(f)(g)索数据保存在索引? 单单讉K索引可以完全满x(chng)询结? (tng)
(tng) (tng) 下面SQL只需要INDEX UNIQUE SCAN 操作.
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT LODGING
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM (tng) LODGING
WHERE LODGING = ‘ROSE HILL?
(tng) 索引范围查询(INDEX RANGE SCAN)
(tng) (tng) (tng) (tng) (tng) 适用于两U情?
1. (tng) (tng) (tng) (tng) (tng) (tng) Z一个范围的(g)?
2. (tng) (tng) (tng) (tng) (tng) (tng) Z非唯一性烦(ch)引的(g)?
(tng)?:
(tng) (tng) (tng) (tng) (tng) SELECT LODGING
(tng) (tng) (tng) (tng) (tng) FROM (tng) LODGING
WHERE LODGING LIKE ‘M%?
WHERE子句条g包括一pd? ORACLE通过索引范围查询的方式查询LODGING_PK . ׃索引范围查询返回一l? 它的效率p比烦(ch)引唯一扫描
低一? (tng)
?:
(tng) (tng) (tng) (tng) (tng) SELECT LODGING
(tng) (tng) (tng) (tng) (tng) FROM (tng) LODGING
WHERE MANAGER = ‘BILL GATES?
(tng)
(tng) q个SQL的执行分两步, LODGING$MANAGER的烦(ch)引范围查?得到所有符合条件记录的ROWID) 和下一步同qROWID讉K表得到LODGING列的? ׃LODGING$MANAGER是一个非唯一性的索引,数据库不能对它执行烦(ch)引唯一扫描. (tng)
(tng) ׃SQLq回LODGING?而它q不存在于LODGING$MANAGER索引? 所以在索引范围查询后会(x)执行一个通过ROWID讉K表的操作. (tng)
(tng) WHERE子句? 如果索引列所对应的值的W一个字W由通配W?WILDCARD)开? 索引不被采?
(tng)SELECT LODGING
(tng) (tng) (tng) (tng) (tng) FROM (tng) LODGING
WHERE MANAGER LIKE ‘％HANMAN?
(tng)在这U情况下QORACLE用全表扫?
27. (tng) (tng) (tng) (tng) (tng) (tng) 基础表的选择
基础?Driving Table)是指被最先访问的?通常以全表扫描的方式被访?. Ҏ(gu)优化器的不同, SQL语句中基表的选择是不一L(fng).
如果你用的是CBO (COST BASED OPTIMIZER),优化器会(x)(g)查SQL语句中的每个表的物理大小,索引的状?然后选用p最低的执行路径.
如果你用RBO (RULE BASED OPTIMIZER) , q且所有的q接条g都有索引对应, 在这U情况下, 基础表就是FROM 子句中列在最后的那个?
举例:
(tng) (tng) (tng) (tng) SELECT A.NAME , B.MANAGER
(tng) (tng) (tng) (tng) FROM WORKER A, (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) LODGING B
(tng) (tng) (tng) (tng) WHERE A.LODGING = B.LODING;
׃LODGING表的LODING列上有一个烦(ch)? 而且WORKER表中没有相比较的索引, WORKER表将被作为查询中的基?
28. (tng) (tng) (tng) (tng) (tng) (tng) 多个q等的烦(ch)?
当SQL语句的执行\径可以用分布在多个表上的多个烦(ch)引时, ORACLE?x)同时用多个?ch)引ƈ在运行时对它们的记录q行合ƈ, (g)索出仅对全部索引有效的记?
在ORACLE选择执行路径?唯一性烦(ch)引的{高(sh)非唯一性烦(ch)? 然而这个规则只?
当WHERE子句中烦(ch)引列和常量比较才有效.如果索引列和其他表的索引cȝ比较. q种子句在优化器中的{是非怽?
如果不同表中两个惛_{的烦(ch)引将被引? FROM子句中表的顺序将军_哪个?x)被率先使? FROM子句中最后的表的索引有最高的优先U?
如果相同表中两个惛_{的烦(ch)引将被引? WHERE子句中最先被引用的烦(ch)引将有最高的优先U?
举例:
(tng) (tng) (tng) (tng) DEPTNO上有一个非唯一性烦(ch)?EMP_CAT也有一个非唯一性烦(ch)?
(tng) (tng) (tng) (tng) SELECT ENAME,
(tng) (tng) (tng) (tng) FROM EMP
(tng) (tng) (tng) (tng) WHERE DEPT_NO = 20
(tng) (tng) (tng) (tng) AND EMP_CAT = ‘A?
q里,DEPTNO索引被最先检?然后同EMP_CAT索引(g)索出的记录进行合q? 执行路径如下:
TABLE ACCESS BY ROWID ON EMP
(tng) (tng) (tng) AND-EQUAL
(tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX RANGE SCAN ON DEPT_IDX
(tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX RANGE SCAN ON CAT_IDX
29. (tng) (tng) (tng) (tng) (tng) (tng) (tng) {式比较和范围比?
(tng) (tng) (tng) (tng) 当WHERE子句中有索引? ORACLE不能合ƈ它们,ORACLE用范围比较.
(tng) (tng) (tng) (tng) 举例:
(tng) (tng) (tng) (tng) DEPTNO上有一个非唯一性烦(ch)?EMP_CAT也有一个非唯一性烦(ch)?
(tng) (tng) (tng) (tng) SELECT ENAME
(tng) (tng) (tng) (tng) FROM EMP
(tng) (tng) (tng) (tng) WHERE DEPTNO >; 20
(tng) (tng) (tng) (tng) AND EMP_CAT = ‘A?
(tng) (tng) (tng) (tng)
(tng) (tng) (tng) (tng) q里只有EMP_CAT索引被用?然后所有的记录逐条与DEPTNO条gq行比较. 执行路径如下:
(tng) (tng) (tng) (tng) TABLE ACCESS BY ROWID ON EMP (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX RANGE SCAN ON CAT_IDX
30. (tng) (tng) (tng) (tng) (tng) (tng) 不明的索引{
当ORACLE无法判断索引的等U高?sh)差?优化器将只用一个烦(ch)?它就是在WHERE子句中被列在最前面?
(tng) (tng) (tng) (tng) 举例:
(tng) (tng) (tng) (tng) DEPTNO上有一个非唯一性烦(ch)?EMP_CAT也有一个非唯一性烦(ch)?
(tng) (tng) (tng) (tng) (tng)
(tng) (tng) (tng) (tng) SELECT ENAME
(tng) (tng) (tng) (tng) FROM EMP
(tng) (tng) (tng) (tng) WHERE DEPTNO >; 20
(tng) (tng) (tng) (tng) AND EMP_CAT >; ‘A?
(tng) (tng) (tng) (tng) q里, ORACLE只用C(jin)DEPT_NO索引. 执行路径如下:
(tng) (tng) (tng) (tng) (tng)
(tng) (tng) (tng) (tng) TABLE ACCESS BY ROWID ON EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX RANGE SCAN ON DEPT_IDX
译者按:
我们来试一下以下这U情?
SQL>; select index_name, uniqueness from user_indexes where table_name = 'EMP';
INDEX_NAME (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) UNIQUENES
------------------------------ ---------
EMPNO (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) UNIQUE
EMPTYPE (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) NONUNIQUE
SQL>; select * from emp where empno >;= 2 and emp_type = 'A' ;
no rows selected
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'EMP'
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'EMPTYPE' (NON-UNIQUE)
(tng)
虽然EMPNO是唯一性烦(ch)?但是׃它所做的是范围比? {要比非唯一性烦(ch)引的{式比较?
31. (tng) (tng) (tng) (tng) (tng) (tng) 强制索引失效
(tng) (tng) (tng)
如果两个或以上烦(ch)引具有相同的{,你可以强制命令ORACLE优化器用其中的一?通过?(g)索出的记录数量少) .
举例:
(tng) (tng) (tng)
SELECT ENAME
FROM EMP
WHERE EMPNO = 7935 (tng) (tng)
AND DEPTNO + 0 = 10 (tng) (tng) (tng) /*DEPTNO上的索引失?/
AND EMP_TYPE || ‘?= ‘A’?/*EMP_TYPE上的索引失?/
q是一U相当直接的提高查询效率的办? 但是你必}慎考虑q种{略,一般来?只有在你希望单独优化几个SQL时才能采用它.
q里有一个例子关于何旉用这U策? (tng)
假设在EMP表的EMP_TYPE列上有一个非唯一性的索引而EMP_CLASS上没有烦(ch)? (tng)
SELECT ENAME
FROM EMP
WHERE EMP_TYPE = ‘A’?
AND EMP_CLASS = ‘X?
优化器会(x)注意到EMP_TYPE上的索引q用它. q是目前唯一的选择. 如果,一D|间以? 另一个非唯一性徏立在EMP_CLASS?优化器必d两个索引q行选择,在通常情况?优化器将使用两个索引q在他们的结果集合上执行排序?qing)合q? 然?如果其中一个烦(ch)引（EMP_TYPEQ接q于唯一性而另一个烦(ch)引（EMP_CLASSQ上有几千个重复的? 排序?qing)合q就?x)成ZU不必要的负? 在这U情况下,你希望优化器屏蔽掉EMP_CLASS索引.
用下面的Ҏ(gu)可以解决问?
SELECT ENAME
FROM EMP
WHERE EMP_TYPE = ‘A’?
AND EMP_CLASS||’?= ‘X?
32. (tng) (tng) (tng) (tng) (tng) (tng) 避免在烦(ch)引列上用计．
WHERE子句中，如果索引列是函数的一部分Q优化器不使用索引而用全表扫描．
举例:
低效Q?
SELECT ?
FROM DEPT
WHERE SAL * 12 >; 25000;
高效:
SELECT ?
FROM DEPT
WHERE SAL (tng) >; 25000/12;
译者按:
q是一个非常实用的规则Q请务必牢记
33. (tng) (tng) (tng) (tng) (tng) (tng) 自动选择索引
如果表中有两个以上（包括两个Q烦(ch)引，其中有一个唯一性烦(ch)引，而其他是非唯一性．
在这U情况下QORACLE用唯一性烦(ch)引而完全忽略非唯一性烦(ch)引．
举例:
SELECT ENAME
FROM EMP
WHERE EMPNO = 2326 (tng) (tng)
AND DEPTNO (tng) = 20 ;
q里Q只有EMPNO上的索引是唯一性的Q所以EMPNO索引用来检索记录．
TABLE ACCESS BY ROWID ON EMP
(tng) (tng) (tng) (tng) (tng) (tng) INDEX UNIQUE SCAN ON EMP_NO_IDX
(tng)
34. (tng) (tng) (tng) (tng) (tng) (tng) 避免在烦(ch)引列上用NOT
通常Q 我们要避免在索引列上使用NOT, NOT?x)生在和在索引列上使用函数相同?
影响. 当ORACLE”遇到”NOT,他就?x)停止用?ch)引{而执行全表扫?
(tng) (tng) 举例:
(tng) (tng) 低效: (q里,不用烦(ch)?
(tng) (tng) SELECT ?
(tng) (tng) FROM DEPT
(tng) (tng) WHERE DEPT_CODE NOT = 0;
(tng) (tng) (tng)
(tng) (tng) 高效: (q里,使用?jin)?ch)?
(tng) SELECT ?
(tng) (tng) FROM DEPT
(tng) (tng) WHERE DEPT_CODE >; 0;
(tng) (tng) 需要注意的?在某些时? ORACLE优化器会(x)自动NOT转化成相对应的关pL作符.
(tng) (tng) NOT >; (tng) to (tng) <=
(tng) (tng) NOT >;= (tng) to (tng) <
(tng) (tng) NOT < (tng) to (tng) >;=
(tng) (tng) NOT <= (tng) to (tng) >;
(tng)
译者按:
(tng) (tng) (tng) (tng) 在这个例子中,作者犯?jin)一些错? 例子中的低效率SQL是不能被执行?
我做?jin)一些测?
(tng) (tng) (tng) (tng) (tng)
SQL>; select * from emp where NOT empno >; 1;
no rows selected
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'EMP'
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'EMPNO' (UNIQUE) (tng) (tng) (tng) (tng) (tng)
SQL>; select * from emp where empno <= 1;
no rows selected
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'EMP'
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'EMPNO' (UNIQUE)
(tng) (tng) (tng) (tng) (tng) 两者的效率完全一P也许q符合作者关于?在某些时? ORACLE优化器会(x)自动NOT转化成相对应的关pL作符?的观?/p>
35. (tng) (tng) (tng) (tng) (tng) (tng) ?gt;;=替代>;
如果DEPTNO上有一个烦(ch)? (tng)
高效:
(tng) (tng) SELECT *
(tng) (tng) FROM EMP
(tng) (tng) WHERE DEPTNO >;=4
(tng) (tng) (tng)
(tng) (tng) 低效:
(tng) (tng) SELECT *
(tng) (tng) FROM EMP
(tng) (tng) WHERE DEPTNO >;3
(tng) (tng) (tng) (tng) (tng) 两者的区别在于, 前者DBMS直接蟩到第一个DEPT{于4的记录而后者将首先定位到DEPTNO=3的记录ƈ且向前扫描到W一个DEPT大于3的记?
36. (tng) (tng) (tng) (tng) (tng) (tng) 用UNION替换OR (适用于烦(ch)引列)
通常情况? 用UNION替换WHERE子句中的OR会(x)起到较好的效? 对烦(ch)引列使用OR造成全表扫描. 注意, 以上规则只针对多个烦(ch)引列有效. 如果有column没有被烦(ch)? 查询效率可能?x)因Z没有选择OR而降? (tng)
(tng) (tng) 在下面的例子? LOC_ID 和REGION上都建有索引.
高效:
(tng) (tng) SELECT LOC_ID , LOC_DESC , REGION
(tng) (tng) FROM LOCATION
(tng) (tng) WHERE LOC_ID = 10
(tng) (tng) UNION
(tng) (tng) SELECT LOC_ID , LOC_DESC , REGION
(tng) (tng) FROM LOCATION
(tng) (tng) WHERE REGION = “MELBOURNE?
低效:
(tng) (tng) SELECT LOC_ID , LOC_DESC , REGION
(tng) (tng) FROM LOCATION
(tng) (tng) WHERE LOC_ID = 10 OR REGION = “MELBOURNE?
如果你坚持要用OR, 那就需要返回记录最的索引列写在最前面.
注意:
WHERE KEY1 = 10 (tng) (tng) (q回最记?
OR KEY2 = 20 (tng) (tng) (tng) (tng) (tng) (tng) (tng) (q回最多记?
ORACLE 内部以上{换ؓ(f)
WHERE KEY1 = 10 AND
((NOT KEY1 = 10) AND KEY2 = 20) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng)
译者按: (tng)
下面的测试数据仅供参? (a = 1003 q回一条记?, b = 1 q回1003条记?
SQL>; select * from unionvsor /*1st test*/
(tng) 2 (tng) (tng) where a = 1003 or b = 1;
1003 rows selected.
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) CONCATENATION
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 3 (tng) (tng) (tng) 2 (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)
(tng) (tng) 4 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 5 (tng) (tng) (tng) 4 (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)
Statistics
----------------------------------------------------------
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) recursive calls
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) db block gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) 144 (tng) consistent gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) physical reads
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) redo size
(tng) (tng) (tng) (tng) (tng) 63749 (tng) bytes sent via SQL*Net to client
(tng) (tng) (tng) (tng) (tng) (tng) 7751 (tng) bytes received via SQL*Net from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 68 (tng) SQL*Net roundtrips to/from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (memory)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (disk)
(tng) (tng) (tng) (tng) (tng) (tng) 1003 (tng) rows processed
SQL>; select * from unionvsor /*2nd test*/
(tng) 2 (tng) where b (tng) = 1 or a = 1003 ; (tng)
1003 rows selected.
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) CONCATENATION
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 3 (tng) (tng) (tng) 2 (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)
(tng) (tng) 4 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 5 (tng) (tng) (tng) 4 (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)
Statistics
----------------------------------------------------------
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) recursive calls
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) db block gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) 143 (tng) consistent gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) physical reads
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) redo size
(tng) (tng) (tng) (tng) (tng) 63749 (tng) bytes sent via SQL*Net to client
(tng) (tng) (tng) (tng) (tng) (tng) 7751 (tng) bytes received via SQL*Net from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 68 (tng) SQL*Net roundtrips to/from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (memory)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (disk)
(tng) (tng) (tng) (tng) (tng) (tng) 1003 (tng) rows processed
SQL>; select * from unionvsor /*3rd test*/
(tng) 2 (tng) where a = 1003
(tng) 3 (tng) union (tng)
(tng) 4 (tng) (tng) select * from unionvsor
(tng) 5 (tng) (tng) where b = 1;
1003 rows selected.
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) SORT (UNIQUE)
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) UNION-ALL
(tng) (tng) 3 (tng) (tng) (tng) 2 (tng) (tng) (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 4 (tng) (tng) (tng) 3 (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UA' (NON-UNIQUE)
(tng) (tng) 5 (tng) (tng) (tng) 2 (tng) (tng) (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'UNIONVSOR'
(tng) (tng) 6 (tng) (tng) (tng) 5 (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'UB' (NON-UNIQUE)
Statistics
----------------------------------------------------------
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) recursive calls
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) db block gets
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 10 (tng) consistent gets (tng) (tng) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) physical reads
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) redo size
(tng) (tng) (tng) (tng) (tng) 63735 (tng) bytes sent via SQL*Net to client
(tng) (tng) (tng) (tng) (tng) (tng) 7751 (tng) bytes received via SQL*Net from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 68 (tng) SQL*Net roundtrips to/from client
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 1 (tng) sorts (memory)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) 0 (tng) sorts (disk)
(tng) (tng) (tng) (tng) (tng) (tng) 1003 (tng) rows processed
用UNION的效果可以从consistent gets?SQL*NET的数据交换量的减看?
37. (tng) (tng) (tng) (tng) (tng) (tng) 用IN来替换OR
下面的查询可以被更有效率的语句替?
低效:
SELECT?
FROM LOCATION
WHERE LOC_ID = 10
OR (tng) (tng) (tng) (tng) LOC_ID = 20
OR (tng) (tng) (tng) (tng) LOC_ID = 30
高效
SELECT?
FROM LOCATION
WHERE LOC_IN IN (10,20,30);
(tng) (tng) (tng) (tng)
译者按:
q是一条简单易记的规则Q但是实际的执行效果q须(g)验，在ORACLE8i下，两者的执行路径g是相同的Q
38. (tng) (tng) (tng) (tng) (tng) (tng) 避免在烦(ch)引列上用IS NULL和IS NOT NULL
避免在烦(ch)引中使用M可以为空的列QORACLE无法用该索引 Q对于单列烦(ch)引，如果列包含空|索引中将不存在此记录. 对于复合索引Q如果每个列都ؓ(f)I，索引中同样不存在此记? 如果臛_有一个列不ؓ(f)I，则记录存在于索引中．
举例:
(tng) 如果唯一性烦(ch)引徏立在表的A列和B列上, q且表中存在一条记录的A,Bgؓ(f)(123,null) , ORACLE不接受下一条具有相同A,B|123,nullQ的记录(插入). 然而如?
所有的索引列都为空QORACLE认为整个键gؓ(f)I空不等于空. 因此你可以插?000
条具有相同键值的记录,当然它们都是I?
(tng) (tng) (tng) (tng) (tng) 因ؓ(f)Ig存在于烦(ch)引列?所以WHERE子句中对索引列进行空值比较将使ORACLE停用该烦(ch)?
举例:
低效: (索引失效)
SELECT ?
FROM DEPARTMENT
WHERE DEPT_CODE IS NOT NULL;
高效: (索引有效)
SELECT ?
FROM DEPARTMENT
WHERE DEPT_CODE >;=0;
39. (tng) (tng) (tng) (tng) (tng) (tng) L使用索引的第一个列
如果索引是徏立在多个列上, 只有在它的第一个列(leading column)被where子句引用?优化器才?x)选择使用该烦(ch)? (tng)
译者按:
q也是一条简单而重要的规则. 见以下实?
SQL>; create table multiindexusage ( inda number , indb number , descr varchar2(10));
Table created.
SQL>; create index multindex on multiindexusage(inda,indb);
Index created.
SQL>; set autotrace traceonly
SQL>; (tng) select * from (tng) multiindexusage where inda = 1;
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'MULTIINDEXUSAGE'
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) INDEX (RANGE SCAN) OF 'MULTINDEX' (NON-UNIQUE)
SQL>; select * from (tng) multiindexusage where indb = 1;
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (FULL) OF 'MULTIINDEXUSAGE'
(tng) (tng) (tng)
很明? 当仅引用索引的第二个列时,优化器用了(jin)全表扫描而忽略了(jin)索引
40. (tng) (tng) (tng) (tng) (tng) (tng) ORACLE内部操作
当执行查询时,ORACLE采用?jin)内部的操? 下表昄?jin)几U重要的内部操作.
ORACLE Clause 内部操作
ORDER BY SORT ORDER BY
UNION UNION-ALL
MINUS MINUS
INTERSECT INTERSECT
DISTINCT,MINUS,INTERSECT,UNION SORT UNIQUE
MIN,MAX,COUNT SORT AGGREGATE
GROUP BY SORT GROUP BY
ROWNUM COUNT or COUNT STOPKEY
Queries involving Joins SORT JOIN,MERGE JOIN,NESTED LOOPS
CONNECT BY CONNECT BY
41. (tng) (tng) (tng) (tng) (tng) (tng) 用UNION-ALL 替换UNION ( 如果有可能的?
当SQL语句需要UNION两个查询l果集合?q两个结果集合会(x)以UNION-ALL的方式被合ƈ, 然后在输出最l结果前q行排序.
如果用UNION ALL替代UNION, q样排序׃是必要了(jin). 效率׃(x)因此得到提高.
举例:
(tng) (tng) 低效Q?
SELECT ACCT_NUM, BALANCE_AMT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEBIT_TRANSACTIONS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE TRAN_DATE = ?1-DEC-95?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) UNION
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT ACCT_NUM, BALANCE_AMT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEBIT_TRANSACTIONS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE TRAN_DATE = ?1-DEC-95?
高效:
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT ACCT_NUM, BALANCE_AMT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEBIT_TRANSACTIONS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE TRAN_DATE = ?1-DEC-95?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) UNION ALL
(tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT ACCT_NUM, BALANCE_AMT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEBIT_TRANSACTIONS
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE TRAN_DATE = ?1-DEC-95?
译者按:
需要注意的是，UNION ALL 重复输Z个结果集合中相同记录. 因此各位q是
要从业务需求分析用UNION ALL的可行?
UNION 对l果集合排序,q个操作?x)用到SORT_AREA_SIZEq块内存. 对于q?
块内存的优化也是相当重要? 下面的SQL可以用来查询排序的消耗量
Select substr(name,1,25) (tng) "Sort Area Name",
(tng) (tng) (tng) substr(value,1,15) (tng) (tng) "Value"
from v$sysstat
where name like 'sort%'
(tng) (tng) (tng) (tng)
42. (tng) (tng) (tng) (tng) (tng) (tng) 使用提示(Hints)
对于表的讉K,可以使用两种Hints.
FULL ?ROWID
FULL hint 告诉ORACLE使用全表扫描的方式访问指定表.
例如:
(tng) (tng) SELECT /*+ FULL(EMP) */ *
(tng) (tng) FROM EMP
(tng) (tng) WHERE EMPNO = 7893;
(tng) (tng) ROWID hint 告诉ORACLE使用TABLE ACCESS BY ROWID的操作访问表.
(tng) (tng) 通常, 你需要采用TABLE ACCESS BY ROWID的方式特别是当访问大表的时? 使用q种方式, 你需要知道ROIWD的值或者用烦(ch)?
(tng) (tng) 如果一个大表没有被讑֮为缓?CACHED)表而你希望它的数据在查询结束是仍然停留
在SGA?你就可以使用CACHE hint 来告诉优化器把数据保留在SGA? 通常CACHE hint ?FULL hint 一起?
例如:
SELECT /*+ FULL(WORKER) CACHE(WORKER)*/ *
FROM WORK;
(tng) (tng) 索引hint 告诉ORACLE使用Z索引的扫描方? 你不必说明具体的索引名称
例如:
(tng) (tng) SELECT /*+ INDEX(LODGING) */ LODGING
(tng) (tng) FROM LODGING
(tng) (tng) WHERE MANAGER = ‘BILL GATES?
(tng) (tng) (tng)
(tng) (tng) 在不使用hint的情况下, 以上的查询应该也?x)用?ch)?然?如果该烦(ch)引的重复D多而你的优化器是CBO, 优化器就可能忽略索引. 在这U情况下, 你可以用INDEX hint强制ORACLE使用该烦(ch)?
(tng) (tng) ORACLE hints q包括ALL_ROWS, FIRST_ROWS, RULE,USE_NL, USE_MERGE, USE_HASH {等.
(tng) (tng) (tng)
译者按:
(tng) (tng) 使用hint , 表示我们对ORACLE优化器缺省的执行路径不满?需要手工修?
q是一个很有技巧性的工作. 我徏议只针对特定?数的SQLq行hint的优?
对ORACLE的优化器q是要有信心(j)(特别是CBO)
43. (tng) (tng) (tng) (tng) (tng) (tng) 用WHERE替代ORDER BY
ORDER BY 子句只在两种严格的条件下使用索引.
ORDER BY中所有的列必d含在相同的烦(ch)引中q保持在索引中的排列序.
ORDER BY中所有的列必d义ؓ(f)非空.
WHERE子句使用的烦(ch)引和ORDER BY子句中所使用的烦(ch)引不能ƈ?
例如:
(tng) (tng) (tng) (tng) (tng) 表DEPT包含以下?
(tng) (tng) (tng) (tng) (tng) (tng) (tng) DEPT_CODE (tng) (tng) (tng) PK (tng) (tng) (tng) NOT NULL
(tng) (tng) (tng) (tng) (tng) (tng) (tng) DEPT_DESC (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) NOT NULL
(tng) (tng) (tng) (tng) (tng) (tng) (tng) DEPT_TYPE (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) NULL
(tng) (tng) (tng) (tng)
(tng) (tng) (tng) (tng) (tng) (tng) 非唯一性的索引(DEPT_TYPE)
(tng) (tng) (tng) (tng) 低效: (索引不被使用)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT DEPT_CODE
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEPT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) ORDER BY DEPT_TYPE
(tng) (tng) (tng) (tng) (tng) (tng) EXPLAIN PLAN:
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SORT ORDER BY (tng)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) TABLE ACCESS FULL
(tng) (tng) (tng) (tng) 高效: (使用索引)
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) SELECT DEPT_CODE
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) FROM DEPT
(tng) (tng) (tng) (tng) (tng) (tng) (tng) WHERE DEPT_TYPE >; 0 (tng)
(tng) (tng) (tng) EXPLAIN PLAN:
(tng) (tng) (tng) (tng) (tng) TABLE ACCESS BY ROWID ON EMP
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX RANGE SCAN ON DEPT_IDX
译者按:
(tng) (tng) (tng) (tng) (tng) ORDER BY 也能使用索引! q的是个容易被忽视的知识点. 我们来验证一?
SQL>; (tng) select * from emp order by empno;
Execution Plan
----------------------------------------------------------
(tng) (tng) 0 (tng) (tng) (tng) (tng) (tng) SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) 1 (tng) (tng) (tng) 0 (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF 'EMP'
(tng) (tng) 2 (tng) (tng) (tng) 1 (tng) (tng) (tng) (tng) INDEX (FULL SCAN) OF 'EMPNO' (UNIQUE)
44. (tng) (tng) (tng) (tng) (tng) (tng) 避免改变索引列的cd.
当比较不同数据类型的数据? ORACLE自动对列q行单的cd转换.
假设 EMPNO是一个数值类型的索引?
SELECT ?
FROM EMP
WHERE EMPNO = ?23?
实际?l过ORACLEcd转换, 语句转化?
SELECT ?
FROM EMP
WHERE EMPNO = TO_NUMBER(?23?
q运的是,cd转换没有发生在烦(ch)引列?索引的用途没有被改变.
现在,假设EMP_TYPE是一个字W类型的索引?
SELECT ?
FROM EMP
WHERE EMP_TYPE = 123
q个语句被ORACLE转换?
SELECT ?
FROM EMP
WHERE TO_NUMBER(EMP_TYPE)=123
因ؓ(f)内部发生的类型{? q个索引不?x)被用? (tng)
译者按:
Z(jin)避免ORACLE对你的SQLq行隐式的类型{? 最好把cd转换用显式表现出? 注意当字W和数值比较时, ORACLE?x)优先{换数值类型到字符cd.
45. (tng) (tng) (tng) (tng) (tng) (tng) 需要当?j)的WHERE子句
某些SELECT 语句中的WHERE子句不用烦(ch)? q里有一些例?
在下面的例子? ?=?不使用索引. C, 索引只能告诉你什么存在于表中, 而不能告诉你什么不存在于表? (tng)
不用烦(ch)?
SELECT ACCOUNT_NAME
FROM TRANSACTION
WHERE AMOUNT !=0;
使用索引:
SELECT ACCOUNT_NAME
FROM TRANSACTION
WHERE AMOUNT >;0;
下面的例子中, ‘||’是字符q接函数. p其他函数那样, 停用?jin)?ch)?
不用烦(ch)?
SELECT ACCOUNT_NAME,AMOUNT
FROM TRANSACTION
WHERE ACCOUNT_NAME||ACCOUNT_TYPE=’AMEXA?
使用索引:
SELECT ACCOUNT_NAME,AMOUNT
FROM TRANSACTION
WHERE ACCOUNT_NAME = ‘AMEX?
AND (tng) ACCOUNT_TYPE=?A?
下面的例子中, ?’是数学函数. p其他数学函数那样, 停用?jin)?ch)?
不用烦(ch)?
SELECT ACCOUNT_NAME, AMOUNT
FROM TRANSACTION
WHERE AMOUNT + 3000 >;5000;
使用索引:
SELECT ACCOUNT_NAME, AMOUNT
FROM TRANSACTION
WHERE AMOUNT >; 2000 ;
下面的例子中,相同的烦(ch)引列不能互相比较,q将?x)启用全表扫?
不用烦(ch)?
SELECT ACCOUNT_NAME, AMOUNT
FROM TRANSACTION
WHERE ACCOUNT_NAME = NVL(:ACC_NAME,ACCOUNT_NAME);
使用索引:
SELECT ACCOUNT_NAME, AMOUNT
FROM TRANSACTION
WHERE ACCOUNT_NAME LIKE NVL(:ACC_NAME,??;
译者按:
如果一定要对用函数的列启用烦(ch)? ORACLE新的功能: Z函数的烦(ch)?Function-Based Index) 也许是一个较好的Ҏ(gu).
(tng)CREATE INDEX EMP_I ON EMP (UPPER(ename)); /*建立Z函数的烦(ch)?/
(tng)SELECT * FROM emp WHERE UPPER(ename) = ‘BLACKSNAIL? /*用烦(ch)?/
46. (tng) (tng) (tng) (tng) (tng) (tng) q接多个扫描
如果你对一个列和一l有限的D行比? 优化器可能执行多ơ扫描ƈ对结果进行合q连?
举例:
(tng) (tng) (tng) SELECT * (tng)
(tng) (tng) (tng) FROM LODGING
(tng) (tng) (tng) WHERE MANAGER IN (‘BILL GATES?’KEN MULLER?;
(tng) (tng) (tng) 优化器可能将它{换成以下形式
(tng) (tng) (tng) SELECT * (tng)
(tng) (tng) (tng) FROM LODGING
(tng) (tng) (tng) WHERE MANAGER = ‘BILL GATES?
(tng) (tng) (tng) OR MANAGER = ’KEN MULLER?
(tng) (tng) (tng) 当选择执行路径? 优化器可能对每个条g采用LODGING$MANAGER上的索引范围扫描. q回的ROWID用来讉KLODGING表的记录 (通过TABLE ACCESS BY ROWID 的方?. 最后两l记录以q接(CONCATENATION)的Ş式被l合成一个单一的集?
Explain Plan :
SELECT STATEMENT Optimizer=CHOOSE
(tng) (tng) CONCATENATION
(tng) (tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF LODGING
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN ) OF LODGING$MANAGER (NON-UNIQUE)
(tng) (tng) (tng) (tng) TABLE ACCESS (BY INDEX ROWID) OF LODGING
(tng) (tng) (tng) (tng) (tng) (tng) (tng) (tng) INDEX (RANGE SCAN ) OF LODGING$MANAGER (NON-UNIQUE)
译者按:
本节和第37节似乎有矛盾之处. (tng)
47. (tng) (tng) (tng) (tng) (tng) (tng) CBO下用更具选择性的索引
Z成本的优化器(CBO, Cost-Based Optimizer)对烦(ch)引的选择性进行判断来军_索引的用是否能提高效率.
如果索引有很高的选择? 那就是说对于每个不重复的索引键?只对应数量很的记录.
比如, 表中共有100条记录而其中有80个不重复的烦(ch)引键? q个索引的选择性就?0/100 = 0.8 . 选择性越? 通过索引键值检索出的记录就少. (tng)
如果索引的选择性很? (g)索数据就需要大量的索引范围查询操作和ROWID 讉K表的
操作. 也许?x)比全表扫描的效率更?
译者按:
下列l验请参?
a. (tng) (tng) (tng) (tng) (tng) (tng) 如果(g)索数据量过30%的表中记录数.使用索引没有显著的效率提高. (tng)
b. (tng) (tng) (tng) (tng) (tng) (tng) 在特定情况下, 使用索引也许?x)比全表扫描? 但这是同一个数量上的
区别. 而通常情况?使用索引比全表扫描要块几倍乃臛_千?
48. (tng) (tng) (tng) (tng) (tng) (tng) 避免使用耗费资源的操?
带有DISTINCT,UNION,MINUS,INTERSECT,ORDER BY的SQL语句?x)启动SQL引擎
执行耗费资源的排?SORT)功能. DISTINCT需要一ơ排序操? 而其他的臛_需要执行两ơ排?
例如,一个UNION查询,其中每个查询都带有GROUP BY子句, GROUP BY?x)触发嵌入排?NESTED SORT) ; q样, 每个查询需要执行一ơ排? 然后在执行UNION? 又一个唯一排序(SORT UNIQUE)操作被执行而且它只能在前面的嵌入排序结束后才能开始执? 嵌入的排序的深度?x)大大?jing)响查询的效率.
通常, 带有UNION, MINUS , INTERSECT的SQL语句都可以用其他方式重写.
译者按:
(tng) (tng) (tng) (tng) 如果你的数据库的SORT_AREA_SIZE调配得好, 使用UNION , MINUS, INTERSECT也是可以考虑? 毕竟它们的可L很?
49. (tng) (tng) (tng) (tng) (tng) (tng) 优化GROUP BY
提高GROUP BY 语句的效? 可以通过不需要的记录在GROUP BY 之前qo(h)?下面两个查询q回相同l果但第二个明显快?jin)许?
低效:
(tng) (tng) SELECT JOB , AVG(SAL)
(tng) (tng) FROM EMP
(tng) (tng) GROUP JOB (tng)
(tng) (tng) HAVING JOB = ‘PRESIDENT?
(tng) (tng) OR JOB = ‘MANAGER?
(tng)高效:
(tng) (tng) SELECT JOB , AVG(SAL)
(tng) (tng) FROM EMP
(tng) (tng) WHERE JOB = ‘PRESIDENT?
(tng) (tng) OR JOB = ‘MANAGER?
(tng) (tng) GROUP JOB (tng)
(tng) (tng) (tng) 译者按:
(tng) (tng) (tng) (tng) 本节?4节相? 可略q? (tng)
50. (tng) (tng) (tng) (tng) (tng) (tng) 使用日期 (tng)
当用日期是,需要注意如果有过5位小数加到日期上, q个日期?x)进C一?
例如:
1.
SELECT TO_DATE(?1-JAN-93?.99999)
FROM DUAL;
Returns:
(tng) (tng) ?1-JAN-93 23:59:59?
2.
SELECT TO_DATE(?1-JAN-93?.999999)
FROM DUAL;
Returns:
(tng) (tng) ?2-JAN-93 00:00:00?
译者按:
(tng) (tng) (tng) 虽然本节和SQL性能优化没有关系, 但是作者的功力可见一?
51. (tng) (tng) 使用昑ּ的游?CURSORs)
使用隐式的游?会(x)执行两次操作. W一ơ检索记? W二ơ检查TOO MANY ROWS q个exception . 而显式游标不执行W二ơ操? (tng)
52. (tng) (tng) 优化EXPORT和IMPORT
使用较大的BUFFER(比如10MB , 10,240,000)可以提高EXPORT和IMPORT的速度.
ORACLE尽可能地获取你所指定的内存大?即在内存(sh)满,也不?x)报?q个D要和表中最大的列相?否则列g(x)被截? (tng)
译者按:
可以肯定的是, 增加BUFFER?x)大大提高EXPORT , IMPORT的效? (曄到q一个CASE, 增加BUFFER?IMPORT/EXPORT快了(jin)10?) (tng)
作者可能犯?jin)一个错? “这个D要和表中最大的列相?否则列g(x)被截? ?
其中最大的列也许是指最大的记录大小.
关于EXPORT/IMPORT的优?CSDN论坛中有一些ȝ性的贴子,比如关于BUFFER参数, COMMIT参数{等, 详情h.
53. (tng) (tng) 分离表和索引
L你的表和烦(ch)引徏立在不同的表I间?TABLESPACES). 决不要将不属于ORACLE内部pȝ的对象存攑ֈSYSTEM表空间里. 同时,保数据表空间和索引表空间置于不同的盘?sh)?
译者按:
“同?保数据表空间和索引表空间置与不同的盘?sh)?”可能改为如下更为准?“同?保数据表空间和索引表空间置与不同的盘控制卡控制的盘?sh)??/p>