Something about oracle joins

Editor's Notes

• #8: Semi join 有短路功能 ：意思是不全找，找到就 OK ，只要对应每行，查找到匹配的记录，则返回，然后继续下一次查找 Semi join 的一个实际使用，找结果集是否存在数据： select count(*) from dual where exists (select 1 from …); Semi join 也可以使用三大 join 算法， OK ，有 nested loops semi,hash joins semi and merge joins semin……. 以及对应的 hint 控制
• #9: Anti Join 无短路 ，必须对应每行，与对子查询表全部查询一遍，然后无匹配，则返回此行记录，有匹配不返回 和 Semi join 一样，它也有三大算法以及对应的 hint 控制
• #10: -- 检查执行计划，未发现 outer, 检查谓词未发现 + -- 语法级 SQL 转换 DROP TABLE a ; DROP TABLE b ; CREATE TABLE a (ID NUMBER,NAME VARCHAR2( 10 ), code VARCHAR2( 10 )); INSERT INTO a VALUES( 1 ,'a','00001'); INSERT INTO a VALUES( 2 ,'b','00002'); CREATE TABLE b (ID NUMBER,NAME VARCHAR2( 10 ), code NUMBER( 10 )); INSERT INTO b SELECT * FROM a ; INSERT INTO b VALUES( 3 ,'c', 3 ); COMMIT; SELECT * FROM a ; SELECT * FROM b ; dingjun123@ORADB> SELECT a.ID,a.NAME,a.code, 2 b.ID,b.NAME,b.code 3 FROM a,b 4 WHERE a.ID(+)=b.ID AND a.NAME(+)=b.NAME 5 AND TO_NUMBER(a.code)=b.code; 已选择 2 行。 已用时间 : 00: 00: 00.01 执行计划 ---------------------------------------------------------- Plan hash value: 652036164 --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 60 | 7 (15)| 00:00:01 | |* 1 | HASH JOIN | | 1 | 60 | 7 (15)| 00:00:01 | | 2 | TABLE ACCESS FULL| A | 2 | 54 | 3 (0)| 00:00:01 | | 3 | TABLE ACCESS FULL| B | 3 | 99 | 3 (0)| 00:00:01 | --------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - access("A"."ID"="B"."ID" AND "A"."NAME"="B"."NAME" AND "B"."CODE"=TO_NUMBER("A"."CODE")) Note ----- - dynamic sampling used for this statement (level=2) 统计信息 ---------------------------------------------------------- 7 recursive calls 0 db block gets 31 consistent gets 0 physical reads 0 redo size 775 bytes sent via SQL*Net to client 416 bytes received via SQL*Net from client 2 SQL*Net roundtrips to/from client 0 sorts (memory) 0 sorts (disk) 2 rows processed dingjun123@ORADB> SELECT a.ID,a.NAME,a.code, 2 b.ID,b.NAME,b.code 3 FROM a,b 4 WHERE a.ID(+)=b.ID AND a.NAME(+)=b.NAME 5 AND TO_NUMBER(a.code(+))=b.code; 已选择 3 行。 已用时间 : 00: 00: 00.01 执行计划 ---------------------------------------------------------- Plan hash value: 843196925 --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 3 | 180 | 7 (15)| 00:00:01 | |* 1 | HASH JOIN OUTER | | 3 | 180 | 7 (15)| 00:00:01 | | 2 | TABLE ACCESS FULL| B | 3 | 99 | 3 (0)| 00:00:01 | | 3 | TABLE ACCESS FULL| A | 2 | 54 | 3 (0)| 00:00:01 | --------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 1 - access("A"."ID"(+)="B"."ID" AND "A"."NAME"(+)="B"."NAME" AND "B"."CODE"=TO_NUMBER("A"."CODE"(+)))
• #13: -- 理解 +, Conclusion : + operator 的确是 outer join 语法 如果漏掉 righ table 端条件没有写 + ，则语义上是先做 outer join, 然后做 filter 如果 righ table 端其他列选择具体值或有对应列的 join ，但是漏掉 + ，则 Oracle 会将语句转为 inner join 如果 righ table 端条件是 IS NULL ，则是先外连接，再做 filter , 当然，如果是 IS NOT NULL ，则也转为内连接 Left table 的单列条件都是 filter INSERT INTO a VALUES(NULL,'c', 3 ); SELECT * FROM a ; SELECT * FROM b ; --1. 是先进行 id 的外连接，然后按 a.id is null 过滤，类似于 anti join ，找没有匹配到的 b 表记录 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .ID IS NULL; ALTER TABLE a DROP CONSTRAINTS uk_a ; ALTER TABLE b DROP CONSTRAINTS uk_b ; ALTER TABLE b MODIFY ID NULL; SELECT a .ID FROM a , b WHERE a .ID(+)= b .ID AND a .ID IS NULL; SELECT a .ID, a .NAME FROM a WHERE NOT EXISTS (SELECT 1 FROM b WHERE a .ID= b .ID); -- 肯定不返回行，因为是内连接，而且是连接键 IS NULL SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID= b .ID AND a .ID IS NULL; -- 一般内连接， + 无效 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .NAME='b'; -- 一般内连接， + 无效 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .NAME= b .NAME; -- 外连接 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .ID(+) IS NULL; --DELETE FROM a WHERE ID=3; -- 列换为 name INSERT INTO a VALUES( 3 ,NULL, 3 ); INSERT INTO b VALUES( 4 ,'d', 4 ); SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .NAME(+) IS NULL; -- 过滤 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a RIGHT JOIN b ON a .ID= b .ID WHERE a .NAME IS NOT NULL; -- 非连接键值 , 下面两个也不同，第 1 个是先外连接后过滤 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .NAME IS NULL; SELECT a .ID, a .NAME, b .ID, b .NAME FROM a RIGHT JOIN b ON a .ID= b .ID WHERE a .NAME IS NULL; -- 第 2 个内连接 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID= b .ID AND a .NAME IS NULL; SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .ID IS NOT NULL; SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND a .NAME IS NOT NULL; --b.id>1 是过滤，先过滤再连接 SELECT a .ID, a .NAME, b .ID, b .NAME FROM a , b WHERE a .ID(+)= b .ID AND b .ID> 1 ;
• #17: 老语法改写：只能是 where 条件 1.Where dept.location_id(+)=loc.location_id AND loc.city=‘Seattle’; -- 错误，因为有过滤 2. SELECT * FROM hr . departments dept , hr . locations loc WHERE dept . location_id (+)= loc . location_id AND loc . city =CASE WHEN ( dept . location_id (+) IS NOT NULL) THEN 'Seattle' ELSE 'Seattle' END; Loc.city= ‘Seattle’ 的前提是前面的 location_id 条件必须匹配或不匹配，这样外连接的条件就完整了 3. 也可以这样改写 SELECT * FROM hr . departments dept , hr . locations loc WHERE loc . location_id = decode ( loc . city ,'Seattle', dept . location_id (+));
• #18: SELECT * FROM hr . departments dept LEFT JOIN hr . locations loc ON dept . location_id = loc . location_id AND loc . city ='Seattle'; SELECT * FROM hr . departments dept LEFT JOIN hr . locations loc ON dept . location_id = loc . location_id AND loc . city (+)='Seattle';
• #19: 从前面 + 语法分析也可以知道，因为是对非主表过滤，所以语义上是内连接 实际计划也转为内连接，并且 Oracle 很聪明，使用了谓词传递
• #20: 因为对主表过滤，所以是先过滤，后外连接 并且也有主表的谓词传递到了从表的谓词
• #21: SQL 的分析步骤很重要 不进行详细分析，写不好 SQL ，甚至错误 不清楚对象的属性，约束，索引等信息，不清楚对象间的关系，写的 SQL 可能不高效 不能寻找合适正确的语法支撑，写不好 SQL 分析重于一切，必须先分析，才开始动手构造 SQL
• #23: drop TABLE test1 ; drop TABLE test2 ; drop TABLE test3 ; CREATE TABLE test1 AS SELECT 1 id, 'aa' NAME FROM dual UNION ALL SELECT 2 , 'bb' FROM dual UNION ALL SELECT 3 , 'cc' FROM dual ; CREATE TABLE test2 AS SELECT 1 id, 'aa' NAME FROM dual UNION ALL SELECT 4 , 'dd' FROM dual ; CREATE TABLE test3 AS SELECT 2 id, 'bb' NAME FROM dual UNION ALL SELECT 5 , 'ee' FROM dual ; SELECT * FROM test1 a , test2 b WHERE a .ID= b .id(+) OR a .NAME IS NOT NULL; SELECT * FROM test1 a , test2 b WHERE a .ID= b .id(+) AND a .NAME IS NOT NULL; SELECT * FROM test1 a , test2 b WHERE a .ID= b .id(+) AND a .NAME(+) IN (SELECT c .NAME FROM test3 c ); SELECT * FROM test1 a , test2 b WHERE a .ID= b .id(+) AND a .NAME IN (SELECT c .NAME FROM test3 c ); SELECT * FROM test1 a , test2 b , test3 c WHERE a .id = b .id(+) AND a .id = c .id(+); DROP TABLE test1 ; DROP TABLE test2 ; DROP TABLE test3 ; CREATE TABLE test1 AS SELECT 1 id, 'aa' NAME FROM dual UNION ALL SELECT 2 , 'bb' FROM dual ; CREATE TABLE test2 AS SELECT 1 id, 'aa' NAME FROM dual UNION ALL SELECT 4 , 'dd' FROM dual ; CREATE TABLE test3 AS SELECT 2 id, 'bb' NAME FROM dual UNION ALL SELECT 1 , 'aa' FROM dual UNION ALL SELECT 4 , 'dd' FROM dual UNION ALL SELECT 5 , 'ee' FROM dual ; --0ra-01417 一个表最多能外连接到一个表 -- 不知道谁才是基表 SELECT * FROM test1 a , test2 b , test3 c WHERE a .id(+) = b .id AND a .id(+) = c .id; -- 用中间结果集改写 SELECT * FROM (SELECT a .id aid , a .NAME aname , b .id bid , b .NAME bname FROM test1 a , test2 b WHERE a .id(+) = b .id) c , test3 d WHERE c . aid (+) = d .id; -- 基表最终是 test3 SELECT * FROM test1 a RIGHT JOIN test2 b ON a .ID = b .ID RIGHT JOIN test3 c ON a .ID = c .ID; -- 基表最终是 test2 SELECT * FROM test1 a RIGHT JOIN test3 b ON a .ID = b .ID RIGHT JOIN test2 c ON a .ID = c .ID;
• #24: a.ID IS NULL 不能写成 a.ID(+) IS NULL, 习惯选择一个连接键判断，其他的也可以，比如 rowid DROP TABLE a ; DROP TABLE b ; CREATE TABLE a (ID NUMBER,NAME VARCHAR2( 10 )); CREATE TABLE b (ID NUMBER,NAME VARCHAR2( 10 )); INSERT INTO a VALUES( 1 ,'a'); INSERT INTO a VALUES( 2 ,'b'); INSERT INTO a VALUES( 3 ,'c'); INSERT INTO b VALUES( 1 ,'a'); INSERT INTO b VALUES( 2 ,'b'); INSERT INTO b VALUES( 4 ,'d'); COMMIT ； SELECT * FROM a ; SELECT * FROM b ; SELECT a .ID, b .ID FROM a FULL JOIN b ON a .ID= b .ID; -- 如果连接条件 1:1, 可以用 union SELECT a .ID, b .ID FROM a , b WHERE a .ID= b .ID(+) UNION SELECT a .ID, b .ID FROM a , b WHERE a .ID(+)= b .ID; -- 非 1:1 应该用 UNION ALL 并且第 2 个语句在 + 号处的列只选出 NULL 看到有些人经常问，把自己的一些体会简单举个例子，详细的东西还是需要自己慢慢体会的，比如 from a , b where a .id= b .id(+) and b .name='a' 这个 b . name 没有 + 号则相当于普通的内连接 ( 先外连接后过滤，这个准确点，比如 b .name is null 那么和普通内连接还是有所不同的 ) ， -- 用连接键 is null 判断，不用考虑 null 问题，因为选择的是不配的值， rowid is null 也可以，他非连接键只能选择 NOT NULL 约束的 ?? 不对， 这个是过滤，所有没有匹配列都为 NULL INSERT INTO a VALUES( 1 ,'a'); SELECT a .ID, b .ID FROM a , b WHERE a .ID= b .ID(+) UNION ALL SELECT a .ID, b .ID FROM a , b WHERE a .ID(+)= b .ID AND a .ID IS NULL; -- 下面的是找以 id 连接在 b 中不在 a 中的数据，相当于 SELECT a .ID, b .ID FROM a , b WHERE a .ID(+)= b .ID AND a .ID IS NULL; ==> SELECT a .ID, b .ID FROM a RIGHT JOIN b ON a .ID= b .ID WHERE a .ID IS NULL; SELECT a .ID, b .ID FROM a , b WHERE a .ID(+)= b .ID AND a .ID(+) IS NULL; ===> SELECT a .ID, b .ID FROM a RIGHT JOIN b ON a .ID= b .ID AND a .ID IS NULL;
• #25: 73993 * 3 + 295, 类似嵌套循环，效率很低，目标表必须做驱动表 全表更新 DROP TABLE a ; DROP TABLE b ; CREATE TABLE a AS SELECT * FROM all_objects ; CREATE TABLE b AS SELECT * FROM user_objects ; CREATE INDEX idx_a ON a ( object_id ); CREATE INDEX idx_b ON b ( object_id ); BEGIN dbms_stats . gather_table_stats ( ownname => USER, tabname => 'a',cascade => TRUE); dbms_stats . gather_table_stats ( ownname => USER, tabname => 'b',cascade => TRUE); END; SELECT COUNT(*) FROM a ; --73993 SELECT COUNT(*) FROM b ; --2301 UPDATE a SET a . object_name =(SELECT b . object_name FROM b WHERE a . object_id = b . object_id ); UPDATE a SET a . object_name =(SELECT b . object_name FROM b WHERE a . object_id = b . object_id ) WHERE EXISTS (SELECT 1 FROM b WHERE a . object_id = b . object_id ); MERGE INTO a USING b ON a . object_id = b . object_id WHEN MATCHED THEN UPDATE a . object_name = b . object_name ; ALTER TABLE b ADD CONSTRAINTS uk_b UNIQUE( object_id ); UPDATE (SELECT a . object_name aname , b . object_name bname FROM a , b WHERE a . object_id = b . object_id ) SET aname = bname ; SELECT 73993 * 3 + 295 FROM dual ;
• #26: 多次访问 B 表
• #27: Merge 一般不错，但是 merge 也有限制，比如连接条件不能返回多行
• #28: BYPASS_UJVC 11g 不能用了 必须保证 b 的连接键值有唯一约束 为了保证视图是可更新的，其定义中不 能包含以下语法结构（ construct ）： ● 集合操作符（ set operator ） ● DISTINCT 操作符 ● 聚合函数（ aggregate function ）或分析型函数（ analytic function ） ● GROUP BY ， ORDER BY ， CONNECT BY ，或 START WITH 字句 ● 在 SELECT 之后的列表中使用 collection expression ● 在 SELECT 之后的列表中使用子查询（ subquery ） ● 连接（ join ）（但是有例外情况）
• #29: In 和 exists 等价，因为 in 没有 null 的问题，从计划可以看出来
• #30: Exists 一样，只要 semi join 与 or 连用就走不了 semi,filter 效率经常低
• #32: ALTER TABLE a MODIFY object_name NULL; ALTER TABLE b MODIFY object_name NULL; 对于 not exists 是 anti join,not in 主要看前后的都要有 not null 约束 SELECT * FROM a WHERE a . object_name NOT IN (SELECT b . object_name FROM b ); SELECT * FROM a WHERE NOT EXISTS (SELECT 1 FROM b WHERE a . object_name = b . object_name ) SELECT * FROM a WHERE a . object_name NOT IN (SELECT b . object_name FROM b WHERE b . object_name IS NOT NULL); -- 这个菜是 anti join SELECT * FROM a WHERE a . object_name IS NOT NULL AND a . object_name NOT IN (SELECT b . object_name FROM b WHERE b . object_name IS NOT NULL);
• #36: ALTER TABLE a MODIFY object_name NULL; ALTER TABLE b MODIFY object_name NULL; SELECT * FROM a WHERE a . object_name IN (SELECT b . object_name FROM b ); SELECT * FROM a WHERE EXISTS (SELECT 1 FROM b WHERE a . object_name = b . object_name ) OR a . object_id < 80000 ; SELECT * FROM a WHERE a . object_name IN (SELECT b . object_name FROM b ) OR a . object_id < 80000 ; SELECT * FROM a WHERE a . object_name IN (SELECT b . object_name FROM b ) OR a . object_id < (SELECT b . object_id FROM b WHERE b . object_id = 80000 ); SELECT * FROM a WHERE a . object_name IN (SELECT b . object_name FROM b ) UNION SELECT * FROM a WHERE a . object_id < (SELECT b . object_id FROM b WHERE b . object_id = 80000 ); /*+precompute_subquery */ SELECT * FROM a WHERE a . object_name NOT IN (SELECT b . object_name FROM b ); SELECT /*+optimizer_features_enable('9.0.0')*/ * FROM a WHERE a . object_name NOT IN (SELECT b . object_name FROM b ); SELECT * FROM a WHERE NOT EXISTS (SELECT 1 FROM b WHERE a . object_name = b . object_name ) SELECT * FROM a WHERE a . object_name NOT IN (SELECT b . object_name FROM b WHERE b . object_name IS NOT NULL); -- 这个菜是 anti join SELECT * FROM a WHERE a . object_name IS NOT NULL AND a . object_name NOT IN (SELECT b . object_name FROM b WHERE b . object_name IS NOT NULL); SELECT * FROM a WHERE NOT EXISTS (SELECT 1 FROM b WHERE a . object_name = b . object_name ) OR a . object_id < 80000 ; SELECT * FROM a WHERE NOT EXISTS (SELECT 1 FROM b WHERE a . object_name = b . object_name ) UNION SELECT * FROM a WHERE a . object_id < 80000 ;
• #37: 一定情况下也可以考虑外连接
• #38: 当然也可以通过索引来消除排序 就是相互匹配