今天做Schema评审的时候发现一个很奇怪的现象,也许是用工具生成的SQL语句,清一色的如下:
CREATE TABLE table_name ( id NUMBER NOT NULL, ...... ...... ) ; CREATE INDEX table_name_PK ON table_name(ID) ; ALTER TABLE table_name ADD CONSTRAINT table_name_PK PRIMARY KEY (ID) USING INDEX table_name_PK ;
通常来说主键(Primary Key,PK)的index是unique index,而现在变成了non-unique index,这有什么不同呢?于是我建了两张1000万数据的表,并用两种不同的index设定为PK的index,语句如下:
create table tab1000w01 as select level id,'killkill Hello world' data from dual connect by level<=1000*10000; create table tab1000w02 as select level id,'killkill Hello world' data from dual connect by level<=1000*10000; CREATE UNIQUE INDEX tab1000w01_pk ON tab1000w01 (PK_ID) ; ALTER TABLE tab1000w01 ADD CONSTRAINT tab1000w01_PK PRIMARY KEY (PK_ID) USING INDEX tab1000w01_pk ; CREATE INDEX tab1000w02_pk ON tab500w02 (PK_ID) ; ALTER TABLE tab1000w02 ADD CONSTRAINT tab1000w02_PK PRIMARY KEY (PK_ID) USING INDEX tab1000w02_pk ;
以下是按照PK查找数据的语句:
select * from tab1000w01 where id=34567; ------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 35 | 3 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| TAB1000W01 | 1 | 35 | 3 (0)| 00:00:01 | |* 2 | INDEX UNIQUE SCAN | IDX_TAB1000W01_PK | 1 | | 2 (0)| 00:00:01 | ------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 2 - access("ID"=34567) Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 4 consistent gets select * from tab1000w02 where id=34567; ------------------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ------------------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 35 | 3 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| TAB1000W02 | 1 | 35 | 3 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | IDX_TAB1000W02_PK | 1 | | 2 (0)| 00:00:01 | ------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 2 - access("ID"=34567) Statistics ---------------------------------------------------------- 0 recursive calls 0 db block gets 5 consistent gets
从执行计划来看,一个是index unique scan,一个是index range scan,从consistent gets来看,一个是4,一个是5,使用unique index节省了1个,不要少看这1个consistent gets,它可是占了总体的20%啊。
不过这是为什么呢?这篇文章很好地介绍这两种索引的异同:Differences Between Unique and Non-Unique Indexes,说到底是这两种索引的结构不同。引用一下这篇文章的分析:
Leaf block dump =============== header address 143336028=0x88b225c kdxcolev 0 KDXCOLEV Flags = - - - kdxcolok 0 kdxcoopc 0x80: opcode=0: iot flags=--- is converted=Y kdxconco 2 kdxcosdc 0 kdxconro 500 kdxcofbo 1036=0x40c kdxcofeo 1042=0x412 kdxcoavs 6 kdxlespl 0 kdxlende 0 kdxlenxt 75520140=0x480588c kdxleprv 75520138=0x480588a kdxledsz 0 kdxlebksz 8036 row#0[8022] flag: ------, lock: 0, len=14 <=== length is 14 bytes for the index row entry col 0; len 4; (4): c3 60 61 1c col 1; len 6; (6): 04 80 50 3c 01 06 <=== rowid is stored as a second column for the index row entry row#1[8008] flag: ------, lock: 0, len=14 col 0; len 4; (4): c3 60 61 1d col 1; len 6; (6): 04 80 50 3c 01 07
non-unique index将 rowid 作为一个字段和数据字段组合成一个“唯一、复合”索引,而unique index的结构如下:
Leaf block dump =============== header address 143336028=0x88b225c kdxcolev 0 KDXCOLEV Flags = - - - kdxcolok 0 kdxcoopc 0x80: opcode=0: iot flags=--- is converted=Y kdxconco 1 kdxcosdc 0 kdxconro 533 kdxcofbo 1102=0x44e kdxcofeo 1112=0x458 kdxcoavs 10 kdxlespl 0 kdxlende 0 kdxlenxt 75527436=0x480750c kdxleprv 75527434=0x480750a kdxledsz 6 kdxlebksz 8036 row#0[8023] flag: ------, lock: 0, len=13, data:(6): 04 80 5e 34 02 82 <=== length is 13 byes and rowid not stored as a second column entry col 0; len 4; (4): c3 60 30 2c row#1[8010] flag: ------, lock: 0, len=13, data:(6): 04 80 5e 34 02 83 col 0; len 4; (4): c3 60 30 2d
从dump文件中可以看到结构不同导致index中的entry的长度是不一样的,unique index稍稍短一点,所以每个block可以容纳更多的index entry,从宏观来看unique index更小一点。