如何优化大数据表的LEFTJOIN

[海鲨数据库架构师:曾凡坤.姒]

有两个表 分别是A和B A表是字典表有2500条记录 B表是日志表每天有300万记录.
现在从A表取出700条记录 去跟B表匹配 主要是检查这700条记录可有日志.通过4个ID来匹配.

B表量大的话要130分钟,小的话也要20分钟. 我想提高速度 把B表按时间来划分6个子部分

with tmp as (select * from  a where xxxx=xxx)
select a.*
from tmp a
left join
(
 select * from b 
 where inserttime >= trunc(sysdate)+0/24
   and inserttime  < trunc(sysdate)+4/24
) b on 
where b.id is null

union all

select a.*
from tmp a
left join
(
 select * from b 
 where inserttime >= trunc(sysdate)+4/24
   and inserttime  < trunc(sysdate)+8/24
) b on 
where b.id is null

性能也没提供很多跟使用一条语句相比.我把它分成24个部分也没提高

请问各位想这样有何提供呢? 只能有SQL语句 ,不能使用PL/SQL 因为它使用在其他程序中

[tangren]

1、你这样拆分不会提搞性能
2、相关条件列上有没建立索引
3、修改成not exists试一下
select * from a where exists(
select 1 from b where b.xx=a.xx and ...)
4、贴出相关的执行计划看一下

[tangren]

是not exists
select * from a where not exists(
select 1 from b where b.xx=a.xx and ...)

[海鲨数据库架构师:曾凡坤.姒]

inserttime 的有索引

[海鲨数据库架构师:曾凡坤.姒]

SELECT STATEMENT, GOAL = ALL_ROWS		Cost=90	CPU cost=252284361	IO cost=61	Cardinality=7716	Bytes=100308	
 TEMP TABLE TRANSFORMATION							
  LOAD AS SELECT							
   TABLE ACCESS FULL	Object name=ETL_RULE	Cost=37	CPU cost=207190251	IO cost=13	Cardinality=1286	Bytes=136316	
  FILTER							
   HASH GROUP BY		Cost=53	CPU cost=45094111	IO cost=48	Cardinality=7716	Bytes=100308	
    VIEW		Cost=52	CPU cost=32037059	IO cost=48	Cardinality=7716	Bytes=100308	
     UNION-ALL							
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	
      FILTER							
       HASH JOIN RIGHT OUTER		Cost=9	CPU cost=5339510	IO cost=8	Cardinality=1286	Bytes=111882	
        TABLE ACCESS FULL	Object name=OP_MAIL_VISIT_17_1104	Cost=2	CPU cost=7121	IO cost=2	Cardinality=1	Bytes=48	
        VIEW		Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=50154	
         TABLE ACCESS FULL	Object name=SYS_TEMP_0FD9D660B_D73396C9	Cost=6	CPU cost=919767	IO cost=6	Cardinality=1286	Bytes=136316	

[海鲨数据库架构师:曾凡坤.姒]

      select * 
        from tmp_actionid_lst a
        where not exists
        (
         select 1
         from [e-mail]   b
         where  a.action_id=b.actionid and a.opcontent_id=b.opcontentid
          and   b.inserttime >=trunc(to_date('20110510000000','yyyymmddhh24miss'))
          and   b.inserttime <trunc(to_date('20110510000000','yyyymmddhh24miss'))+1
         )

这个不错