一、诡异dblink起源
今天开发拿过来一条sql，说有诡异现象，sql如下

INSERT INTO TAB_CS_CALLLOG
select c.user_logon,/*工号*/c.user_name,/*姓名*/a.call_id,/*通话id*/
      a.caller_nbr,/*主叫号码*/a.called_nbr,/*被叫号码*/d.start_time,/*呼入时间*/
      b.call_time,/*接听时间*/b.end_time,/*结束时间*/b.call_dura,/*时长*/
      to_number(b.call_time-d.start_time)*24*3600,/*等待时长*/
      decode(c.user_logon,null,0,1),/*类型*/
case
        when substr (a.CALLED_NBR,7, 2) = '00'
          then '广东移动'
        when substr (a.CALLED_NBR,7, 2) = '01'
          then '浙江移动'
        when substr (a.CALLED_NBR,7, 2) = '02'
          then '福建'
        when substr (a.CALLED_NBR,7, 2) = '03'
          then '四川'
        when substr (a.CALLED_NBR,7, 2) = '04'
          then '河南'
        when substr (a.CALLED_NBR,7, 2) = '05'
          then '湖北'
        when substr (a.CALLED_NBR,7, 2) = '06'
          then '北京CSIP'
        when substr (a.CALLED_NBR,7, 2) = '07'
          then '陕西'
        when substr (a.CALLED_NBR,7, 2) = '08'
          then '吉林'
        when substr (a.CALLED_NBR,7, 2) = '09'
          then '江西'
        when substr (a.CALLED_NBR,7, 2) = '10'
          then '宁夏'
        when substr (a.CALLED_NBR,7, 2) = '11'
          then '太原'
        when substr (a.CALLED_NBR,7, 2) = '12'
          then '江苏移动'
        when substr (a.CALLED_NBR,7, 2) = '13'
          then 'e掌管'
        when substr (a.CALLED_NBR,7, 2) = '15'
          then 'e多商'
        when substr (a.CALLED_NBR,7, 2) = '18'
          then '江苏联通'
      end 区域
from cscnew.a@cs a,cscnew.b@cs b,cscnew.c@cs c,cscnew.d@cs d
where a.call_serial=b.call_serial(+)
and b.call_serial=d.call_serial(+)
and b.user_id=c.user_id(+)
and substr(a.CALLED_NBR,1, 6) = '951654'
and  length (a.CALLED_NBR) = 15
and b.fail_reason is null
and a.end_time>=to_date('20110822000000','yyyymmddhh24miss')
and a.end_time<to_date('20110823000000','yyyymmddhh24miss');

然后我进行了测试，证实了她所说的诡异：
1、直接执行select语句需要1.7S左右，但是加上insert inot后，执行时间需要6分钟
2、直接select结果集为602条，加上insert into后，结果集为598条（少4条），如果直接执行select，除掉and b.fail_reason is null限制条件也刚好602条
二、查询相关资料，得到dblink的一些解释
1、dblink执行有两种方式，一种是在远处数据库执行完，然后结果返回，另一种是把远程的表下载到本来，然后执行
2、如果把远程的表下载到本地，空值或者null可能会发生变化（怀疑是空值转化为null，未证实）
三、根据这些解释，进行猜想
1、只执行select的时候，应该是在远程执行完，传输结果回来；而执行insert into的时候，是把远程的表全部下载到本地，然后执行出结果，而数据量本身比较大，所以比较慢
2、在把表从远程下载到本地的过程中，fail_reason 列的null值可以发生了变化，或者空值变为了null，所以数据多了4条
四、事实证明猜想
1、查询远程表大小，发现a表50m，b表400m，c表10m，d表100m左右，传输过来需要一定的时间
2、既然猜测是由于要把表传输到本地而导致这样的结果产生，那么处理方法就是让程序在远程计算出结果，然后传输到本地，查询了一些资料，上面说insert into会导致driving_site提示无效，那么我想到一个用视图的办法解决这个问题：在目标端建立一个关于本查询中无参数的视图，然后在本地通过dblink调用视图，这样总该先在远程执行出结果传输到本地了吧。
2.1）建立目标端视图

create or replace view v_tab
select c.user_logon,/*工号*/c.user_name,/*姓名*/a.call_id,/*通话id*/
      a.caller_nbr,/*主叫号码*/a.called_nbr,/*被叫号码*/d.start_time,/*呼入时间*/
      b.call_time,/*接听时间*/b.end_time,/*结束时间*/b.call_dura,/*时长*/
      to_number(b.call_time-d.start_time)*24*3600,/*等待时长*/
      decode(c.user_logon,null,0,1),/*类型*/
case
        when substr (a.CALLED_NBR,7, 2) = '00'
          then '广东移动'
        when substr (a.CALLED_NBR,7, 2) = '01'
          then '浙江移动'
        when substr (a.CALLED_NBR,7, 2) = '02'
          then '福建'
        when substr (a.CALLED_NBR,7, 2) = '03'
          then '四川'
        when substr (a.CALLED_NBR,7, 2) = '04'
          then '河南'
        when substr (a.CALLED_NBR,7, 2) = '05'
          then '湖北'
        when substr (a.CALLED_NBR,7, 2) = '06'
          then '北京CSIP'
        when substr (a.CALLED_NBR,7, 2) = '07'
          then '陕西'
        when substr (a.CALLED_NBR,7, 2) = '08'
          then '吉林'
        when substr (a.CALLED_NBR,7, 2) = '09'
          then '江西'
        when substr (a.CALLED_NBR,7, 2) = '10'
          then '宁夏'
        when substr (a.CALLED_NBR,7, 2) = '11'
          then '太原'
        when substr (a.CALLED_NBR,7, 2) = '12'
          then '江苏移动'
        when substr (a.CALLED_NBR,7, 2) = '13'
          then 'e掌管'
        when substr (a.CALLED_NBR,7, 2) = '15'
          then 'e多商'
        when substr (a.CALLED_NBR,7, 2) = '18'
          then '江苏联通'
      end 区域
from cscnew.a a,cscnew.b b,cscnew.c c,cscnew.d d
where a.call_serial=b.call_serial(+)
and b.call_serial=d.call_serial(+)
and b.user_id=c.user_id(+)
and substr(a.CALLED_NBR,1, 6) = '951654'
and  length (a.CALLED_NBR) = 15
and b.fail_reason is null

2.2）本地调用远程视图

INSERT INTO TAB_CS_CALLLOG
select * from v_tab@cs a where
a.end_time>=to_date('20110822000000','yyyymmddhh24miss')
and a.end_time<to_date('20110823000000','yyyymmddhh24miss');

2.3）执行结果2.3S完成数据插入，而且条数也是598条，证明我的猜想是正确的，更重要的是解决了今天这个让人疑惑的问题