一种并发控制的精典案例

并发控制一直是个比较头痛的问题，下文演示了一种并发案例，希望大家能从中有所收获。

SQL> create t(id number,nick varchar2(32),status number,email varchar2(32),gmt_modified date);

Table created.

SQL> desc t;
 Name                                      Null?    Type
 ----------------------------------------- -------- ----------------------------
 ID                                                 NUM
 K                                               VARCHAR2(32)
 STATUS                                             NUMBER
 EML                                              VARCHAR2(32)
 _MODIFD                                       E

SQL> begin  
  2      for i in 1..5
  3           insert into t values(i,'test',0,'test@taobao.',sysdate);
  4      end loop;
  5      ;
  6  end;
  7  /

QL procedure successfully completed.

SQL> alter nls_date_='YYYY-MM-DD HH24:MI:SS';

Session altered.

SQL> select * from t;

ID NK       STATUS EMAIL               GMT_MODIFIED
-- ------ ---------- ------------------- -------------------
 1 test            0 [e-mail]     2008-04-09 13:28:27
 2 test            0 [e-mail]     2008-04-09 13:28:27
 3 test            0 [e-mail]     2008-04-09 13:28:27
 4 test            0 [e-mail]     2008-04-09 13:28:27
 5 test            0 [e-mail]     2008-04-09 13:28:27
        
--session 1执行,但不提交        
SQL> update t set status=1,gmt_modified=sysdate where id in (
  2                                         select id from (select id from t where nick='test' and status=0 order by id asc)
  3                                                          where rownum < 2)
                                           and status=0;

1 updated.

SQL> select * from t;

ID NICK       STATUS EMAIL                GMT_MODIFIED
-- ------ ---------- -------------------- -------------------
 1 test            1 [e-mail]      2008-04-09 13:36:31
 2 test            0 [e-mail]      2008-04-09 13:28:27
 3 test            0 [e-mail]      2008-04-09 13:28:27
 4 test            0 [e-mail]      2008-04-09 13:28:27
 5 test            0 [e-mail]      2008-04-09 13:28:27

--session 2也执行这样的语句,出现等待。
SQL> update t set status=1,gmt_modified=sysdate where id in (
  2                                         select id from (select id from t where nick='test' and status=0 order by id asc)
  3                                                          where rownum < 2)
                                                and status=0;

--session 1提交
commit;

--session 2执行完成，查看结果
SQL> select * from t;

 ID NICK      STATUS EMAIL               GMT_MODIFIED
--- ----- ---------- ------------------- -------------------
  1 test           1 [e-mail]     2008-04-09 13:48:30  --session 1更新了这条
  2 test           1 [e-mail]     2008-04-09 13:48:37  --session 2更新了这条
  3 test           0 [e-mail]     2008-04-09 13:28:27
  4 test           0 [e-mail]     2008-04-09 13:28:27
  5 test           0 [e-mail]     2008-04-09 13:28:27

　　上面运行的更新语句

update t set status=1,gmt_modified=sysdate where id in (                                                              
                                       select id from (select id from t where nick='test' and status=0 order by id asc)
                                                        where rownum < 2)                                             
                                           and status=0;  --注意此条件

　　当session 1执行，但并不提交后，session 2执行相同的语句，由于update语句中有select子查询，众所周知，select要进行一致性读，即也会读取到记录1,但这又是一个update语句，会进行当前读，session 2也想去更新这一行，但发现这一行已被另外一个事务更新，所以出现等待。但当session 1提交后，通过以上实验，证明session 2会再次触发一致性读与当前读，去更新了满足条件的第二行,而不是第一行。

　　大家还可以做其它实验，比如说session 1执行如下的语句：

update t set status=1,gmt_modified=sysdate where id in (                                                              
                                       select id from (select id from t where nick='test' and status=0 order by id asc)
                                                        where rownum < 2)                                             
                                           and status=0;  --注意此条件

　　session 2执行如下的语句：

update t set status=1,gmt_modified=sysdate where id in (                                                              
                                       select id from (select id from t where nick='test' and status=0 order by id asc)
                                                        where rownum < 4)                                             
                                           and status=0;  --注意此条件

　　结果是sesion 1更新了1条;session 2更新了2条,而session 2实际想更新3条。

　　本篇实验也提供了一个很好处理并发的方法,使用此方法，可以保证多个更新在并发时不会更新同一行，不过要在更新后，加一个判断，如果实际更新的条数 != 想要更新的条数(即sql语句中的rownum),那就说明出现了并发，并且剩余的满足条件的记录数不满足应用要求，这时候就需要在应用层加上进一步的并发处理了。我们在实际中出现的一个案例，恰好就是没有对实际更新的条目进行判断，导致出现了应用层数据逻辑错误。

【责编:Ken】

--------------------next---------------------