[原]ORA-00060: Deadlock detected(场景1:单表并发更新)

      先说说什么是死锁(Deadlock),关于死锁的定义google、baidu可以轻易查到,我也不想引经据典,我用一个简单故事来说明一下死锁。

      话说一个风和日丽的下午,小明和小强打架,由于情节严重被老师批评教育不说还得放学后写悔过书,大家知道写悔过书要纸和笔,恰巧在刚才那场大战中小明将小强全部作业本和草稿纸撕碎,而小强将小明所有笔摔个稀巴烂,现在两人势如水火,别说借东西,话都不想说,于是就这么等啊等,等到老师来看他们的悔过书写得怎么样,结果当然是“还没开始写”,于是老师说:“小明你把纸借给小强,让小强写”。

      哈哈整个故事有点牵强,但这就是一个经典死锁场景,如果老师不出来调停,他两就只能这样等下去,俗语叫“等死”了。

 

      说回正题,Oracle 特殊的锁管理模式使发生死锁的几率大大减少,但是,要相信“一切皆有可能”,以后几篇博客对会分析导致 ORA-00060 的各种场景和处理方法。

      今天说说第一个场景,也是网上能找到最多的场景,我都不太好意思在标题上加个“[原]”标志了。

      还是使用经典的 scott demo 吧:

      开两个会话,看看各自的sid:

select sid from v$mystat where rownum=1;

      我这里两个会话的 sid 分别是 126 和128 。看看示例数据:

scott$mydb@test02 SQL> set pagesize 50 ; 
scott$mydb@test02 SQL> select * from emp ; 

     EMPNO ENAME      JOB              MGR HIREDATE                   SAL       COMM     DEPTNO
---------- ---------- --------- ---------- ------------------- ---------- ---------- ----------
      7369 SMITH      CLERK           7902 1980-12-17 00:00:00        800                    20
      7499 ALLEN      SALESMAN        7698 1981-02-20 00:00:00       1600        300         30
      7521 WARD       SALESMAN        7698 1981-02-22 00:00:00       1250        500         30
      7566 JONES      MANAGER         7839 1981-04-02 00:00:00       2975                    20
      7654 MARTIN     SALESMAN        7698 1981-09-28 00:00:00       1250       1400         30
      7698 BLAKE      MANAGER         7839 1981-05-01 00:00:00       2850                    30
      7782 CLARK      MANAGER         7839 1981-06-09 00:00:00       2450                    10
      7788 SCOTT      ANALYST         7566 1987-04-19 00:00:00       3000                    20
      7839 KING       PRESIDENT            1981-11-17 00:00:00       5000                    10
      7844 TURNER     SALESMAN        7698 1981-09-08 00:00:00       1500          0         30
      7876 ADAMS      CLERK           7788 1987-05-23 00:00:00       1100                    20
      7900 JAMES      CLERK           7698 1981-12-03 00:00:00        950                    30
      7902 FORD       ANALYST         7566 1981-12-03 00:00:00       3000                    20
      7934 MILLER     CLERK           7782 1982-01-23 00:00:00       1300                    10

      我想为工资(SAL)最低的两个人(ENAME:Smith、James EMPNO:7369、7900)加工资1元(我承认这次工资的涨幅最不上CPI),而另一位毫不知情的 Manager Blake 也想给他们加1元的工资,我在 sid 为 126 进行操作,而 Blake 在 sid 为 128 的会话中操作,执行顺序如下表:

        +--------------------------------------------+--------------------------------------------+
        |         Session 1 (sid=126)                |            Session 2 (sid=128)             |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7369; |                                            |
        +--------------------------------------------+--------------------------------------------+
        |                                            | update emp set sal=sal+1 where empno=7900; |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7900; |                                            |
        +--------------------------------------------+--------------------------------------------+
        |                                            | update emp set sal=sal+1 where empno=7369; |
        +--------------------------------------------+--------------------------------------------+
        | ORA-00060: deadlock detected               |                                            |
        |            while waiting for resource      |                                            |
        +--------------------------------------------+--------------------------------------------+

      这样我们就成功地触发了一个ORA-00060,从aler文件中可以看到一条类似如下的信息:

ORA-00060: Deadlock detected. More info in file /u01/app/admin/mydb/udump/mydb_ora_7531.trc.

      我们看看 trc 文件,其中最有用的一部分是 Deadlock graph :

[Transaction Deadlock]
Current SQL statement for this session:
update emp set sal=sal+1 where empno=7900
The following deadlock is not an ORACLE error. It is a
deadlock due to user error in the design of an application
or from issuing incorrect ad-hoc SQL. The following
information may aid in determining the deadlock:
Deadlock graph:
                       ---------Blocker(s)--------  ---------Waiter(s)---------
Resource Name          process session holds waits  process session holds waits
TX-00010013-00005f96        32     126     X             34     128           X
TX-00030001-000065ed        34     128     X             32     126           X
session 126: DID 0001-0020-00003C79     session 128: DID 0001-0022-0000772F
session 128: DID 0001-0022-0000772F     session 126: DID 0001-0020-00003C79
Rows waited on:
Session 128: obj - rowid = 0000C7CF - AAAMfPAAEAAAAAgAAA
  (dictionary objn - 51151, file - 4, block - 32, slot - 0)
Session 126: obj - rowid = 0000C7CF - AAAMfPAAEAAAAAgAAL
  (dictionary objn - 51151, file - 4, block - 32, slot - 11)
Information on the OTHER waiting sessions:
Session 128:
  pid=34 serial=31980 audsid=319634 user: 54/SCOTT
  O/S info: user: oracle, term: pts/8, ospid: 22771, machine: test02
            program: sqlplus@test02 (TNS V1-V3)
  application name: SQL*Plus, hash value=3669949024
  Current SQL Statement:
  update emp set sal=sal+1 where empno=7369
End of information on OTHER waiting sessions.

      大家可以看到引起死锁的语句(一个巴掌拍不响,一般死锁都要2条或以上的语句才能引起死锁),还有注意产生的object,这里的objn是51151,正是emp表:

scott$mydb@test02 SQL> select object_id,object_name from user_objects where object_name='EMP';   

 OBJECT_ID OBJECT_NAME
---------- ------------------------------
     51151 EMP

      如果八卦一点,还可以看看引起死锁的相关者都在等什么而“等死”了:

Rows waited on:
Session 128: obj - rowid = 0000C7CF - AAAMfPAAEAAAAAgAAA
  (dictionary objn - 51151, file - 4, block - 32, slot - 0)
Session 126: obj - rowid = 0000C7CF - AAAMfPAAEAAAAAgAAL
  (dictionary objn - 51151, file - 4, block - 32, slot - 11)

      session 128 (也就是session 2,sid=128) 等着要ROWID=AAAMfPAAEAAAAAgAAA的行锁,而session 126 (也就是session 1,sid=126)等着要ROWID=AAAMfPAAEAAAAAgAAL的行,验证一下:

scott$mydb@test02 SQL> select rowid,empno from emp where empno in (7369,7900) ;

ROWID                   EMPNO
------------------ ----------
AAAMfPAAEAAAAAgAAA       7369
AAAMfPAAEAAAAAgAAL       7900

      对照一下上面的表格,注意一下 update 语句的顺序,你就明白了。

      大家可以类比一下之前所说的故事,假设empno为 7369 和 7900 的行是纸和笔,session 1 和 session 2是小明和小强,最后老师Oracle跑出来调停。

 

      解决死锁的方法之一是给资源编号,然后按照固定的顺序进行访问,简单来说,就是先改编号小的再改编号大的(翻过来亦然),如下表:

        +--------------------------------------------+--------------------------------------------+
        |         Session 1 (sid=126)                |            Session 2 (sid=128)             |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7369; |                                            |
        +--------------------------------------------+--------------------------------------------+
        |                                            | update emp set sal=sal+1 where empno=7369; |
        |                                            | Waiting....                                |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7900; |                                            |
        +--------------------------------------------+--------------------------------------------+
        | commit/rollback;                           |                                            |
        +--------------------------------------------+--------------------------------------------+
        |                                            | update emp set sal=sal+1 where empno=7900; |
        +--------------------------------------------+--------------------------------------------+

      大家可以看到加了两次工资,而且session 2 被 session 1 阻塞了。对于用户感受不好,如果session 1 一直结束事务(commit/rollback),session 2 只能一直等啊等,比deadlock后,oracle出面调停还要糟糕,那怎么办呢?

      可以 select … for update nowait 语句测试一下需要更改的行是否被锁定,如果没有被锁定那这个语句会马上给这行加锁,如果该已经被锁定那就马上返回 ORA-00054: resource busy and acquire with NOWAIT specified ,如下表所示:

        +--------------------------------------------+--------------------------------------------+
        |         Session 1 (sid=126)                |            Session 2 (sid=128)             |
        +--------------------------------------------+--------------------------------------------+
        | select * from emp where empno in(7369,7900)|                                            |
        | for update nowait ;                        |                                            |
        +--------------------------------------------+--------------------------------------------+
        |                                            | select * from emp where empno in(7369,7900)|
        |                                            | for update nowait ;                        |
        |                                            | ORA-00054: resource busy and acquire       |
        |                                            |            with NOWAIT specified           |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7369; |                                            |
        +--------------------------------------------+--------------------------------------------+
        | update emp set sal=sal+1 where empno=7900; |                                            |
        +--------------------------------------------+--------------------------------------------+ 

      有人可能会说,这样做我的程序改动太大了,毕竟要引入一个select … for update nowait 和 ORA-00054 的判断,有没有更好的办法呢?

      有,更经典的处理死锁的算法——“鸵鸟算法”,简单来说就是“不管”,反正Oracle最终会出来调停的。

posted @ 2010-09-12 23:57  killkill  阅读(10381)  评论(3编辑  收藏  举报