oracle 死锁和锁等待的区别(转载)

oracle 死锁和锁等待的区别(转载)

所谓的锁等待:就是一个事务a对一个数据表进行ddl或是dml操作时，系统就会对该表加上表级的排它锁，此时其他的事务对该表进行操作的时候会等待a提交或是回滚后，才可以继续b的操作

所谓的死锁：当两个或多个用户相互等待锁定的数据时就会发生死锁，这时这些用户被卡在不能继续处理业务，oracle可以自动检测死锁并解决他们，通过回滚一个死锁中的语句，释放锁定的数据，回滚的话会遇到ora-00060 deadlock detected while waiting for resource

模拟锁等待：

   两个事务a和b，分别创建t1,t2,并且初始化一条数据，

   a 更改t1的数据，此时并不提交，这时候b更改相同的一列，此时b一直处于等待的状态

我们可以查询锁等待的内容：

wait_lock.sql

select
      (select username from v$session where sid = a.sid) username,
      a.sid,
      (select serial# from v$session where sid = a.sid) serial#,
      a.type,
      a.id1,
      a.id2,
      a.lmode,
      a.request,
      a.block,
      b.sid blocking_sid
from v$lock a,
      ( select * from v$lock
        where request > 0
        and type <> 'MR'
      ) b
where a.id1 = b.id1(+)
   and a.id2 = b.id2(+)
   and a.lmode > 0
   and a.type <> 'MR'
order by username,a.sid,serial#,a.type

此时可以查询到锁等待的现象，最后一列不为空的就是等待的事件

此时我们可以跟a用户提示让其提交事务或是回滚，也可以直接杀死

alter system kill session 'sid,serial#';

保持现状不动，在a事务更改t2表此时在a事务会产生

SQL> update t1 set id=1000 where id=1;
update t1 set id=1000 where id=1
        *
第 1 行出现错误:
ORA-00060: 等待资源时检测到死锁

此时oracle已经帮我解决了这个死锁问题

死锁的产生需要四个必须的条件：

1 ，mutual execution(互斥)资源不能被共享，只能由一个进程使用

2，hold and wait(请求并持续)已经得到资源的进程可以再次申请新的资源

3，no pre-emption(不可剥夺)已经分配的资源不能被相应的进程强制剥夺

4，circular wait(循环等待条件)系统中若干进程组成环路，该环路中的每个进程都在等待相邻进程正占用的资源

定位死锁：

  系统级别的定位

    Select username,lockwait,status,machine,program from v$session where sid in (select session_id from v$locked_object)

      Username死锁的用户，lockwait死锁的状态，status中active表示死锁，machine死锁所在的机器，program死锁来自于那个程序

   语句级别的定位

      Select sql_text from v$sql where hash_value in (select sql_hash_value from v$session where sid in (select session_id from v$locked_object));

  进程级别的定位

   Select s.username,l.object_id,l.session_id,s.serial#,l.oracle_usrename,l.os_user_name,l.process from v$locked_object l,v$session s where l.session_id=s.sid;

处理死锁的一般策略

   1，鸵鸟算法忽略该问题

   2，定位死锁并且恢复

  3，仔细对资源进行动态分配，避免死锁

   4，破坏死锁中的一个条件

如果oracle解决不了的死锁，我们需要定位到进程级别，找到对应的sid和serial#

alter system kill 'sid,serail#'

失败的话，找到对应的进程强制关闭

Select p.spid from v$session s, v$process p where s.sid=xx and s.paddr=p.addr

ps -ef | grep spid

kill -9 xx



查询oracle的死锁

lock.sql

SELECT    bs.username "Blocking User", bs.username "DB User",
           ws.username "Waiting User", bs.SID "SID", ws.SID "WSID",
           bs.serial# "Serial#", bs.sql_address "address",
           bs.sql_hash_value "Sql hash", bs.program "Blocking App",
           ws.program "Waiting App", bs.machine "Blocking Machine",
           ws.machine "Waiting Machine", bs.osuser "Blocking OS User",
           ws.osuser "Waiting OS User", bs.serial# "Serial#",
           ws.serial# "WSerial#",
           DECODE (wk.TYPE,
                   'MR', 'Media Recovery',
                   'RT', 'Redo Thread',
                   'UN', 'USER Name',
                   'TX', 'Transaction',
                   'TM', 'DML',
                   'UL', 'PL/SQL USER LOCK',
                   'DX', 'Distributed Xaction',
                   'CF', 'Control FILE',
                   'IS', 'Instance State',
                   'FS', 'FILE SET',
                   'IR', 'Instance Recovery',
                   'ST', 'Disk SPACE Transaction',
                   'TS', 'Temp Segment',
                   'IV', 'Library Cache Invalidation',
                   'LS', 'LOG START OR Switch',
                   'RW', 'ROW Wait',
                   'SQ', 'Sequence Number',
                   'TE', 'Extend TABLE',
                   'TT', 'Temp TABLE',
                   wk.TYPE
                  ) lock_type,
           DECODE (hk.lmode,
                   0, 'None',
                   1, 'NULL',
                   2, 'ROW-S (SS)',
                   3, 'ROW-X (SX)',
                   4, 'SHARE',
                   5, 'S/ROW-X (SSX)',
                   6, 'EXCLUSIVE',
                   TO_CHAR (hk.lmode)
                  ) mode_held,
           DECODE (wk.request,
                   0, 'None',
                   1, 'NULL',
                   2, 'ROW-S (SS)',
                   3, 'ROW-X (SX)',
                   4, 'SHARE',
                   5, 'S/ROW-X (SSX)',
                   6, 'EXCLUSIVE',
                   TO_CHAR (wk.request)
                  ) mode_requested,
           TO_CHAR (hk.id1) lock_id1, TO_CHAR (hk.id2) lock_id2,
           DECODE
              (hk.BLOCK,
               0, 'NOT Blocking',          /**//* Not blocking any other processes */
               1, 'Blocking',              /**//* This lock blocks other processes */
               2, 'Global',           /**//* This lock is global, so we can't tell */
               TO_CHAR (hk.BLOCK)
              ) blocking_others
      FROM v$lock hk, v$session bs, v$lock wk, v$session ws
     WHERE hk.BLOCK = 1
       AND hk.lmode != 0
       AND hk.lmode != 1
       AND wk.request != 0
       AND wk.TYPE(+) = hk.TYPE
       AND wk.id1(+) = hk.id1
       AND wk.id2(+) = hk.id2
       AND hk.SID = bs.SID(+)
       AND wk.SID = ws.SID(+)
       AND (bs.username IS NOT NULL)
       AND (bs.username <> 'SYSTEM')
       AND (bs.username <> 'SYS')
ORDER BY 1;

这些语句的执行最好是在plsql或是sqldeveloper如果是直接在数据库里面执行的需要格式化表，否则会很让你眼花的。