【转】外键缺少索引引发的死锁
客户的10.2.0.4 RAC for AIX环境频繁出现ORA-60死锁问题，导致应用程序无法顺利执行。
经过一系列的诊断，发现最终问题是由于外键上没有建立索引所致，由于程序在主子表上删除数据，缺少索引导致行级锁升级为表级锁，最终导致大量的锁等待和死锁。
下面通过一个例子简单模拟一下问题：
SQL> create table t_p (id number primary key, name varchar2(30));
Table created.
SQL> create table t_f (fid number, f_name varchar2(30), foreign key (fid) references t_p);
Table created.
SQL> insert into t_p values (1, 'a');
1 row created.
SQL> insert into t_f values (1, 'a');
1 row created.
SQL> insert into t_p values (2, 'b');
1 row created.
SQL> insert into t_f values (2, 'c');
1 row created.
SQL> commit;
Commit complete.
SQL> delete t_f where fid = 2;
1 row deleted.
这时在会话2同样对子表进行删除：
SQL2> delete t_f where fid = 1;
1 row deleted.
回到会话1执行主表的删除：
SQL> delete t_p where id = 2;
会话被锁，回到会话2执行主表的删除：
SQL2> delete t_p where id = 1;
会话同样被锁，这时会话1的语句被回滚，出现ORA-60死锁错误：
delete t_p where id = 2
*
ERROR at line 1:
ORA-00060: deadlock detected while waiting for resource
SQL> rollback;
Rollback complete.
将会话1操作回滚，会话2同样回滚并建立外键列上的索引：
1 row deleted.
SQL2> rollback;
Rollback complete.
SQL2> create index ind_t_f_fid on t_f(fid);
Index created.
重复上面的步骤会话1删除子表记录：
SQL> delete t_f where fid = 2;
1 row deleted.
会话2删除子表记录：
SQL2> delete t_f where fid = 1;
1 row deleted.
会话1删除主表记录：
SQL> delete t_p where id = 2;
1 row deleted.
会话2删除主表记录：
SQL> delete t_p where id = 1;
1 row deleted.
所有的删除操作都可以成功执行，关于两种情况下锁信息的不同这里就不深入分析了，重点就是在外键列上建立索引。
虽然有一些文章提到过，如果满足某些情况，可以不在外键列上建立的索引，但是我的观点一向是，既然创建了外键，就不要在乎再多一个索引，因为一个索引所增加的代价，与缺失这个索引所带来的问题相比，是微不足道的。
参考：
http://yangtingkun.itpub.net/post/468/524858
http://www.cnblogs.com/angzi/archive/2006/12/09/587612.html
【补充】Oracle 10g和Oracle 9i trc日志内容的差别
Oracle 10g 10.2.0.3.0：
Oracle 9i 9.2.0.7.0：
经过一系列的诊断，发现最终问题是由于外键上没有建立索引所致，由于程序在主子表上删除数据，缺少索引导致行级锁升级为表级锁，最终导致大量的锁等待和死锁。
下面通过一个例子简单模拟一下问题：
SQL> create table t_p (id number primary key, name varchar2(30));
Table created.
SQL> create table t_f (fid number, f_name varchar2(30), foreign key (fid) references t_p);
Table created.
SQL> insert into t_p values (1, 'a');
1 row created.
SQL> insert into t_f values (1, 'a');
1 row created.
SQL> insert into t_p values (2, 'b');
1 row created.
SQL> insert into t_f values (2, 'c');
1 row created.
SQL> commit;
Commit complete.
SQL> delete t_f where fid = 2;
1 row deleted.
这时在会话2同样对子表进行删除：
SQL2> delete t_f where fid = 1;
1 row deleted.
回到会话1执行主表的删除：
SQL> delete t_p where id = 2;
会话被锁，回到会话2执行主表的删除：
SQL2> delete t_p where id = 1;
会话同样被锁，这时会话1的语句被回滚，出现ORA-60死锁错误：
delete t_p where id = 2
*
ERROR at line 1:
ORA-00060: deadlock detected while waiting for resource
SQL> rollback;
Rollback complete.
将会话1操作回滚，会话2同样回滚并建立外键列上的索引：
1 row deleted.
SQL2> rollback;
Rollback complete.
SQL2> create index ind_t_f_fid on t_f(fid);
Index created.
重复上面的步骤会话1删除子表记录：
SQL> delete t_f where fid = 2;
1 row deleted.
会话2删除子表记录：
SQL2> delete t_f where fid = 1;
1 row deleted.
会话1删除主表记录：
SQL> delete t_p where id = 2;
1 row deleted.
会话2删除主表记录：
SQL> delete t_p where id = 1;
1 row deleted.
所有的删除操作都可以成功执行，关于两种情况下锁信息的不同这里就不深入分析了，重点就是在外键列上建立索引。
虽然有一些文章提到过，如果满足某些情况，可以不在外键列上建立的索引，但是我的观点一向是，既然创建了外键，就不要在乎再多一个索引，因为一个索引所增加的代价，与缺失这个索引所带来的问题相比，是微不足道的。
参考：
http://yangtingkun.itpub.net/post/468/524858
http://www.cnblogs.com/angzi/archive/2006/12/09/587612.html
【补充】Oracle 10g和Oracle 9i trc日志内容的差别
Oracle 10g 10.2.0.3.0：
DEADLOCK DETECTED ( ORA-00060 )
[Transaction Deadlock]
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
TM-0000dd55-00000000 16 146 SX SSX 17 148 SX SSX
TM-0000dd55-00000000 17 148 SX SSX 16 146 SX SSX
session 146: DID 0001-0010-00000008 session 148: DID 0001-0011-00000006
session 148: DID 0001-0011-00000006 session 146: DID 0001-0010-00000008
Rows waited on:
Session 148: no row
Session 146: no row
Information on the OTHER waiting sessions:
Session 148:
pid=17 serial=39 audsid=540046 user: 54/SCOTT
O/S info: user: SKYHOME\sky, term: SKYHOME, ospid: 3028:7000, machine: WORKGROUP\SKYHOME
program: plsqldev.exe
application name: PL/SQL Developer, hash value=1190136663
action name: Command Window - New, hash value=254318129
Current SQL Statement:
delete t_p where id = 1
End of information on OTHER waiting sessions.
Current SQL statement for this session:
delete t_p where id = 2
Oracle 9i 9.2.0.7.0：
DEADLOCK DETECTED
Current SQL statement for this session:
delete t_p where id = 2
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
TM-0000260e-00000000 21 51 SX SSX 23 20 SX SSX
TM-0000260e-00000000 23 20 SX SSX 21 51 SX SSX
session 51: DID 0001-0015-0000043D session 20: DID 0001-0017-00000397
session 20: DID 0001-0017-00000397 session 51: DID 0001-0015-0000043D
Rows waited on:
Session 20: no row
Session 51: no row
Information on the OTHER waiting sessions:
Session 20:
pid=23 serial=53179 audsid=197296 user: 87/scott
O/S info: user: sky, term: SKYHOME, ospid: 5540:4984, machine: WORKGROUP\SKYHOME
program: plsqldev.exe
client info: 127.0.0.1
application name: PL/SQL Developer, hash value=1190136663
action name: Command Window - New, hash value=254318129
Current SQL Statement:
delete t_p where id = 1
End of information on OTHER waiting sessions.