Oracle shutdown 过程中 DB hang住解决方法

一.故障处理过程

数据库版本：Oracle10.2.0.5.4

操作系统： Redhat5.4

中午同事对表添加字段，然后造成70多个存储等对象无效，系统不能正常运行，查看了一下同事添加字段的表，并没有锁没有释放。

尝试编译无效对象，部分可以编译，部分不能编译，对象一直处于编译状态。Kill掉其他的session 后也是无法编译，因为影响业务，决定重启数据库。

--关闭DB:

[oracle@qs-xezf-db1 ~]$ sqlplus / as sysdba

SQL*Plus: Release 10.2.0.5.0 - Productionon Thu Mar 29 13:23:14 2012

Connected to:

Oracle Database 10g Enterprise EditionRelease 10.2.0.5.0 - 64bit Production

With the Partitioning, OLAP, Data Miningand Real Application Testing options

SYS@xezf(qs-xezf-db1)> alter systemswitch logfile;

System altered.

SYS@xezf(qs-xezf-db1)> alter systemswitch logfile;

System altered.

SYS@xezf(qs-xezf-db1)> alter systemswitch logfile;

System altered.

--个人习惯，关闭db前先切换下日志。

SYS@xezf(qs-xezf-db1)> shutdownimmediate

--一直挂在这

--跟踪LOG：

[oracle@qs-xezf-db1 bdump]$ tail -falert_xezf.log

Stopping background process QMNC

Thu Mar 29 13:24:08 CST 2012

Stopping background process CJQ0

Thu Mar 29 13:24:10 CST 2012

Stopping background process MMNL

Thu Mar 29 13:24:11 CST 2012

Stopping background process MMON

License high water mark = 201

Thu Mar 29 13:24:12 CST 2012

Job queue slave processes stopped

Thu Mar 29 13:29:11 CST 2012

Active call for process 11403 user 'oracle'program 'oracle@qs-xezf-db1'

System State dumped to trace file/u01/app/oracle/admin/xezf/udump/xezf_ora_13569.trc

Thu Mar 29 13:29:29 CST 2012

SHUTDOWN: waiting for active calls tocomplete.

Thu Mar 29 13:44:15 CST 2012

MMNL absent for 1207 secs; Foregroundstaking over

--等了20分钟的时间，最后还是挂住了，在这个状态等了10分钟，还是不能结束。

决定强制关闭DB:

[oracle@qs-xezf-db1 ~]$ sqlplus / as sysdba

SQL*Plus: Release 10.2.0.5.0 - Productionon Thu Mar 29 13:53:13 2012

Connected.

SQL> shutdown abort

ORACLE instance shut down.

--启动DB:

SQL> startup

ORACLE instance started.

Total System Global Area 5251268608 bytes

Fixed Size 2103600 bytes

Variable Size 1275070160 bytes

Database Buffers 3959422976 bytes

Redo Buffers 14671872 bytes

Database mounted.

Database opened.

SQL>

--在编译无效对象，系统恢复正常。

二．分析过程

2.1 分析system state dump

在关闭DB的时候，自动生成了一个system dump 文件，使用ass.awk 分析：

Oracle 使用ass.awk 工具查看system state dump 说明

http://blog.csdn.net/tianlesoftware/article/details/7237729

[oracle@qs-xezf-db1 src]$ awk -f ass109.awk/u01/app/oracle/admin/xezf/udump/xezf_ora_13569.trc

Starting Systemstate 1

.............................................

Ass.Awk Version 1.0.9 - Processing/u01/app/oracle/admin/xezf/udump/xezf_ora_13569.trc

System State 1

~~~~~~~~~~~~~~~~

2: waiting for 'pmon timer'

3: waiting for 'rdbms ipc message'

4: waiting for 'rdbms ipc message'

5: waiting for 'rdbms ipc message'

6: waiting for 'rdbms ipc message'

7: waiting for 'rdbms ipc message'

8: last wait for 'rdbms ipc message'

9: last wait for 'smon timer'

10: waiting for 'rdbms ipc message'

14:

15:

17: waiting for 'rdbms ipc message'

18: waiting for 'rdbms ipc message'

23: waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

26: last wait for 'ksdxexeotherwait'

38: waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

39: waiting for 'SQL*Net message fromclient'

40: waiting for 'SQL*Net message fromclient'

41: waiting for 'SQL*Net message fromclient'

75: waiting for 'SQL*Net message fromclient'

88: waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

95: waiting for 'SQL*Net message fromclient'

96: waiting for 'SQL*Net message fromclient'

97: waiting for 'SQL*Net message fromclient'

99: waiting for 'SQL*Net message fromclient'

104:waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

106:waiting for 'SQL*Net message fromclient'

107:waiting for 'SQL*Net message fromclient'

108:waiting for 'SQL*Net message fromclient'

117:waiting for 'SQL*Net message fromclient'

Cmd: Update

129:waiting for 'SQL*Net message fromclient'

Cmd: Update

130:waiting for 'SQL*Net message fromclient'

133:waiting for 'SQL*Net message fromclient'

138:waiting for 'SQL*Net message fromclient'

140:last wait for 'db file sequential read'(1,974b,1)

Cmd: Insert

154:waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

159:waiting for 'SQL*Net message fromclient'

160:waiting for 'SQL*Net message fromclient'

162:waiting for 'SQL*Net message fromclient'

171:waiting for 'SQL*Net message fromclient'

196:waiting for 'SQL*Net message fromclient'

204:waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

216:waiting for 'SQL*Net message fromclient'

Cmd: PL/SQL Execute

220:waiting for 'SQL*Net message from client'[Latchfor]

Blockers

~~~~~~~~

Above is a list of all the processes. If they are waiting for a resource

then it will be given in square brackets. Below is a summary of the

waited upon resources, together with the holder of that resource.

Notes:

~~~~~

o A process id of '???' implies that the holder was not found in the

systemstate.

Resource Holder State

Latch for ??? Blocker

Object Names

~~~~~~~~~~~~

Latch for SHUTDOWN: waiting foractive calls to complete.

4767 Lines Processed.

[oracle@qs-xezf-db1 src]$

2.2 分析：SHUTDOWN: waiting for active calls tocomplete.

在此次shutdown过程中，在这一步停了近15分钟。

Thu Mar 29 13:29:11 CST 2012

Active call for process 11403 user 'oracle'program 'oracle@qs-xezf-db1'

System State dumped to trace file/u01/app/oracle/admin/xezf/udump/xezf_ora_13569.trc

Thu Mar 29 13:29:29 CST 2012

SHUTDOWN: waiting for active calls tocomplete.

Thu Mar 29 13:44:15 CST 2012

MMNL absent for 1207 secs; Foregroundstaking over

MOS上解释如下：

Alert Log: Shutdown Waiting for ActiveCalls to Complete [ID 1039389.6]

2.2.1 Problem Description

You areattempting to shut down the database and the database hangs. The

alert log contains the following message:

--我们尝试关闭DB时，数据库hang住了，然后alert log 就会出现如下信息：

SHUTDOWN: Waiting for active calls to complete

There are noother error messages in the alert log.

    并且没有其他的错误信息了。

2.2.2 Solution Description: --解决方法描述

--------------------------------------------------------

Caveat: This note is intended for Unix platforms primarily, but we shouldaddress this issue on Windows platforms as well. On Windows, stopping the service is a validworkaround for these errors.

--------------------------------------------------------

警告：这个方式针对Unix 平台，但是windows 平台同样适用。

Locate and kill any client connections tothe database at the Unix level, as follows:

解决方法是在Unix 系统级别查找并kill 掉任何客户端的连接，具体方法如下：

1. Locate any client connections to thedatabase using ps, and grep for any processes belonging to this .

Example: ps -ef | grep SID

2. Look for processes that include a'Local=No' designation.

Example:osupport 6235 1 0 Nov 24 0:01 oracleSID (LOCAL=NO)

2.2.3 Explanation –造成这个现象的原因

The database iswaiting for pmon to clean up processes, but pmon is unable to clean them. Theclient connections to the server are causing the shutdown immediate or normalto hang. Killing them allows pmon toclean up and release the associated Oracle processes and resources.

--数据库等待PMON 进程清理这些LOCAL=NO的进程，但是PMON 进程不能清理他们，这些进程就会导致数据库shutdown 时hang住，kill 掉这些进程，然后让PMON 来清理和释放其他的数据库进程和资源，完成数据库关闭操作。

关于PMON进程的更多内容，参考：

Oracle 后台进程说明

http://blog.csdn.net/tianlesoftware/article/details/5587788

What resources are we talking about?

--组成DB 关闭的资源包括：

1) Any non committed transactions must berolled back

2) Any temporary space (sort segments /lobs / session temporary tables) must be freed

3) The session itself and any associatedmemory consumed by the session.

4) Internal locks / enqueues must becleaned up

Often Oracle(SMON or PMON depending on whether Shared Server is used) will wait for the OSto terminate the process(es) associated with the session. I the OS never returns, or fails to terminatethem, then the instance shutdown will hang with this message (Shutdown Waitingfor Active Calls to Complete)

Other meansexist to achieve a quick shutdown, as outlined in Note386408.1 - What Is The Fastest Way To Cleanly Shutdown An Oracle Database?

Note that inE-Business Environments, the same messages can be produced if the databaseshutdown process starts before the concurrent manager stops in the applicationserver .

2.3 分析：MMNL absent for 1207 secs; Foregroundstaking over

在我遇到的这次故障中，我等待了10分钟，依旧挂在这，之后就shutdown Abort db了。

MOS 说明：

"MMNL absent for %u secs; Foregroundstaking over" Messages in Alert.log [ID 465891.1]

When there is ahang situation for a long period of time, oracle writes this message tothe alert.log: "MMNL absent for %u secs; Foregrounds takingover".

--当数据库hang住一段时间后，Oracle 就会写这个信息到alert log。

Thu Nov 1 15:20:56 2007
MMNL absent for 3805 secs; Foregrounds taking over
Thu Nov 1 15:58:13 2007
Starting ORACLE instance (normal)

If this is ahang situation during shutdown (after MMNL is stopped) or during databaseopen (when MMNL is not yet started), messages are written to the alert.logevery 5 minutes.

--在数据库关闭过程中（MMNL 已经停止）或者在数据库open时（MMNL 还没有启动），这个消息会每隔5分钟写入一次alert log。

Any completehang situations which take a long time trigger these messages.

2.3.1 Cause –导致原因

These messagesare for informational purposes only to indicate a hang situation which lasted along period of time and MMNL is unable to run its job at that moment. When thehang lasts for several hours before the instance is completely stopped/killed,the tail of the alert.log may flood with these messages. Although the messagetext contains MMNL process, in this case MMNL is actually the victim of thehang issue rather than the cause.

--这个信息唯一的作用是用来说明数据库已经hang住了，并且MMNL 也不能执行它的job。

2.3.2 Solution --解决方法

In case of ahang taking a very long time, there is no way to prevent these messages to bewritten to alert.log. Hang situation must be resolved to stop the messages.Starting with 11.1 version, messages are written to a trace file rather thanthe alert.log to prevent messages to be written to alert.log repeatedly manytimes.

--在某些情况下，可能会hang住很长时间，并且没有方法阻止数据库往alert log里写入这些信息。从Oracle 11gR1开始，该消息将写入tracefile，而不是alert log。从而避免重复的往alert log里写信息。