Oracle systemstate dump介绍

当数据库出现严重的性能问题或者hang起的时候，那么我们非常需要通过systemstate dump来知道进程在做什么，在等待什么，谁是资源的持有者，谁阻塞了别人。在出现上述问题时，及时收集systemstate dump非常有助于问题原因的分析。一般Oracle Support工程是也是需要你提供systemstate dump生成的trace文件做分析，关于systemstate dump的资料，其实没有非常详细的官方介绍资料，都是一些零零散散的介绍。

当数据库出现严重性能问题或hang起的时候，服务器端sqlplus连接数据库要么非常慢，要么根本无法连接。ORACLE 10g 开始，sqlplus提供了这么一个功能参数-prelim,在sqlplus无法连接的情况下，连接登录到数据库。下面关于这些知识点的一个总结

There are two ways to connect to sqlplus using a preliminary connection.

sqlplus -prelim / as sysdba

sqlplus /nolog

set _prelim on

connect / as sysdba

用sysdba登录到数据库上：

$sqlplus / as sysdba

或者

$sqlplus -prelim / as sysdba <==当数据库已经很慢或者hang到无法连接

下面是在metalink上的介绍如何在单机或RAC环境下做Systemstate或Hanganalyze（详细信息，请见下面参考资料）

Collection commands for Hanganalyze and Systemstate: Non-RAC:
Sometimes, database may actually just be very slow and not actually hanging. It is therefore recommended, where possible to get 2 hanganalyze and 2 systemstate dumps in order to determine whether processes are moving at all or whether they are "frozen".

Hanganalyze
sqlplus '/ as sysdba'
oradebug setmypid
oradebug unlimit
oradebug hanganalyze 3
-- Wait one minute before getting the second hanganalyze
oradebug hanganalyze 3
oradebug tracefile_name
exit

Systemstate
sqlplus '/ as sysdba'
oradebug setmypid
oradebug unlimit
oradebug dump systemstate 266
oradebug dump systemstate 266
oradebug tracefile_name
exit

Collection commands for Hanganalyze and Systemstate: RAC
There are 2 bugs affecting RAC that without the relevant patches being applied on your system, make using level 266 or 267 very costly. Therefore without these fixes in place it highly unadvisable to use these level

For information on these patches see:
Document 11800959.8 Bug 11800959 - A SYSTEMSTATE dump with level >= 10 in RAC dumps huge BUSY GLOBAL CACHE ELEMENTS - can hang/crash instances
Document 11827088.8 Bug 11827088 - Latch 'gc element' contention, LMHB terminates the instance

Note: both bugs are fixed in 11.2.0.3.

Collection commands for Hanganalyze and Systemstate: RAC with fixes for bug 11800959 and bug 11827088
For 11g:
sqlplus '/ as sysdba'
oradebug setorapname reco
oradebug unlimit
oradebug -g all hanganalyze 3
oradebug -g all hanganalyze 3
oradebug -g all dump systemstate 266
oradebug -g all dump systemstate 266
exit
Collection commands for Hanganalyze and Systemstate: RAC without fixes for Bug 11800959 and Bug 11827088
sqlplus '/ as sysdba'
oradebug setorapname reco
oradebug unlimit
oradebug -g all hanganalyze 3
oradebug -g all hanganalyze 3
oradebug -g all dump systemstate 258
oradebug -g all dump systemstate 258
exit
For 10g, run oradebug setmypid instead of oradebug setorapname reco:
sqlplus '/ as sysdba'
oradebug setmypid
oradebug unlimit
oradebug -g all hanganalyze 3
oradebug -g all hanganalyze 3
oradebug -g all dump systemstate 258
oradebug -g all dump systemstate 258
exit
In RAC environment, a dump will be created for all RAC instances in the DIAG trace file for each instance.

那么我们现在来看一个例子吧：

[oracle@DB-Server ~]$ sqlplus -prelim / as sysdba

SQL*Plus: Release 10.2.0.5.0 - Production on Wed Mar 2 16:31:03 2016

Copyright (c) 1982, 2010, Oracle.  All Rights Reserved.

SQL> oradebug setmypid

Statement processed.

SQL> oradebug unlimit

Statement processed.

SQL> oradebug dump systemstate 266

Statement processed.

SQL> oradebug dump systemstate 266

Statement processed.

SQL> oradebug tracefile_name

/u01/app/oracle/admin/SCM2/udump/scm2_ora_13598.trc

SQL> exit

Disconnected from ORACLE

告警日志里面会看到类似这样的信息：

Wed Mar 02 16:32:08 CST 2016

System State dumped to trace file

Wed Mar 02 16:32:48 CST 2016

System State dumped to trace file /u01/app/oracle/admin/xxx/udump/scm2_ora_13598.trc

$ORACLE_BASE/admin/ORACLE_SID/udump/ 下找到对应的trc文件，如下所示，你会看到大量系统中所有进程的进程状态等信息。每个进程对应跟踪文件中的一段内容，反映该进程的状态信息，包括进程信息，会话信息，enqueues信息（主要是lock的信息）等等。

systemstate dump有多个级别：

2: dump (不包括lock element)

10: dump

11: dump + global cache of RAC

256: short stack （函数堆栈）

258: 256+2 -->short stack +dump(不包括lock element)

266: 256+10 -->short stack+ dump

267: 256+11 -->short stack+ dump + global cache of RAC

level 11和 267会 dump global cache, 会生成较大的trace 文件，一般情况下不推荐。一般情况下，如果进程不是太多，推荐用266，因为这样可以dump出来进程的函数堆栈，可以用来分析进程在执行什么操作。但是生成short stack比较耗时，如果进程非常多，比如2000个进程，那么可能耗时30分钟以上。这种情况下，可以生成level 10 或者 level 258， level 258 比 level 10会多收集short short stack, 但比level 10少收集一些lock element data.

使用systemstate dump生成的trace文件可能会非常大，一般都会几百兆甚至更大，虽然通过system state dump收集了进程的相关，但是如何有效的解读相关信息，并诊断问题是一个不小的难题和挑战！