NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

问题描述:

上午刚刚到办公室,就有监控人员邮件反馈,昨晚NDMCDB407数据库被重启过,让我分析一下数据库重启的原因。由于昨晚业务有版本上线,所以短信警告关闭了,所以没有短信下发到我手机上,而且故障时相关人员也没有通知到我。

1     检查alert日志

从alert日志中,可以看到,先是在03:29时有一个job运行失败了:
Fri Aug 22 03:29:29 2014
Errors in file/opt/oracle/diag/rdbms/ndmcdb/NDMCDB/trace/NDMCDB_j000_28856.trc:
ORA-12012: error on auto execute of job 31
ORA-04023: ObjectNDMC.DELETE_ANONY_RSHARE_INFO could not be validated or authorized
ORA-06512: at "NDMC.PROC_NDMC_CANCEL_OPEN",line 5
ORA-06512: at line 1
然后在03:49时,出现了连接超时失败,而且一直持续到05:00:08:
Fri Aug 22 03:49:43 2014
***********************************************************************
 
Fatal NI connect error 12170.
 
 VERSION INFORMATION:
       TNS for Linux: Version 11.1.0.7.0 - Production
       Oracle Bequeath NT Protocol Adapter for Linux: Version 11.1.0.7.0 -Production
       TCP/IP NT Protocol Adapter for Linux: Version 11.1.0.7.0 - Production
 Time: 22-AUG-2014 03:49:43
 Tracing not turned on.
  Tnserror struct:
   ns main err code: 12535
   
TNS-12535: TNS:operation timed out
   ns secondary err code: 12606
   nt main err code: 0
   nt secondary err code: 0
   nt OS err code: 0
 Client address: (ADDRESS=(PROTOCOL=tcp)(HOST=192.168.130.87)(PORT=36628))
WARNING: inbound connection timed out(ORA-3136)
Fri Aug 22 03:49:44 2014
……
而且出现了连接数耗尽了:
Fri Aug 22 03:49:50 2014
ORA-00020: maximum number of processes 0exceeded
   ns secondary err code: 12560
   ns secondary err code: 12560
   ns main err code: 12537
Fri Aug 22 03:49:50 2014
……
Fri Aug 22 03:51:48 2014
 
***********************************************************************
 
Fatal NI connect error 12537, connectingto:
 (LOCAL=NO)
 
 VERSION INFORMATION:
       TNS for Linux: Version 11.1.0.7.0 - Production
       Oracle Bequeath NT Protocol Adapter for Linux: Version 11.1.0.7.0 -Production
       TCP/IP NT Protocol Adapter for Linux: Version 11.1.0.7.0 - Production
 Time: 22-AUG-2014 03:51:48
 Tracing not turned on.
  Tnserror struct:
   ns main err code: 12537
   
TNS-12537: TNS:connection closed
ns secondaryerr code: 12560
   nt main err code: 0
   nt secondary err code: 0
   nt OS err code: 0
ORA-609 : opiodr aborting process unknownospid (30476_47044991385184)
Fri Aug 22 04:14:15 2014
ORA-28 : opiodr aborting process unknownospid (24925_46986315964000)
Fri Aug 22 04:16:27 2014
ORA-28 : opiodr aborting process unknownospid (22475_47013891882592)
Fri Aug 22 04:16:28 2014
ORA-28 : opiodr aborting process unknownospid (21356_47116835528288)
Fri Aug 22 04:16:29 2014
ORA-28 : opiodr aborting process unknownospid (24947_47774766210656)
ORA-28 : opiodr aborting process unknownospid (14958_47053435166304)
……
Fri Aug 22 05:00:05 2014
ORA-28 : opiodr aborting process unknownospid (25765_46941307182688)
Fri Aug 22 05:00:08 2014
ORA-28 : opiodr aborting process unknownospid (4949_47396524895840)
于是在05:04数据库被关闭,从日志来看,这是正常关闭的,初步怀疑是人为关闭或是VCS双机自动将数据库关闭了:
Fri Aug 22 05:04:10 2014
Stopping background process SMCO
Stopping background process FBDA
Shutting down instance: further logonsdisabled
Fri Aug 22 05:04:12 2014
Stopping background process CJQ0
Stopping background process QMNC
Stopping background process MMNL
Stopping background process MMON
Shutting down instance (immediate)
License high water mark = 1220
Stopping Job queue slave processes, flags =7
Fri Aug 22 05:04:20 2014
Waiting for Job queue slaves to complete
Job queue slave processes stopped
Fri Aug 22 05:09:11 2014
License high water mark = 1220
USER (ospid: 25110): terminating theinstance
Termination issued to instance processes.Waiting for the processes to exit
Fri Aug 22 05:09:21 2014
Instance termination failed to kill one ormore processes
Instance terminated by USER, pid = 25110

2     检查messages日志

大概在05:03:51时,人为的想将双机切换到备机中:

Aug 22 05:03:51 NDMCDB11 user_cmd:2014-08-22 05:03:51 hagrp -switch RCS_DB_SG -to system by root from [oraclepts/9 Aug 22 04:29 (192.168.128.142)]
Aug 22 05:04:01 NDMCDB11/usr/sbin/cron[15348]: (root) CMD (su - root -c'/opt/watchdog/watchdog_schedule -n OS,oracle' >/dev/null 2>&1)
Aug 22 05:04:01 NDMCDB11 su: (to root) rooton none
Aug 22 05:04:03 NDMCDB11 su: (to oracle)root on none
Aug 22 05:04:09 NDMCDB11 user_cmd:2014-08-22 05:04:09 hagrp -switch RCS_DB_SG -to NDMCDB12 by root from [oraclepts/9 Aug 22 04:29 (192.168.128.142)]
Aug 22 05:04:09 NDMCDB11 su: (to oracle)root on none

但双机切换失败,最后是直接将双机停止,重启VCS:

Aug 22 05:06:18 NDMCDB11 user_cmd:2014-08-22 05:06:18 hastop -all by root from [oracle pts/9 Aug 22 04:29(192.168.128.142)]
……
Aug 22 05:07:02 NDMCDB11 user_cmd:2014-08-22 05:07:02 hastat by root from [oracle pts/9 Aug 22 04:29(192.168.128.142)]

所以,到这里就已经确定,数据库这所以重启了,完全是由于人为将VCS集群重启引起的。那么为什么要VCS群集重启呢?数据库到底有没有问题呢?再来看看。

 

最后,经向升级人员操作确认,在升级时,有一个存储过程需要跑,但执行后,数据库基本响应就非常慢了,一直运行到3:29左右,人为cancel掉了,所以这也就是为什么会出现这样的报错了:

Fri Aug 22 03:29:29 2014
Errors in file/opt/oracle/diag/rdbms/ndmcdb/NDMCDB/trace/NDMCDB_j000_28856.trc:
ORA-12012: error on auto execute of job 31
ORA-04023: ObjectNDMC.DELETE_ANONY_RSHARE_INFO could not be validated or authorized
ORA-06512: at"NDMC.PROC_NDMC_CANCEL_OPEN", line 5
ORA-06512: at line 1

3     查看系统负载

CPU负载:

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

内存负载:

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

可见,系统在3:49左右,出现了CPU及内存均被耗尽的情况,这个时间段,刚好数据库出现了大量连接超时失败,甚至是出现了连接数超过阀值:

Fri Aug 22 03:49:50 2014
ORA-00020: maximum number of processes 0exceeded
   ns secondary err code: 12560
   ns secondary err code: 12560
   ns main err code: 12537
Fri Aug 22 03:49:50 2014

4     分析AWR

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

从这里看,数据库在2点到3点时,已经非常的繁忙,但从之前有系统负载来看,2点到3点时,CPU及内存使用率都不算很高的。接着看:

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

指标都没有什么特别高的。

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

从top 5 event中,看到了有大量的cursor: pin S wait on X等待,可见出现mutex争用,但通常这只是表象而已,并非根因。

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

绝大部分时间都在做SQL的解析,而且解析还失败了,这就是数据库hang住的根因。正常来说,一个数据库的绝大部分时间应该是用于SQL的执行,所以这个是占用最多时间的:sql execute elapsedtime等。

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

不存在较高的versioncount。

 

那么数据库什么时候出现的不停解析SQL,并且解析失败了呢?

查了DBA_HIST_ACTIVE_SESS_HISTORY,分析了下历史会话信息,发现在02:57:00至03:00:00出现的问题:

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

经过确认,恰巧就是执行存储过程的时间点左右。

至此,数据库从3:00开始,已经是不正常的,数据库不停的在解析SQL,SQL都还没有到执行这一步,数据库已经处于无响应的状态,连接会话都被阻塞住了,直到连接数达到了最大连接数,最后被升级操作人员重启了VCS集群。

 

5     分析结论

(1)数据库down机主要还是人为进行了VCS切换失败后,进行了VCS重启操作引起。

(2)这套数据库故障的根因,还是为什么数据库在2:58左右时出现解析SQL失败上。从目前的日志分析来看,看不出是什么原因。

 

NDMCDB数据库hang住故障分析 - cursor: pin S wait on X

上一篇:PLSQL developer常用技巧


下一篇:MySQL在大型网站的应用架构演变