Comments on: Library cache latches gone in Oracle 11g http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/ Oracle troubleshooting, internals and performance tuning Mon, 15 Mar 2010 01:50:11 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Tanel Poder http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-3992 Tanel Poder Sun, 07 Mar 2010 03:42:27 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-3992 <a href="#comment-3976" rel="nofollow">@Richard</a> Hi Richard, Yes, latchprofx does work on 11g as well. In 11gR2 there was a new CBO transformation enabled which made older latchprofx versions not work and return only one row, but I have fixed this issue and uploaded a new version (v1.20) which works on 11gR2 too. 11gR1 doesn't have this problem. So download the latest version and make sure it is v1.20 (or higher) @Richard
Hi Richard,

Yes, latchprofx does work on 11g as well. In 11gR2 there was a new CBO transformation enabled which made older latchprofx versions not work and return only one row, but I have fixed this issue and uploaded a new version (v1.20) which works on 11gR2 too. 11gR1 doesn’t have this problem. So download the latest version and make sure it is v1.20 (or higher)

]]> By: Tanel Poder http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-3991 Tanel Poder Sun, 07 Mar 2010 03:40:46 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-3991 <a href="#comment-3529" rel="nofollow">@David Rydzewski</a> David, this is a late reply, but I don't think Mutexes have level sanity checking like latches do. Latches don't really need it either for functioning, but this is more for debugging and sanity checks/assertions in case something goes wrong. @David Rydzewski
David, this is a late reply, but I don’t think Mutexes have level sanity checking like latches do. Latches don’t really need it either for functioning, but this is more for debugging and sanity checks/assertions in case something goes wrong.

]]> By: Richard http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-3976 Richard Thu, 04 Mar 2010 15:24:53 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-3976 Hi Tanel - Does LatchProfX work with 11g? Hi Tanel – Does LatchProfX work with 11g?

]]> By: David Rydzewski http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-3529 David Rydzewski Mon, 04 Jan 2010 16:30:47 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-3529 Hi Tandel - I don't have the specifics in front of me, but I remember reading Steve Adams' book in regards to latching and it mentioned deadlock prevention in latches based on the latch level. Basically, latches need to be taken in certain order and if violated, Oracle throws an internal error. Do you know if the mutex mechanism also does this checking or is it assumed that the more granular level of the mutex (and careful coding) does not require this checking? Thanks, Dave Hi Tandel –

I don’t have the specifics in front of me, but I remember reading Steve Adams’ book in regards to latching and it mentioned deadlock prevention in latches based on the latch level. Basically, latches need to be taken in certain order and if violated, Oracle throws an internal error.

Do you know if the mutex mechanism also does this checking or is it assumed that the more granular level of the mutex (and careful coding) does not require this checking?

Thanks,
Dave

]]> By: Tanel Poder http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1737 Tanel Poder Fri, 05 Jun 2009 21:03:59 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1737 Hi DBA, It sure sounds like a bug. There was an issue with mutexes that when a get failed the mutex getter never slept for long time. It spinned for short time and then just yielded CPU (which is not a sleep - but just means the process goes off CPU, but not to sleep but to the end of the runqueue - trying to get back onto CPU immediately) That would explain why you had 99% CPU usage which didn't "go away". Even if you shut down the client and the sessions remain (processes aren't killed) then the sessions may be stuck in this mutex get loop and never check whether they should exit. In 11.1.0.7 the yielding problem should have been fixed, but maybe not for all cases. If it happens again, I would try to dump x$mutex_sleep_history and take hanganalyze dumps. If your db is completely hung, then you can connect with "sqlplus -prelim" to take a hanganalyze dump without estabilishing a session in the instance. Hi DBA,

It sure sounds like a bug. There was an issue with mutexes that when a get failed the mutex getter never slept for long time. It spinned for short time and then just yielded CPU (which is not a sleep – but just means the process goes off CPU, but not to sleep but to the end of the runqueue – trying to get back onto CPU immediately)

That would explain why you had 99% CPU usage which didn’t “go away”. Even if you shut down the client and the sessions remain (processes aren’t killed) then the sessions may be stuck in this mutex get loop and never check whether they should exit.

In 11.1.0.7 the yielding problem should have been fixed, but maybe not for all cases.

If it happens again, I would try to dump x$mutex_sleep_history and take hanganalyze dumps. If your db is completely hung, then you can connect with “sqlplus -prelim” to take a hanganalyze dump without estabilishing a session in the instance.

]]> By: DBA http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1736 DBA Fri, 05 Jun 2009 19:42:40 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1736 Hi Tanel, I posted a "library cache: mutex X" problem above. Have you come across that before? Thanks. Hi Tanel,

I posted a “library cache: mutex X” problem above. Have you come across that before?

Thanks.

]]> By: Очередная подборка ссылок « Тетрадочка DBA http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1611 Очередная подборка ссылок « Тетрадочка DBA Fri, 22 May 2009 11:45:30 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1611 [...] – HOW TO SUPPORT TWO-TASK COMMON ERRORS - С блога Танела Подера: Library cache latches gone in Oracle 11g - Note: 177610.1 – Oracle Forms in Applications FAQ - Раздел на oracle.com – [...] [...] – HOW TO SUPPORT TWO-TASK COMMON ERRORS – С блога Танела Подера: Library cache latches gone in Oracle 11g – Note: 177610.1 – Oracle Forms in Applications FAQ – Раздел на oracle.com – [...]

]]> By: DBA http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1601 DBA Tue, 19 May 2009 21:14:58 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1601 Hi Tanel, We are using 11.1.0.7 on Linux_x86_64. We are having a problem with "library cache: mutex X" wait. The CPU usage goes high to 99% and stays there until I shutdown the database (ABORT!). The client is shutdown but the sessions are active in the database and are waiting on that mutex wait event. Some people are saying there is a bug in 11g that was supposed to be fixed in 11.1.0.7. http://jhdba.wordpress.com/2008/03/31/11g-library-cache-mutex-x-known-bug/ Have you come across such behavior with 11.1.0.7? Hi Tanel,

We are using 11.1.0.7 on Linux_x86_64. We are having a problem with “library cache: mutex X” wait. The CPU usage goes high to 99% and stays there until I shutdown the database (ABORT!). The client is shutdown but the sessions are active in the database and are waiting on that mutex wait event.
Some people are saying there is a bug in 11g that was supposed to be fixed in 11.1.0.7. http://jhdba.wordpress.com/2008/03/31/11g-library-cache-mutex-x-known-bug/
Have you come across such behavior with 11.1.0.7?

]]> By: Tanel Poder http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1430 Tanel Poder Thu, 23 Apr 2009 05:18:00 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1430 Yep I suspected that as another reason. Thanks for letting me know. Yep I suspected that as another reason. Thanks for letting me know.

]]> By: Dion Cho http://blog.tanelpoder.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/comment-page-1/#comment-1429 Dion Cho Thu, 23 Apr 2009 05:10:08 +0000 http://tanelpoder.wordpress.com/2008/08/03/library-cache-latches-gone-in-oracle-11g/#comment-1429 <i>(I had to sample v$mutex_sleep_history instead of v$mutex_sleep on 11g as the latter was empty on 11g… it may be that the mutex get operations have been tuned further to not maintain the counters just to save even few more CPU cycles every get) </i> It's a bug. Metalink bug# 7202166 (I had to sample v$mutex_sleep_history instead of v$mutex_sleep on 11g as the latter was empty on 11g… it may be that the mutex get operations have been tuned further to not maintain the counters just to save even few more CPU cycles every get)

It’s a bug. Metalink bug# 7202166

]]>