Hi Chris,

We are running Java 5 on CentOS 5. We have never seen any thrashing or
resource overload on the server through our monitoring system. Good to know
that this time shift issue is repeatable by triggering GC. I will build a
testing environment with Java 6 and see how it goes. Thanks for your tips. 

Regards,

Tyler 

-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Christopher
Ekberg
Sent: Friday, 9 January 2009 6:26 AM
To: Sequoia general mailing list
Subject: Re: [Sequoia] Controller Timeout Issue?

Ha, we struggled with that too.  Still are.

1.  Is this on Linux?
2.  Are you using Java 5?
3.  Do you notice that your system load shoots through the roof?  (We  
saw system load numbers of like 300 on our systems, which became  
unresponsive for over a minute)

We suspect this is a GC bug in the VM; we mostly eliminated this  
problem by commenting out an explicit call to System.gc() in Sequoia's  
code (can't remember quite where right now).  We still see the problem  
on rare occasions under load; our backends get trashed and we have to  
restore the databases from another copy.  We were also able to trigger  
this "load bomb" by hitting the "Perform GC" button in jconsole a few  
times in rapid succession; it's as if this VM gets very unhappy if  
something triggers a GC while it's already doing GC.

You can also see those warnings if your system clock changed, but I'm  
hoping you are having the same problem we are.  We're hoping that this  
is a GC bug in Java 5 and are considering switching to Java 6, which  
apparently rewrote a lot of the GC stuff.  Has anyone else had this  
problem and worked around it?

-Chris


On Jan 7, 2009, at 10:23 PM, Tyler Shen wrote:

> G'day,
>
> We have been running Sequoia in our production environment for weeks
> until this morning our application crashed with the following
> messages:
> 2009-01-07 10:44:16,490 INFO
> [org.karora.moomba.ui.internal.FolderLayout] - Adding view to
> layout: com.xbio.nba.abdr.site.ui.view.memberdetailsv...@224d0286
> 25s time shift detected, from Wed Jan 07 10:44:29 EST 2009 to Wed
> Jan 07 10:44:56 EST 2009
> Brutal changes of date/time can lead to erroneous controller failure
> detections!
> 2009-01-07 10:44:59,141 INFO
> [org.karora.moomba.ui.internal.FolderLayout] - Adding view to
> layout:
> com.xbio.nba.abdr.checklist.ui.view.tasktypesetupviewp...@6729246b
> 2009-01-07 10:44:59,148 INFO
> [org.karora.moomba.ui.internal.PageLayout] - Adding view to layout:
> View(com.xbio.nba.abdr.site.ui.view.MemberNavigator)
> 2009-01-07 10:44:59,172 WARN
> [org.hibernate.util.JDBCExceptionReporter] - SQL Error: 0, SQLState:
> null
> 2009-01-07 10:44:59,172 ERROR
> [org.hibernate.util.JDBCExceptionReporter] - Protocol corruption
> while trying to send command: 0. Check the previous command
> 18s time shift detected, from Wed Jan 07 10:44:40 EST 2009 to Wed
> Jan 07 10:45:00 EST 2009
> Brutal changes of date/time can lead to erroneous controller failure
> detections!
> 29s time shift detected, from Wed Jan 07 10:44:29 EST 2009 to Wed
> Jan 07 10:45:00 EST 2009
> Brutal changes of date/time can lead to erroneous controller failure
> detections!
> 2009-01-07 10:45:32,600 INFO
> [com
> .xbio
> .nba
> .abdr.patienttransition.service.impl.PatientTransitionServiceImpl] -
> completeOutstandingTransitions executed
> 2009-01-07 10:45:32,713 WARN
> [org.hibernate.util.JDBCExceptionReporter] - SQL Error: 0, SQLState:
> null
> 2009-01-07 10:45:32,713 ERROR
> [org.hibernate.util.JDBCExceptionReporter] - Protocol corruption
> while trying to send command: 0. Check the previous command
>
> We have the server running on NTP. How such a time shift occurred is
> beyond me. Having briefly gone through the source code, I noticed
> that tweaking ControllerTimeout can make the Sequoia more tolerant
> once a time shit is detected.
> Where can I change ControllerTimeout and is this the proper way to
> go around the issues? Thanks!
>
> Regards,
>
> Tyler Shen
>
>
> <ATT00001.txt>


_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia

_______________________________________________
Sequoia mailing list
[email protected]
https://forge.continuent.org/mailman/listinfo/sequoia

Reply via email to