[Wrapper-user] RE : don't have time to get heap dump

[DJP JEAN-P. D. <Dom...@de...>]

Just to say that by desactivating the ping feature, I managed to get my dump.
dom


-------- Message d'origine--------
De:	DJP JEAN-PROST Dominique
Date:	lun. 09/02/2009 12:01
À:	wra...@li...
Cc:	Leif Mortenson
Objet:	Re: [Wrapper-user] don't have time to get heap dump

Glad to see we understood each other.
I will try to desactivate the ping timeout, as getting this dump is getting urgent.

regards,
dom

-----Message d'origine-----
De : Leif Mortenson [mailto:le...@ta...]
Envoyé : lundi 9 février 2009 11:34
À : DJP JEAN-PROST Dominique
Cc : wra...@li...
Objet : Re: [Wrapper-user] don't have time to get heap dump


Dom,
No problem.  It is always good to understand what is happening.

The ping timeout is used while the JVM application is running
normally.   How are you initiating your memory dump?   I was thinking
it was the memory dump that takes place as the JVM is exiting.    If
the JVM continues after the dump is complete then you are correct, you
would want to do something like extend the ping timeout.
http://wrapper.tanukisoftware.org/doc/english/prop-ping-timeout.html

Cheers,
Leif

2009/2/9 DJP JEAN-PROST Dominique <Dom...@de...>:
> Leif,
>
> I will try what you suggest. But as I always try to understand what I do, let me try to reexpose my question.
> You explain me the shutdown process of the jvm (hooks, ...). I did use this parameter in another environment because when I stopped the wndows service, jvm shutdown was quite long.
>
> Now for my OOME : I setup jvm so that if it encounters an memory problem, it creates the memory dump. This setup has nothing to do whith shutdown or something related for what I understand. So what I understand is that when memory problem occurs, jvm tries to make the dump. During this time, jvm does not respond anymore, so wrapper decide to kill the process, and would normally start another one, but I configured it not to because I thought it would let time to the jvm to make the dump : it doesn't.
>
> I'm not trying to tell your're wrong, but I don't realy understand the motivation of using jvm_exit parameter, and not for instance the ping timeout parameter.
>
> Can you elaborate about this please ?
> best regards,
> Dom
>
> -----Message d'origine-----
> De : Leif Mortenson [mailto:le...@ta...]
> Envoyé : samedi 7 février 2009 15:09
> À : wra...@li...
> Objet : Re: [Wrapper-user] don't have time to get heap dump
>
>
> Dom,
> Please try this property:
> wrapper.jvm_exit.timeout=600
>
> The WrapperManager's shutdown hook makes a call to the native Wrapper
> process that the Java application is stopped.  From that point, there
> is a default 15 second timeout that the Wrapper will wait for the JVM
> process to terminate.  If it fails to do so then the Wrapper will
> forcibly kill it.  This is what you are seeing.
> The above property extends this timeout.
>
> When the JVM completes normal operation, it will start doing the full
> heap dump.   This can take several minutes for large applications.
> During this time, the JVM process still exists, but is in a state
> which appears to be frozen.
>
> Cheers,
> Leif
>
> 2009/2/7 DJP JEAN-PROST Dominique <Dom...@de...>:
>> Well, when the dump is getting done, the jvm is not trying to exit, so I don't think what you suggest could help me, or I didn't really understand.
>>
>> For what I understand, jvm gets the oome, it then tries to dump memory on disk, but wrapper does something (it kills the process ?) before the dump gets done.
>>
>> Where am I worng ?
>> dom
>>
>> -----Message d'origine-----
>> De : Leif Mortenson [mailto:le...@ta...]
>> Envoyé : vendredi 6 février 2009 17:56
>> À : wra...@li...
>> Objet : Re: [Wrapper-user] don't have time to get heap dump
>>
>>
>> Dom,
>> The problem here is that the Wrapper itself does not know that the JVM
>> is doing a heap dump on exit.  It tells the Wrapper that it is
>> shutting down but then the java process does not terminate.  The
>> Wrapper by default waits for 15 seconds before deciding that the java
>> process is frozen and kills it.  This is what you are seeing.
>>
>> Normally this is the desired action.  In the case of large heap dumps
>> however, the timeout needs to be extended.   Try adding the following
>> to your wrapper.conf:
>>
>> wrapper.jvm_exit.timeout=600
>>
>> That will tell the Wrapper to wait up to 10 minutes for the jvm
>> process to exit.  Another useful timeout may be:
>>
>> wrapper.shutdown.timeout=600
>>
>> They are both described here:
>> http://wrapper.tanukisoftware.org/doc/english/prop-jvm-exit-timeout.html
>> http://wrapper.tanukisoftware.org/doc/english/prop-shutdown-timeout.html
>>
>> Let me know how this works for you.
>> Cheers,
>> Leif
>>
>> 2009/2/7 DJP JEAN-PROST Dominique <Dom...@de...>:
>>> Hello,
>>>
>>> I'm actually meeting Out of Memory in my production environment. I configured jvm so that I get a heap dump when the oome is raised.
>>> So the I met out of memory, but wrapper did not let jvm the time to dump the heap :
>>>
>>> INFO   | jvm 1    | 2009/02/06 15:56:33 | java.lang.OutOfMemoryError: Java heap space
>>> INFO   | jvm 1    | 2009/02/06 15:56:33 | Dumping heap to c:\dumps\java_pid2308.hprof ...
>>> ERROR  | wrapper  | 2009/02/06 15:56:48 | JVM appears hung: Timed out waiting for signal from JVM.
>>> ERROR  | wrapper  | 2009/02/06 15:56:48 | JVM did not exit on request, terminated
>>> STATUS | wrapper  | 2009/02/06 15:56:49 | JVM Restarts disabled.  Shutting down.
>>>
>>> I thought wrapper.disable_restarts=TRUE would let time the jvm to do the dump, but it didn't.
>>> What parameter should I use so that I get the full dump ?
>>> What would be really cool is that I got the heap dump and that wrapper automatically restart the jvm *after*.
>>>
>>> Can someone help me ?
>>> Best regards,
>>> dom


Consultez nos nouveaux sites internet : 
http://www.dexia-sofaxis.com 
http://www.dexia-sofcap-sofcah.com

Tous ensemble pour l’environnement : n’imprimer ce courriel que si nécessaire.

Dexia Sofaxis disclaimer : http://www.dexia-sofaxis.com/disclaimer.html