Hi Achim
1. wrapper.on_exit.default=RESTART is a good start. However, it is
better if we can control when exactly we want karaf to restart by
providing configure exit code? ( ie <karaf> shutdown 999 )
Otherwise, we may run into infinite restart
2. Due to deployment logistic and complexity, we have decided to use
single kar deployment.
Thanks
-D
On Fri, Dec 16, 2011 at 12:37 AM, Achim Nierbeck
<[email protected]> wrote:
> Hi Dan,
>
> 1) well the default one is for std. Shutdown, this way you tell the service
> wrapper that if the system is going down for a shutdown no matter what the
> exit code says it does a reboot.
>
> 2) regarding your re-deployment. Consider this, every time a bundle is
> updated all previous Class definitions
> are ignored but since all Class definitions and Static members are stored in
> the PermGen space this memory
> is increased everytime you do a redeploy. This is allright if you just
> update a bundle or two but if you update
> lot's of bundles with lot's of class-definitions you run out of Memory
> here.
> There is one JVM that doesn't have a permGen space it's the JRockit, though
> I don't know if it is still available
> for free as it used to be :)
> One other side note, you might consider updating only the bundle in question
> instead of updating the complete set
> of bundles in a feature when going productive :)
>
> Regards, Achim
>
>
> 2011/12/16 Dan Tran <[email protected]>
>>
>> Hi Achim
>>
>> 1. I am not really understand what wrapper.on_exit.default=RESTART
>> does, but with wrapper.on_exit.99=RESTART, and If can get my bundle to
>> shutdown karaf with exit code 99. That would be test.
>>
>> 2. Our usage of karaf is having a feature ( with a bunch of bundles
>> it has many classes 200+ and growing ). We remotely access karaf via
>> JMX to update our feature feature by having karaf to contact maven
>> repo to pickup the update.
>> How often do we do the update? quite often during development.
>> But not expect a lots once it get to customer size. This is how we
>> deliver bug fix and added feature.
>>
>> Thanks
>>
>> -Dan
>>
>>
>>
>> On Thu, Dec 15, 2011 at 2:37 PM, Achim Nierbeck <[email protected]>
>> wrote:
>> > Hi Dan,
>> >
>> > well the first entry on your link allready does what you need :)
>> > Instead of
>> >
>> > wrapper.on_exit.default=SHUTDOWN
>> >
>> >
>> > do a
>> >
>> > wrapper.on_exit.default=RESTART
>> >
>> >
>> > this way your karaf will always restart through the servicewrapper.
>> >
>> > No back to your PermGen leak issue.
>> > Could you give me a hint of what you actually are doing.
>> > If I did understand correctly you use the features service to deploy a
>> > bunch
>> > of bundles.
>> > Now how often do you actually re-deploy this feature set?
>> > How many classes are contained in your bundles?
>> >
>> > regards, Achim
>> >
>> >
>> >
>> > Am 15.12.2011 21:23, schrieb Dan Tran:
>> >
>> >> We should be able to configure wrapper to restart karaf if karaf exits
>> >> with an expected code. see
>> >> http://wrapper.tanukisoftware.com/doc/english/prop-on-exit-n.html
>> >>
>> >> The question here is how can we do that a karaf's bundle? Can we call
>> >> shutdown with an exit code?
>> >>
>> >> Thanks
>> >>
>> >> -Dan
>> >>
>> >>
>> >> On Thu, Dec 15, 2011 at 9:21 AM, Jean-Baptiste Onofré<[email protected]>
>> >> wrote:
>> >>>
>> >>> Thanks for the update Karaf.
>> >>>
>> >>> For the wrapper, you can use respawn (on Unix).
>> >>> For the perm, it looks like a framework "issue" (we have to dig
>> >>> around).
>> >>>
>> >>> Regards
>> >>> JB
>> >>>
>> >>>
>> >>> On 12/15/2011 06:18 PM, Dan Tran wrote:
>> >>>>
>> >>>> Here are notes after playing with dev:restart:
>> >>>>
>> >>>> 1. Permgen still leaks
>> >>>>
>> >>>> 2. Works only in main karaf shell ( ie invoke bin/karaf ). When I
>> >>>> ran it as a service via wrapper and use bin/client to assess karaf
>> >>>> shell, dev:restart kill my karaf.
>> >>>>
>> >>>> So it seems we need a cold restart.
>> >>>>
>> >>>> -D
>> >>>>
>> >>>> On Thu, Dec 15, 2011 at 9:02 AM, Dan Tran<[email protected]>
>> >>>> wrote:
>> >>>>>
>> >>>>> I am using 2.2.x, can I invoke dev:restart in my own bundle? any
>> >>>>> side
>> >>>>> affect on this since my bundle also get restart.
>> >>>>>
>> >>>>> Thanks
>> >>>>>
>> >>>>> -D
>> >>>>>
>> >>>>> On Thu, Dec 15, 2011 at 8:48 AM, Jean-Baptiste
>> >>>>> Onofré<[email protected]>
>> >>>>> wrote:
>> >>>>>>
>> >>>>>> It's already possible:
>> >>>>>>
>> >>>>>> - in the console: dev:restart for 2.2.x version, system:restart for
>> >>>>>> trunk
>> >>>>>> (3.0)
>> >>>>>> - for the MBean, it's in org.apache.karaf:type=dev in Karaf 2.2.x
>> >>>>>> (restart()
>> >>>>>> operation), or org.apache.karaf:type=system in Karaf 3.0.0 (trunk,
>> >>>>>> restart()
>> >>>>>> operation)
>> >>>>>>
>> >>>>>> Regards
>> >>>>>> JB
>> >>>>>>
>> >>>>>>
>> >>>>>> On 12/15/2011 05:39 PM, Dan Tran wrote:
>> >>>>>>>
>> >>>>>>>
>> >>>>>>> Looks like the permgen issue is here for a long time. How about
>> >>>>>>> adding ability for karaf to restart itself ( Hudson, and eclipse
>> >>>>>>> can
>> >>>>>>> do this i think ). May be via webconsole, or JMX
>> >>>>>>>
>> >>>>>>> Eager to hear how we could implement such feature for karaf.
>> >>>>>>>
>> >>>>>>> -D
>> >>>>>>>
>> >>>>>>> ---------- Forwarded message ----------
>> >>>>>>> From: Dan Tran<[email protected]>
>> >>>>>>> Date: Thu, Dec 15, 2011 at 12:23 AM
>> >>>>>>> Subject: Re: Infamous permgen leak
>> >>>>>>> To: [email protected]
>> >>>>>>>
>> >>>>>>>
>> >>>>>>> It is about feature to be loaded in and out of karaf dynamically.
>> >>>>>>> So
>> >>>>>>> I guess is a known issue to 'java', karaf is not an exception :(
>> >>>>>>>
>> >>>>>>> Wonder if ACE can solve this issue
>> >>>>>>>
>> >>>>>>>
>> >>>>>>> -D
>> >>>>>>>
>> >>>>>>> On Thu, Dec 15, 2011 at 12:15 AM, Jean-Baptiste
>> >>>>>>> Onofré<[email protected]>
>> >>>>>>> wrote:
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>> Hi Dan,
>> >>>>>>>>
>> >>>>>>>> just to understand your issues:
>> >>>>>>>> - if you talk about the feature service itself, I'm agree that it
>> >>>>>>>> shouldn't
>> >>>>>>>> consume perm
>> >>>>>>>> - but the loaded classes (loaded by the feature service) consumes
>> >>>>>>>> perm,
>> >>>>>>>> and
>> >>>>>>>> it's a normal behavior (it's exactly the same with on Tomcat with
>> >>>>>>>> hot-deployment for instance)
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>> Regards
>> >>>>>>>> JB
>> >>>>>>>>
>> >>>>>>>> On 12/15/2011 08:36 AM, Dan Tran wrote:
>> >>>>>>>>>
>> >>>>>>>>>
>> >>>>>>>>>
>> >>>>>>>>> Hello Karaf Team
>> >>>>>>>>>
>> >>>>>>>>> Each time I reinstall/update a feature at runtime, my karaf
>> >>>>>>>>> loses
>> >>>>>>>>> some
>> >>>>>>>>> permgen. Amount depending on number to classes I have. Is there
>> >>>>>>>>> any
>> >>>>>>>>> one able to address this issue?
>> >>>>>>>>>
>> >>>>>>>>> Setting JVM to -XX:+CMSPermGenSweepingEnabled
>> >>>>>>>>> -XX:+CMSClassUnloadingEnabled doesnot help.
>> >>>>>>>>>
>> >>>>>>>>> -Dan
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>> --
>> >>>>>>>> Jean-Baptiste Onofré
>> >>>>>>>> [email protected]
>> >>>>>>>> http://blog.nanthrax.net
>> >>>>>>>> Talend - http://www.talend.com
>> >>>>>>
>> >>>>>>
>> >>>>>>
>> >>>>>> --
>> >>>>>> Jean-Baptiste Onofré
>> >>>>>> [email protected]
>> >>>>>> http://blog.nanthrax.net
>> >>>>>> Talend - http://www.talend.com
>> >>>
>> >>>
>> >>> --
>> >>> Jean-Baptiste Onofré
>> >>> [email protected]
>> >>> http://blog.nanthrax.net
>> >>> Talend - http://www.talend.com
>> >
>> >
>> >
>> > --
>> > --
>> > *Achim Nierbeck*
>> >
>> >
>> > Apache Karaf<http://karaf.apache.org/> Committer& PMC
>> > OPS4J Pax Web<http://wiki.ops4j.org/display/paxweb/Pax+Web/>
>> > Committer&
>> > Project Lead
>> > blog<http://notizblog.nierbeck.de/>
>> >
>
>
>
>
> --
>
> Apache Karaf <http://karaf.apache.org/> Committer & PMC
> OPS4J Pax Web <http://wiki.ops4j.org/display/paxweb/Pax+Web/> Committer &
> Project Lead
> blog <http://notizblog.nierbeck.de/>
