Hi Christian

Oh, that it were so simple!  When WebSphere crashes, it takes out the whole
server, and I don't believe it has the time to create a core dump file.  I
have tried killing it before it gets too far down this process with a kill
-3 <pid> ( as I can see the memory usage and paging quickly climb on topas),
but although the coredump file looked OK to me, neither tool recognised it
as such.

I have to say, I think this is a WebSphere issue as well as a GeoServer
issue, and am trying to document the bug as well as I can in order to raise
it with IBM.  I don't believe that a java app should be capable of bringing
down a server, and I'm sure its just a case of tuning the WAS properly.

Thanks

Rob


On 6 August 2010 08:55, <[email protected]> wrote:

> Hi, I have a similar configuration using AIX, Linux/ppc, IBM sdk, openjdk ,
> DB2 and Websphere or jetty.
>
> To make it short. If your Websphere crashes it creates an heapdump,
> composed of 2 or 3 files. You should find these files in your profile
> directory (something like /opt/IBM/Webspere/AppServer/profiles/server1).
>
> There is an IBM tool
> http://www.alphaworks.ibm.com/tech/heaproots
>
> for analyzing the dump. For large heapdumps , please dont use the GUI
> version, otherwise the utility would create an heapdump for itself :-(.
> Command Line operation is needed.
>
> Within the last 3 years, this utility helps me to solve any problem
> concerning memory problems and finding the code causing these problems.
>
> I would recommend you to invest some time, I am sure this is not the last
> heapdump you have to study. Believe me, it is the best way to find the
> problem.
>
> If you have questions, please ask, I will support you.
>
> I can also analyze your heapdump, but this requires at minimum one day and
> this is part of my commercial support for customers.
>
> Hope this helps
>
>
>
>
>
>
>
>
>
>
> Quoting Andrea Aime <[email protected]>:
>
>  Rob ha scritto:
>>
>>> I'm seeing a serious problem with my geoserver install, where it is
>>> crashing the system it is running on, following certain types of WMS
>>> requests.
>>>
>>> Given the fact that I am probably the only person on the list running
>>> GeoServer (v2.0.0) under WebSphere App Server, using IBM Java, on AIX, I
>>> am not expecting anybody to point me in the direction of a silver
>>> bullet.  However, if someone could confirm my theory as to what is
>>> happening that is causing the issue in the first place, it might help.
>>> I'll try and talk through an example, and my hypothesis - made with no
>>> knowledge of the geoserver code whatsoever, I hasten to add! :)
>>>
>>>
>>> FOR AN EXAMPLE WAYWARD REQUEST
>>>
>>> I am rendering some vector data stored in Oracle Spatial.  This data is
>>> quite detailed, and from the geoserver debug logs, I can identify the
>>> exact SQL used. I can run the SQL in sqlplus and the records return
>>> pretty much instantaneously.  There are eight polygons, and eight lines.
>>>  The lines and polygons have between 50 and 300 vertices, and are styled
>>> in different ways.  The MBR for these records is approx 2km by 1.5km.
>>>
>>> These records render beautifully, and quickly (1-2 seconds) at 1:2000.
>>> [To my simple mind, this suggests that the geometries are valid, and
>>> that the styling is OK.  I remember seeing an issue with symbols at
>>> these kind of scales causing problems, but none of these records would
>>> be styled using symbols.  Basic lines and fills only]
>>>
>>> As I zoom in (and in, and in)  the rendering starts to take more time
>>> each time, until I'm at something like 1:50.  Rendering is now taking 20
>>> seconds.  If I zoom in until I'm at 1:2 or 1:1, I can see the memory on
>>> the machine be eaten up, until WebSphere starts paging like crazy and
>>> eventually the machine hangs.
>>>
>>> My theory is that GeoServer is trying to create an image at the same
>>> scale as the target bbox, but containing the entire geometries pulled
>>> back from Oracle.  So if I request a 600x600 pixel image at 1:1 scale,
>>> it tries to build an internal image that would contain the 2km x 1.5km -
>>> i.e., a (600 x 2000=1200000) x (600 x 1500=900000) image, before it
>>> would cut out the 600x600 bbox of interest.
>>>
>>> Is this how it works?
>>>
>>> GeoServer does have some WMS resource consumption limits that I can
>>> configure, and I have these currently set to
>>>
>>> Max rendering memory - 73728 KB (up from 64Mb, as I was seeing some
>>> OutOfMemory issues on something else)
>>> Max rendering time - 20s (down from 60s - see below)
>>> Max rendering errors - 100 (down from 1000)
>>>
>>> Changing the max rendering time seems to have helped, as GeoServer cant
>>> eat all the available memory and pagespace inside 20 seconds (so far
>>> anyway - touch wood!), but I would have hoped that if it was a memory
>>> issue, that the limit stated here would be adhered to.  I'd rather
>>> GeoServer barfed an OutOfMemory error than it taking down my entire
>>> server.  Should this limit have stopped the problem?
>>>
>>> The next thing for me to try (within GeoServer at least) is to try and
>>> set up some MinScaleDenominators in the SLDs so that these features
>>> aren't displayed beyond 1:10.
>>>
>>> Are there any other suggestions on how to make this work/fallover
>>> gracefully?
>>>
>>
>> My guess is that you're using a dash array on the lines of those
>> polygons. There is a well known java2d bug that makes the rendering
>> bomb out.
>>
>> We could clip the line before passing it to the renderer, but doing it
>> properly is not easy and so far nobody every had spare time to work on
>> it, nor paid time.
>>
>> Cheers
>> Andrea
>>
>>
>> --
>> Andrea Aime
>> OpenGeo - http://opengeo.org
>> Expert service straight from the developers.
>>
>>
>> ------------------------------------------------------------------------------
>> The Palm PDK Hot Apps Program offers developers who use the
>> Plug-In Development Kit to bring their C/C++ apps to Palm for a share
>> of $1 Million in cash or HP Products. Visit us here for more details:
>> http://p.sf.net/sfu/dev2dev-palm
>> _______________________________________________
>> Geoserver-users mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/geoserver-users
>>
>>
>
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>
>
>
------------------------------------------------------------------------------
This SF.net email is sponsored by 

Make an app they can't live without
Enter the BlackBerry Developer Challenge
http://p.sf.net/sfu/RIM-dev2dev 
_______________________________________________
Geoserver-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/geoserver-users

Reply via email to