The Dreaded PermGen Error

[Brian Reindel]

We have experienced similar permgen issues on JBoss EAP 6.1 in domain
mode when we don't stop and start the host controller where the IdP is
deployed. Occasionally we'll even have to stop the whole domain.
However, this isn't a problem unique to Shibboleth. Most containers
don't do well with deployments unless you do a full stop, especially
if you have a considerable number of library dependencies like Spring
that initialize a bunch of beans at startup.

As far as which classes are consuming that memory and potentially
"leaking", there are very few ways to uncover that, but the first is
to configure the JVM to do a heapdump when it runs out of memory. The
second is to implement a production grade monitoring tool. Those can
be expensive though, and do add some overhead.


On Tue, Nov 12, 2013 at 11:32 AM, Christopher Bongaarts <cab at umn.edu> wrote:
> On 11/12/2013 8:27 AM, Peter Schober wrote:
>> * Christopher Bongaarts <cab at umn.edu> [2013-11-12 17:22]:
>>> On 11/12/2013 5:42 AM, Timo Tunturi wrote:
>>>> Using ecmascript in the resolver caused leakage from hell into
>>>> permgen for me. Porting scripts to type mapped or template got rid
>>>> of that.
>>> I think removing the optimizer from the rhino .jar might take care of
>>> that problem.
>> Just a data point: I've not seen those problems when using JavaScript
>> in the resolver and I did not remove the rhino optimzer.
>
> Nor have I, and we have several Script attributes.  I guess the rhino
> thing was strictly a Terracotta issue.
>
> --
> %%  Christopher A. Bongaarts   %%  cab at umn.edu          %%
> %%  OIT - Identity Management  %%  http://umn.edu/~cab  %%
> %%  University of Minnesota    %%  +1 (612) 625-1809    %%
>
> --
> To unsubscribe from this list send an email to users-unsubscribe at shibboleth.net