The Dreaded PermGen Error

[Cantor, Scott]

On 11/12/13, 5:32 AM, "Peter Schober" <peter.schober at univie.ac.at> wrote:

>* Joshua Riffle <jriffle at apu.edu> [2013-11-11 17:24]:
>> 1) How do you detect or prevent the evil PermGen error beforehand?
>
>Fyi: http://plumbr.eu/blog/busting-permgen-myths

I think paraphrasing that: if we had a situation where the IdP in normal
use was leaking permgen space, we would see that as a major bug. Whether
it was our fault, or a library, or the container.

We have a lot of existence proofs that this is not the case, though it
sounds like there may be specific features, Terracotta for instance, that
may lead to this. I myself have not seen this issue with the use of
scriptlets, though. Not in Javascript anyway.

OTOH, we know that restarting the servlet context does leak. I don't know
that we know why exactly, but we don't see context restarts as a supported
mechanism, and in place of that we do reloading of pieces of the IdP if
you want it to do that. Those mechanisms *should not* leak, and I don't
think they do at this point.

As far as V3 is concerned, I don't think we planned on trying to fix the
context reloading issue. We are discussing exactly what to do about
reloading the subsystems, and there's some discussion of moving that to a
more Spring-like approach that just reloads the entire Spring context (not
the same as the servlet context) instead of pieces individually. Honestly
it just depends on what turns out to be easiest and safest.

We're really not done enough yet to start doing real analysis/testing of
whether we have leaks or not in the new code base.

-- Scott