Looking at http://java.sun.com/docs/hotspot/gc/index.html (Other considerations)
you seem to be right.
To delay the OOME you have to set the -XX:MaxPermSize option (default is 64m)
(http://java.sun.com/docs/hotspot/VMOptions.html). But the OOME will occur
anyway, as der permanent heap gets filled up.
So the only way the avoid it seems to turn caching back on.
Cheers,
Markus
> >Seems like tapestry is doing something funny under the covers, since so
> >many people seem to be having the same problem.... But of course,
> >Tapestry is "sitting on top" of a lot of other code (JBoss, hibernate,
> >etc.... In my case), but it's just too hard to ignore since once
> >disable-caching=false there is no problem, but when, true, the OOME
> >occurs.....
>
> When I looked at this in Jan 31 (see archives) the error I ran into was an
> OOME. When I looked at the actual stack-trace I could see that it was the
> permanent heap that was out of memory. This is significant since the
> permanent heap is used for storing classes and types. I have never seen a
> profiler that analyses the permanent heap in detail - it would be interesting
> to know if someone knows if such a tool exists.
> The permanent heap is most probably never garbaged collected, this would
> amount to throwing away class definitions. This is usually not a problem
> since an ordinary application has a finite number of classes defined and when
> they are loaded that is it.
> Now, when in development mode I assume, and this is just a guess, that
> Tapestry is enhancing and creating _new_ classes every time a page is loaded.
> This will fill the permanent heap after a while and when that is full you'll
> get OOME.
> When Tapestry can cache the classes it creates the permanent heap will not
> fill up with new classes and everything should work as expected.
> So this will only be a problem when in development mode, I wouldn't even
> consider it a bug in Tapestry (this can happen with JSP compilers also),
> merely a fact of life as a developer :-)
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
