Memory Leak with rapid connects and disconnects

[mctozzy]

We believe we've found a mem leak which is causing a bit of a problem!
We are using H2 in a pattern where connects and disconnects are short
lived. (Although we're going to change this).

What's happening is that in method traceSystem.getTrace() it adds a
new trace to the hashmap for each module that calls this method. We
see a module corresponding to each H2 session going through this call,
so the hashmap just keeps getting bigger and bigger with no removes
when connections are closed.

What's even worse is we suspect this is also causing a creep up in the
permgen memory pool (because we suspect that hashmap keys are
intern'd) , which is (by default) 64MB. So in our case, the permgen
pool blew up before the regular heap. Which led to a lot of head
scratching because it's much harder to see what's going on in permgen
than in the normal heap.

Cheers,
Milt

[Thomas Mueller]

Hi,

Thanks a lot for your help! I will add a test case for this problem.
However so far I can not make it go out of memory with my test, even I
am opening and closing a lot of connections. Anyway, I will fix it.

Thomas

[mctozzy]

>
> Thanks a lot for your help! I will add a test case for this problem.
> However so far I can not make it go out of memory with my test, even I
> am opening and closing a lot of connections. Anyway, I will fix it.
>

I should have clarified that it might actually take a long time (days)
before you actually run out of memory, because the amount of mem
leaked per new connection is not large. We found this leak using
yourkit java profiler, while we were looking for leaks in general.

[Thomas Mueller]

Hi,

I also use the YourKit Java Profiler, a great tool in my view. I will
fix this problem of course, even if the leak is not that big.

Thanks a lot!
Thomas

On 5/3/07, mctozzy <mct...@gmail.com> wrote:
>
> >