memory leak?

[Cantor, Scott]

On 9/12/11 10:49 AM, "Steve Wills" <swills at webassign.net> wrote:
>
>Is anyone aware of any memory leaks in shibd?

I haven't run it under valgrind in a while, but there were very few, if
any, the last time I looked, and what few there were were not in my code.

I'm going to find some time to run it under valgrind again to see if
anything has changed.

> We've noticed ours seems to grow continuously in terms of memory usage
>and have had it crash once due to running the system out of RAM, and it
>seems to be on track to happen again. We see shibd using 659m of memory
>now. Is this within normal expectations?

That is not a lot of reserve in the scheme of things if you have a lot of
sessions in memory. Linux also does not release RAM back to the OS when
its freed, unlike Windows. Whatever heap it allocates stays allocated
until the process is freed, so whatever peak usage exists while dictate
how much it's using. It will reuse the heap when the load dies down and
then picks up again.

Other factors depend on what you choose to do. If you use long timeouts,
then the sessions will live for a long time and build up. If you use
Apache in prefork mode, you are asking for a huge number of threads and
the default stack size on Linux is quite large. That's also adjustable and
doesn't need to be more than 100K or so, if that.

-- Scott