memory leak?

[Cantor, Scott]

On 9/12/11 11:33 AM, "Steve Wills" <swills at webassign.net> wrote:
>
>Thanks for taking a look. That matches what I'm seeing, which is that RAM
>usage of shibd always goes up. Traffic peaks at night, then goes down in
>the early morning each day, but over the course of a week or so, memory
>usage of shibd only goes up and never seems to level off as I would
>expect. We didn't see this with shib1, so I assume the issues you mention
>didn't exist there?

They did, the code just had more leaks plugged over time.

I've found several more leaks, enough to be significant.

That doesn't change the Linux point, the virtual memory usage will never
decrease in a Linux environment.

The leaks aren't tiny, but hundreds of sessions would be equivalent to a
single over-large stack if you're using prefork, which you didn't specify.

>>I guess that depends on how many new sessions we're seeing and how much
>>RAM there is? Could you estimate how much RAM per session we're talking
>>about?

It's not small, probably upwards of 30k or more per session if you receive
a lot of attributes. You can decrease usage somewhat by making sure it's
not caching assertions when they aren't needed.

All the leaks I found are being patched and I'll do more testing over the
course of work on 2.5.

-- Scott