Re: Squid 3.1.0.5 uses memory without bounds

lör 2009-02-14 klockan 11:35 +0100 skrev Steinar H. Gunderson:
> On Fri, Feb 13, 2009 at 09:20:24AM -0700, Alex Rousskov wrote:
> > Was --enable-valgrind used last time you reconfigured and rebuilt
> > Squid from scratch?
>
> Yes. It didn't actually appear to do much, though.

In squid-3 it's --enable-valgdind-debug

If you get a valgrind report in cache.log (or stderr) when visiting the
memory statistics cahemgre page (mgr:mem) then it's compiled correctly.
I think there is also a leak summary in the page itself.

With this enabled and running under valgrind the memory allocation
policy changes slightly to allow for better leak tracing.

> The Valgrind traces are inaccurate for the still reachable objects, I'm
> pretty sure. (The pprof traces are better there.) For the actual leaks I have
> more faith in them. :-)

There is two items being the main interesting ones here, both in a
running snapshot, not at shutdown:

a) Confirmed leaks by valgrind.

b) Memory pools which don't stop growing. Identified by their age being
0.0 all the time.

and additionally

c) Memory access errors reported by valgrind.

Regards
Henrik
Received on Sun Feb 15 2009 - 17:41:58 MST