Re: commSetTimeout and cbdata

On Tuesday 09 April 2002 10:42, Chemolli Francesco (USI) wrote:
> > > Maybe this should be fixed in 2.5 as well. Unfortunately the
> > > cbdata branch is a bit too intrusive for 2.5 at this time I
> > > think..
> >
> > This may be the leak in NTLM I was referring to a while back -
> > kinkie had a _lot_ of unfreed connection structs. If we can fix
> > it in 2/5, that
> > would be great.
>
> Please please do.
>
> This is a showstopper for heavy NTLM deployment IMO.
> As it is now, I'm leaking about 3Mb/hour (with 150
> to 200 hits/second peak).

I don't see how fixing commSetTimeout would stop any leaks. The only
thing fixing this will stop is some (not very) random memory
trashing, and maybe cause some new leaks (but I think I have that all
covered now).

If you have the problem that a lot of connection structures (or other
cbdata registered structures) get piled up but nothing else (like
filedescriptors etc) then you are most likely missing a
cbdataUnlock() somewhere. The new cbdata code makes it slightly
easier to trap such errors and deal with them automatically.

If you have a leak not showing up in "Memory Utilization" then there
is a missing safe_free() somewhere. memprof immediately where the
problem is in such case with only a very small performance penalty.

Regards
Henrik
Received on Tue Apr 09 2002 - 03:19:09 MDT