Antwort: Re: Antwort: Re: [Samba] 3.0.4: smbd's + nscd's = 100% CPU; load> 4

[Jeremy Allison]

On Wed, Jul 14, 2004 at 12:57:22AM +0200, Dragan.Krnic at bahn.de wrote:
> >> I checked my /etc/group for irregular entries and repaired
> >> all of them. Since then the problem didn't occur any more.
> >> Even so it wasn't very frequent.
> >>
> >> I still have a bunch of strace logs of smbd and of nscd.
> >> Cann I upload it somewhere for you if you can use it?
> >>
> >> Or do you suggest to replant the erroneous entries and
> >> wait for the loop ?
> >
> > Actually if you could add one erroneous entry and catch it
> > in gdb in the act of looping that would be very helpful. I'm
> > still trying to determine if the bug is in glibc or smbd.
> 
> Noblesse oblige.
> 
> OK. I've never debugged runaway programs.
> Does it stay still when I attach it?
> Should I compile a debug version for that?
> Can you be just a little more specific
> as to what I should try to clarify?
> 
> You seem to believe, that it is not
> looping in samba sources, because you
> can't see where it possibly could.
> 
> So probably, if the attached process
> can be stepped, I should step a full
> circle between two socket calls and
> see if it ever lands back in samba
> code. Is that it?

Yes, that's the idea. You need to compile with -g
option and then to attach to the process once it's
looping. Then set a breakpoint in the smbd code at
the point I referred to and see if it ever arrives
there. If it does follow it along until it loops again.

Thanks,

Jeremy.