[distcc] Re: Distcc freezes during compile

[Martin Pool]

On  3 Feb 2004, Stephen torri <storri at torri.org> wrote:
> On Tue, 2004-02-03 at 09:34, Dag Wieers wrote:
> > > This is a kernel bug -- applications should never hang the machine.
> > > Switch to stock 2.4.24 and see if the problem can still be reproduced.
> > > If it can, let us know.  If that works, see if the 2.4.24 preemption
> > > patch works.
> > 
> > There has been a kernel memory leak in the 2.4.18/2.4.20 RH kernels. 
> > Depending on the use your machine could run out of memory in a couple of 
> > hours up to weeks. I've seen it happen on boxes that have high IO-load and 
> > CPU-usage, mainly with rrdtool and mrtg. You can find more information 
> > here:
> 
> I will try to run a non-custom kernel today.

Thanks.

> I find it strange that it takes distcc 2.12.1 to cause this "bug" to
> occur while 2.5 did the same job.

Why is it strange that one program provokes a bug and another doesn't?

> I need an actual proof first in order to have a simple test case
> which I can then send to the colleagues which developed this custom
> kernel patch.

What patch is it?

I think it's more likely that your kernel is at fault than distcc is:

 - 2.4.18 is old and buggy.

 - Preemption for 2.4 is not very stable.

 - Preemption bugs are very likely to cause hard freezes.

 - Applications should never be able to hang the kernel, especially
   unprivileged applications.  By definition if this happens it must
   be a kernel bug.

 - Every other similar bug has turned out to be a kernel bug.

 - No one else has reported similar failures, and other people are
   using 2.12.1 but probably few other people use 2.4.18-preempt.

You can probably derive a test case from the diff between 2.5 and
2.12.1.  From past experience it is likely to be something in the
network IO, perhaps sendfile.  However, if you can reproduce the bug
it is probably more useful to do a binary search for the kernel patch
that introduces the bug.

-- 
Martin 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
Url : http://lists.samba.org/archive/distcc/attachments/20040204/5d904d11/attachment.bin