Re: Building packages with buildd and qemu-kvm

[Samuel Thibault]

Svante Signell, le Wed 13 Apr 2011 11:05:30 +0200, a écrit :
> On Wed, 2011-04-13 at 10:43 +0200, Samuel Thibault wrote:
> > Svante Signell, le Wed 13 Apr 2011 07:46:54 +0200, a écrit :
> > > make[1]: *** [lapack_testing] Error 141
> > > 
> > > Both tests end with a signal being sent for a kill from fakeroot in the
> > > first case and in libfakreoot during a read operation. Both had happen
> > > during redirection of stdin and stdout. Why issuing a kill in the
> > > middle of execution? Are these functions also handled by libpipe? If
> > 
> > 141 is 128 + 13, i.e. SIGPIPE. This is supposed to happen when the
> > reader part of a pipe has disappeared while the writer was trying to
> > write to the pipe. It's not supposed to happen on the reader part.
> 
> How to interpret your answer, is there a problem with libpipe or not?

Err, yes. "It's not supposed to happen" => there's a bug. Since it's on
a pipe, it's most probably on pflocal or libpipe. glibc could be at
stake, but I doubt it, since it doesn't do anything different for files
or for pipes.

> BTW: The kill command comes from a timer called from in debian/rules. I
> tried to increase TIMEOUT from 1800 to 3600, but the build seems to bail
> out at the same place for me. Trying again with TIMEOUT:=4800 now.

Ah! then it's most probably a way simpler, different story: the reader
probably just got hung, and debian/rules simply killed it out of
patience, which thus made the writer get SIGPIPE, which is normal. So
the real bug is actually happening simply in the atlas process at the
read end. When you see no progress from the build, check out the last
started process and what they are doing.

> > > I've also seen very strange things happen not related to the actual
> > > builds. A hurd/gnumach/libc bug??
> > 
> > The pipe issue is probably more from libpipe.
> 
> I don't get it, libpipe is a hurd library, right?

Yes, and?
It's used by pflocal to implement pipes.

Samuel