[BUG?] sendfile / distcc

[BUG?] sendfile / distcc

[Peter Zijlstra]

Hi,

I've been seeing this for a while now, but its still present with -rc5+
kernels. Its always the local machine failing, never the remotes.

[root@twins linux-2.6-rt]# nice ./k opteron build
time distmake x86_64 O=opteron-build ARCH=x86_64
testing twins:    distccd running
testing opteron:  DOWN
testing taijtu:   distccd running
testing dyad:     distccd running
testing lappy:    host running
testing ubuntu:   host running
DISTCC_HOSTS= twins/8 taijtu/32 dyad/32
DISTCC_DIR=/dev/shm/distcc
DISTCC_SLOTS=72

...

distcc[9931] (dcc_writex) ERROR: failed to write: Connection reset by peer
distcc[9931] Warning: failed to distribute /mnt/build/linux-2.6-rt/kernel/fork.c to twins/8, running locally instead

...

distcc[12078] (dcc_pump_sendfile) ERROR: sendfile failed: Connection reset by peer
distcc[12078] (dcc_readx) ERROR: unexpected eof on fd4
distcc[12078] (dcc_r_token_int) ERROR: read failed while waiting for token "DONE"
distcc[12078] Warning: failed to distribute /mnt/build/linux-2.6-rt/init/version.c to twins/8, running locally instead

Re: [BUG?] sendfile / distcc

[Andrew Morton]

On Wed, 10 Sep 2008 11:27:51 +0200 Peter Zijlstra <peterz@infradead.org> wrote:

> Hi,
> 
> I've been seeing this for a while now, but its still present with -rc5+
> kernels. Its always the local machine failing, never the remotes.
> 
> [root@twins linux-2.6-rt]# nice ./k opteron build
> time distmake x86_64 O=opteron-build ARCH=x86_64
> testing twins:    distccd running
> testing opteron:  DOWN
> testing taijtu:   distccd running
> testing dyad:     distccd running
> testing lappy:    host running
> testing ubuntu:   host running
> DISTCC_HOSTS= twins/8 taijtu/32 dyad/32
> DISTCC_DIR=/dev/shm/distcc
> DISTCC_SLOTS=72
> 
> ...
> 
> distcc[9931] (dcc_writex) ERROR: failed to write: Connection reset by peer
> distcc[9931] Warning: failed to distribute /mnt/build/linux-2.6-rt/kernel/fork.c to twins/8, running locally instead
> 
> ...
> 
> distcc[12078] (dcc_pump_sendfile) ERROR: sendfile failed: Connection reset by peer
> distcc[12078] (dcc_readx) ERROR: unexpected eof on fd4
> distcc[12078] (dcc_r_token_int) ERROR: read failed while waiting for token "DONE"
> distcc[12078] Warning: failed to distribute /mnt/build/linux-2.6-rt/init/version.c to twins/8, running locally instead
> 

distcc is very good at breaking networking.  I assume 2.6.26 is OK?

Re: [BUG?] sendfile / distcc

[Peter Zijlstra]

On Wed, 2008-09-10 at 23:44 -0700, Andrew Morton wrote:
> On Wed, 10 Sep 2008 11:27:51 +0200 Peter Zijlstra <peterz@infradead.org> wrote:
> 
> > Hi,
> > 
> > I've been seeing this for a while now, but its still present with -rc5+
> > kernels. Its always the local machine failing, never the remotes.
> > 
> > [root@twins linux-2.6-rt]# nice ./k opteron build
> > time distmake x86_64 O=opteron-build ARCH=x86_64
> > testing twins:    distccd running
> > testing opteron:  DOWN
> > testing taijtu:   distccd running
> > testing dyad:     distccd running
> > testing lappy:    host running
> > testing ubuntu:   host running
> > DISTCC_HOSTS= twins/8 taijtu/32 dyad/32
> > DISTCC_DIR=/dev/shm/distcc
> > DISTCC_SLOTS=72
> > 
> > ...
> > 
> > distcc[9931] (dcc_writex) ERROR: failed to write: Connection reset by peer
> > distcc[9931] Warning: failed to distribute /mnt/build/linux-2.6-rt/kernel/fork.c to twins/8, running locally instead
> > 
> > ...
> > 
> > distcc[12078] (dcc_pump_sendfile) ERROR: sendfile failed: Connection reset by peer
> > distcc[12078] (dcc_readx) ERROR: unexpected eof on fd4
> > distcc[12078] (dcc_r_token_int) ERROR: read failed while waiting for token "DONE"
> > distcc[12078] Warning: failed to distribute /mnt/build/linux-2.6-rt/init/version.c to twins/8, running locally instead
> > 
> 
> distcc is very good at breaking networking.  I assume 2.6.26 is OK?

I think it was - whew .26 was a long time ago already ;-)

I'll build and boot one today to confirm.

Re: [BUG?] sendfile / distcc

[Peter Zijlstra]

On Thu, 2008-09-11 at 10:26 +0200, Peter Zijlstra wrote:
> On Wed, 2008-09-10 at 23:44 -0700, Andrew Morton wrote:
> > On Wed, 10 Sep 2008 11:27:51 +0200 Peter Zijlstra <peterz@infradead.org> wrote:
> > 
> > > Hi,
> > > 
> > > I've been seeing this for a while now, but its still present with -rc5+
> > > kernels. Its always the local machine failing, never the remotes.
> > > 
> > > [root@twins linux-2.6-rt]# nice ./k opteron build
> > > time distmake x86_64 O=opteron-build ARCH=x86_64
> > > testing twins:    distccd running
> > > testing opteron:  DOWN
> > > testing taijtu:   distccd running
> > > testing dyad:     distccd running
> > > testing lappy:    host running
> > > testing ubuntu:   host running
> > > DISTCC_HOSTS= twins/8 taijtu/32 dyad/32
> > > DISTCC_DIR=/dev/shm/distcc
> > > DISTCC_SLOTS=72
> > > 
> > > ...
> > > 
> > > distcc[9931] (dcc_writex) ERROR: failed to write: Connection reset by peer
> > > distcc[9931] Warning: failed to distribute /mnt/build/linux-2.6-rt/kernel/fork.c to twins/8, running locally instead
> > > 
> > > ...
> > > 
> > > distcc[12078] (dcc_pump_sendfile) ERROR: sendfile failed: Connection reset by peer
> > > distcc[12078] (dcc_readx) ERROR: unexpected eof on fd4
> > > distcc[12078] (dcc_r_token_int) ERROR: read failed while waiting for token "DONE"
> > > distcc[12078] Warning: failed to distribute /mnt/build/linux-2.6-rt/init/version.c to twins/8, running locally instead
> > > 
> > 
> > distcc is very good at breaking networking.  I assume 2.6.26 is OK?
> 
> I think it was - whew .26 was a long time ago already ;-)
> 
> I'll build and boot one today to confirm.

It seems I can reproduce similar errors with .26 - I'll be traveling
soon and this box is one of those without serial console so It'll have
to wait until I get back.