[mythtv-users] Best options/howto for diskless frontends

[Paul Gardiner]

On 21/02/2013 20:11, phipps-hutton at sky.com wrote:
>
> Quoting Paul Gardiner <lists at glidos.net>:
>> It tries very hard to work. The boot gets past the nfs mount stage and
>> gets some way through the init, but then it just stops. No shell prompt,
>> and I can't ssh in, so it's difficult to tell where it went wrong.
>> Here's a photo of the screen:
>> http://intranet.glidos.net/~paul/boot-failure.jpg.
>>
>> On the server side, I can see three
>> successful dhcp runs, with a tftp access between first and second, and
>> an nfs mount between second and third. I feel like it's close, but I'm
>> out of ideas what to try to get it the rest of the way.
>
> It looks like a deadlock due to it trying to configure eth0. It probably
> runs if-up which usually disabled the interface before enabling it. As
> soon a sit tries to access any files while the interface is down it will
> deadlock. Your first DHCP is from the BIOS doing the PXE booting, the
> second is from the initrd when given the boot=nfs argument. The third is
> probably from if-up which will try to write the result to disk and
> deadlock.
>
> I got around this by editing /etc/network/interfaces to say
>
> iface eth0 inet manual
>
> instead of
>
> iface eth0 inet dhcp
>
> so if-up leaves it alone. You may have to run
>
> echo "nameserver 192.168.0.1" | resolvconf -a eth0
>
> afterward if you want to access the internet (replace 192.168.0.1 with
> the IP address of your router).

You genius. That was it. I got around it in a slightly different way. I
stuck the disk back in and ran yast2 to perform the change from dhcp to
static. That way I could set up dns and default route in an environment
that I was more familiar with. I then rsync'd again and noticed just
a file in /etc/sysconfig had changed. It had the same effect: no longer
did it deadlock.

Then my one remaining problem was that, out of the two kernels on my
system, I'd picked the wrong one for mkinitrd (I'd assumed it would be
the one that had a soft link called vmlinuz). It mostly worked with the
wrong kernel, but the nvidia module wouldn't load, presumably because
I'd built it for the other kernel.

Anyway, now it works. Thank you.

I'm not sure I completely understand the exact nature of the deadlock.
I know the nfs mount was operational before that stage. Would the
execution of if-up have temporarily disconnected it?

Cheers,
	Paul.