Starting a FAI installation using kexec - technical thread
Torsten Schlabach
tschlabach at gmx.net
Thu May 28 12:56:15 CEST 2009
Hi again!
> propose, but also kexec seems not to be what you need - kexec just
> changes the kernel, but keeps all processes and stuff runnning,
No, it doesn't. It would be wonderful if a tool would exist which "just
changes the kernel" on a running system. Starting a new kernel with
kexec is like a reboot: The newly loaded and executed kernel will
re-initialize the memory, start a new init process, etc. Anything which
has existed in memory prior to calling kexec -e will be gone.
A minor difference between shutdown 0 -r and kexec is that with a kexec
"reboot" there will be no re-initialization of the hardware because the
BIOS will not re-start. In most cases this shouldn't be a problem though.
> So, you say the main problem/change for you now is that FAI doesn't
> have a special kernel anymore.
Well, I don't say this is a problem. It's more that I am asking for
clarification because I need to find out what the difference is between
what has been working for me in older FAI versions (with a monolithic
and specially configured install kernel) and the new versions.
To ask a simple question:
There is no FAI install kernel anymore; FAI is using a stock Debian
kernel now, right?
Is there mabye a FAI initrd or is the initrd also the "stock" one that
comes from the corresponding Debian package. Or is my initrd which is
used in a PXE boot generated with any special settings possibly when I
execute make-fai-nfsroot?
I hope these two questions can be answered straight with a yes or no by
someone who knows that.
What I think my problem might come from is that in each "NFS root file
system" howto it says "the NFS support must be statically compiled into
the kernel". But that's definitely not the case for the stock Debian
kernel. So I wonder if that remark in most howtos is just void today.
Also quite honestly I could use some pointer to understanding that /live
dir in the initrd which is used by FAI today ... What's in there which
is no longer in the NFS root for example? Does the whole install run out
of the /live dir now?
Regards,
Torsten
Henning Sprang schrieb:
> Torsten Schlabach wrote:
>> But the basic-installed systems wasn't installed using FAI.
>
> That's not necessary for softupdates.
>
>> It needs an
>> entirely new partitioning of the hard disk,
>
> so you want to wipe the whole system anyway. O.K.
>
>> it might have an entirely
>> different kernel than what we want to install,
>
> You can install another one
>
>> it might not even be
>> Debian yet.
>
> No problem, softupdate can manage other systems as well :)
>
>
> But anyway, I now understand, you want to do something different than I
> propose, but also kexec seems not to be what you need - kexec just
> changes the kernel, but keeps all processes and stuff runnning, it
> doesn't boot from scratch, doesn't it? (at least that is what I
> understand what it's made for).
>
> So, you say the main problem/change for you now is that FAI doesn't have
> a special kernel anymore.
>
> But you can do exactly the same with the normal debian kernel. It can
> boot from nfsroot just as the special FAI kernel did, and it should be
> possible to give the required kernel parameters on the grub command line
> - just as you did before with the special FAI kernel.
> Only that the normal debian kernel also needs an initrd. But when
> booting it via grub that is even easier than with PXE.
>
> Did you try this already before trying kexec?
>
> Another idea might be to run the provider's rescue system and try to run
> setup-storage directly from there.
> But you seem to have a very heterogeneous environment, and cannot rely
> on anything.
>
> Apart from this being an interesting problem, to solve, I wonder if it
> would simpler to force customers to rent th Server from a set of
> known-working providers that fullfill some requirements - developing
> something that works with every random system can become very expensive
> - either for you or for the customers.
>
>
> Henning
More information about the linux-fai
mailing list