[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Can the NFS media location be passed to anaconda as a kernel arg?



On 4/28/2010 4:36 PM, Wes Hardin wrote:
> On 04/28/2010 01:15 PM, Kyle McDonald wrote:
>> I want to setup PXE network installs that are interactive, but I'd like
>> to limit the amount of questions the user has to enter for each one. The
>> PXE boot already passes some kernel args to configure the serial
>> console, and force a vnc install. I'm wondering if I can also supply the
>> NFS location for the media as a kernel arg so that users don't need to
>> type it in everytime (It's always the same, and it's long.)
>>
>> Also I know when doing KickStart installs, you can use ks-device=eth1 to
>> avoid being prompted for the network device when there is more than one.
>> Can I use this same option so that the interactive user doesn't have to
>> know which eth to pick?
>>
>> On a related note, is there a way to have anaconda figure out which
>> interface the PXE boot booted the machine on and just go with that one?
>> Right now I need to make 2 different PXE boot config files for the 2
>> types of machines I have because one boots on eth0 and the other on
>> eth1.
>
> I don't think media location is something you pass in on the kernel
> line. Instead, what you can pass in is the path to a kickstart file
> which then answers a lot of those repetitive questions, such as the
> location of the install tree. The kickstart file can be as complete or
> as minimal as you wish.  I have one that only contains the network
> location of the install tree and waits for the installer to answer the
> rest.
That blows.

I already have a PXE config file for each release of RHEL SLES and
Solaris I need to install. These are required to direct the network boot
to the right release. For SLES and Solaris, I can put the media location
in this PXE config file along with the install config file (AutoYast or
JumpStart), and I can get away with just 1 install config file for all
my machines (I use the begin and finish scripts to tweak the profile
based on what release was booted so 1 file really helps me out since
when the I do need to make a code update I only need to edit it once.)

But for RHEL (and CENTOS) I have to have 2 (32 and 64 bit) KS config
files for each release that only differ by the 'nfs ....' line.

Now If I want to have an interactive install where the media location is
already set, I need to go and make 2 more KS files for every release?
That's # releases * 4 files to update when and if the rest of the file
needs to change. And the 'nfs ...' line is one of the things that the
%pre can't generate when it writes out the file to be included.

Whereas to add an interactive option for Solaris and SLES, I just copy
and paste the boot section in the PXE config file and remove the kernel
arg that specifies the automated install info.

That's a shame.

Has this ever been requested before? If so anyone know the rationale for
not doing it?

 -Kyle
>
> As for using the same interface that you PXE'd from, check out the
> ksdevice=bootif option, described briefly here:
>
> http://wiki.centos.org/TipsAndTricks/KickStart
>
> This document from Dell mentions it too:
>
> http://linux.dell.com/files/whitepapers/nic-enum-whitepaper-v2.pdf
>
> I don't claim to have figured this out myself, it's set by default on
> all my Cobbler PXE profiles.
>
> /* Wes Hardin */


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]