David Timms wrote:
Howard
Wilkinson wrote:
...
First the version of yum available in updates
currently is incompatible with the version of anaconda. The problem is
that this version of yum does not create the headers directory and
anaconda assumes it will exist.
anaconda-11.2.0.66-1.i386.rpm
yum-3.2.1-1.fc7
Which headers directory ? {the dvd/everything repo doesn't have such a
folder}
The headers directory is populated (referenced?) by anaconda during a
kickstart only and is expected to have been created by yum, not sure
what it is used for but I found an obscure references to this in one of
the bulletin boards and tested it out. I have not tried the development
combination yet as this requires some serious rebuilding in our
environment.
To get round this we have reverted to
yum-3.2.0-1.fc7.noarch.rpm for now but will probably try out the
development releases of anaconda and yum next week.
It's very close {< week?} to freeze time for next fedora, if this an
issue, prompt bugzilla check/report would be good.
Where do I put this some more pointers would be helpful?
More worrying is a problem with dependencies
in building the kernel initrd images. The newly built Everything regime
has the latest kernel - kernel-2.6.21-1.3228.fc7 - as available from
updates.
I don't think this is correct:
http://download.fedora.redhat.com/pub/fedora/linux/releases/7/Everything/i386/os/Fedora/
kernel*-2.6.21-1.3194.fc7.*.rpm
However, if you have inserted the newer kernel yourself, or are making
updates repo available during the install then that is different.
I have taken the Everything build and replaced the packages in the
image with those available in updates from a week ago (or thereabouts)
and then rebuilt the images so I can kickstart directly from it. We use
NFS based builds so additional repositories are a problem. Same issue
arises with the older kernel if I rebuild the image so the pkgorder is
not running with enough arguments I think (worked this out last night)
now need to confirm that I can force mdadm to be loaded earlier.
In the kickstart.cfg can you request particular version install:
kernel-2.6.21-1.3194.fc7.*.rpm ?
I already forced the new kernel by removing the old one and allowing
the new one to be the only one in the set.
This issue might also be checked/reported in bugzilla. Developers
expected that people might want to enable repos during install, but
perhaps had not considered the updates repo being active, and hence
installing only the updated kernel/yum. IMHO ideally, I would want the
release kernel and the latest released update kernel installed. If you
use this process without the updates repo active, does it work ?
But when the kickstart install runs the
mkinitrd cannot find /sbin/mdadm as the mdadm package is not a
dependency of the kernel (and should not be for all builds but ....[1])
This has the effect of allowing the build to complete but when the
system reboots LVM cannot find the logical volume groups because the
raid has not started.
/We build our systems with mirrored boot partitions on /dev/md0 and
mirrored /dev/md1 on the remainder of the disc with logical volumes
on this portion of the disc including the root partition. Hence the
problem above.[1]/
To get round this we are rebuilding the kernel to require the mdadm
package - but this is really not what we should be doing. Somewhere in
the RPM, YUM, Anaconda, regime we need to be able to specify that mdadm
(and lvm2) needs to be installed prior to the kernel installation, only
for our particular configurations not in general. Any ideas anybody?
I tried the soft raid /boot before in fc5 and found it no go. From
memory there was work planned for either f7 or f8 to make this
possible. Perhaps checkout http://fedoraproject.org/wiki/Grub2ToDo
We have had this configuration working since FC4 and build all of our
systems this way!
DaveT.
SO how do I feed package order data into the buildinstallimage script -
this may fix the second problem.
Howard
--
Signature
Howard Wilkinson
|
Phone:
|
+44(20)76907075
|
Coherent Technology Limited
|
Fax:
|
|
23 Northampton Square,
|
Mobile:
|
+44(7980)639379
|
United Kingdom, EC1V 0HL
|
Email:
|
howard@xxxxxxxxxxx
|
|