On Wed, 18 Dec 2002, Traylen, SM (Steve) wrote:
> The globus error is expected since as you say the globus rpms
> are not installed yet. This should not matter since the globus rpms
> should be installed at the first reboot during installation. It is
> that bit that must be failing.
>
> Is there anything intresting the /var/obj/log/updaterpms
> log. Conflicts in rpms or a lack of space on the node are
> usually the culprit.
I couldn't spot anything in the logs, but after looking through the XML
profiles I spotted what was wrong and it now seems ok.
To get this far, I had to:
Do the BASE-rpm install first (I thought that was fixed) to avoid the
RPM dependency loop [FAIL] error due to the Atlas RPM's. This required
commenting out the CE includes from my CE's profile file in
/var/obj/conf/profile/source
I then had to uncomment the CE stuff and add the line +update.rpmcfg
CE-rpm before the ComputingElement-cfg.h include otherwise it still did a
BASE-rpm install, which fails during the Globus and EDG configuration due
to missing RPMs.
I then had to modify lcfg-client-rpm.h (and lcfg-server-rpm.h) because
the obj-infoproviders-2.0.0-4 RPM was actually obj-infoproviders-2.0.0-5
with the wrong filename. This caused an endless loop in LCFG as it
installs the obj-infoproviders-2.0.0-4 file, but gets
obj-infoproviders-2.0.0-5 in the RPM database, which needs downgrading to
obj-infoproviders-2.0.0-4, ...
In the end I solved this by changing the -4 to -5
I note that the obj-infoproviders-2.0.0-4 RPM has disappeared from
datagrid.in2p3.fr since Tuesday when I grabbed them, but that it's still
wrong (-4) in the CVS edg-release copy of lcfg-*-rpm.h
So, I think we need to add the recommended way of sidestepping the Atlas
dependency problems to the install instructions; and tell people to change
to obj-infoproviders-2.0.0-5 in lcfg-*-rpm.h until the CVS catches up.
SE and WN next ...
Cheers
Andrew
|