[Lustre-discuss] builds: lustre-2.0.2alpha (1.9.181)+ HEAD (1.9.190) with kerberos enabled

[Brian J. Murrell]

On Fri, 2009-05-22 at 21:32 -0400, Josephine Palencia wrote:
> 
> A.  Centos5.3, Lustre-2.0.2alpha (1.9.181) with kerberos enabled
> 
> Re:  lustre-module complaining of wrong kernel
> 
> [root at mds02w x86_64]# rpm -ivh 
> lustre-modules-1.9.181-2.6.18_128.1.6_lustre_1.9.181_200905190414.x86_64.rpm
> error: Failed dependencies:
>          kernel = 2.6.18-128.1.6-lustre-1.9.181 is needed by 
> lustre-modules-1.9.181-2.6.18_128.1.6_lustre_1.9.181_200905190414.x86_64

What does "rpm -q --provides kernel-lustre" report?

As you might have noticed, we have transitioned to building our kernel
with the vendor's spec.  In that transition there was a bug, 19163, in
which we didn't provide the "kernel = <version>" that is provided
implicitly with the vendor's native binary package.  I don't recall
whether the fix for that bug landed on HEAD prior to or after 2.0.2a
though.

But now that you point this all out, I think it would be better for
lustre-modules to require "kernel-lustre = <version>" rather than
"kernel = <version>".  That way we know we have the lustre version of
the kernel.

> The rpms built completely from the source.
> [root at mds02w x86_64]# pwd
> /usr/src/redhat/RPMS/x86_64
> [root at mds02w x86_64]# ls
> kernel-2.6.18128.1.6lustre1.9.181-2.x86_64.rpm
               ^
Hrm.  This is strange.  There should be a dash between those.  Is it
really missing in your /usr/src/redhat/RPMS/x86_64 dir?

> If I force --nodeps on lustre-modules install, system crashes as expected.

I wouldn't say it's expected to crash.  All you have so far is an RPM
dependency mismatch.  Forcing that installation should not cause the
system to crash.  What does the crash look like?

> If I proceed with make install instead of make rpms, the resulting 
> system system crashes when I attempt to load lustre module.

> [root at mds02w ~]# Assertion failure in journal_start() at 
> fs/jbd/transaction.c:283: "handle->h_transaction->t_journal == journal"
> ----------- [cut here ] --------- [please bite here ] ---------

This looks like  a real bug, and if the crash you report from scenario
A, after forcing the lustre-modules RPM to install is the same as this
one, this crash is unrelated to the RPM dependencies.

Cheers,
b.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20090523/4559a67a/attachment.pgp>