[Lustre-discuss] SLES11 'make rpms' dies

[Joe Landman]

Brian J. Murrell wrote:
> On Wed, 2010-03-24 at 13:31 -0400, Joe Landman wrote: 

[...]

> which, as you can see, is older than the kernel you are trying to build
> for.  Maybe you know that.  I thought I would just point it out for
> completeness in any case.

I was and am aware of this.  And this is why I had to start working on 
the build.

This kernel (the .39-0.3.1) has several bugs that have been resolved in 
.42 and .45.  Our use case does trigger these bugs, and we can solve our 
problem by using the .42 or .45 kernels.

Anyone with an NFS system will very likely want to migrate to the .42 or 
.45 kernels due to these issues.  Hence we either have to wait for the 
"official" builds to catch up to the released bits, or perform builds 
ourselves.

[...]

n of function ‘jbd2_journal_callback_set’
> 
> Could be due to the newer kernel you are attempting to build for.

I wouldn't think they were that much different.

[...]

> If the problem is simply too new a kernel that requires some porting on
> Lustre's part, perhaps, but not likely.  If you can find a bug tracking
> the effort to support a newer SLES11 kernel, it will tell you if the
> patches have landed (i.e. are in git) yet.
> 
> A quick search of BZ yielded bug 21862 which appears to be opened to
> track the update to 2.6.27.42 although it appears no work has yet been
> done.

I did see this, and this is unfortunate.  The other possibility is to 
backport the needed fixes into the .39 kernel.  I will investigate this, 
as it doesn't appear that the .42 or .45 will build correctly.


-- 
Joe Landman
landman at scalableinformatics.com