[Lustre-discuss] Problems going from source to RPMs

[David Bogen]

All:

In an effort to test the patches against lustre that might improve NFS  
performance I've been attempting to build from source on a RHEL5 system.

I got the lustre (1.6.4.3) and kernel (linux-2.6.18-53.1.13) source  
from Sun (that's now a really, um, less-than-ideal process) and  
unpacked both.

Patched the lustre source with patches 14006, 14007, and 14591.

Followed the Lustre manual <http://preview.tinyurl.com/6zlgt6> and got  
a bit confused by step 1 in 4.3.1.3 since there doesn't seem much  
point in selecting a kernel config and then never using it (section  
4.3.2 has you linking in a completely different .config), so decided  
to copy in my chosen kernel config (kernel-2.6.18-2.6-rhel5-i686- 
smp.config) and run with it.

Linked in the proper patches and series (2.6-rhel5.series) directories.

Used quilt (0.46) to apply the patches

Ran the four make commands in the kernel source directory.

Switched to the top-level lustre directory (lustre-1.6.4.3).  Ran the  
following configure command:

./configure --enable-quota --with-linux=/mnt/space/user1/ 
linux-2.6.18-53.1.13

Other than this error:

checking linux/quota.h presence... yes
configure: WARNING: linux/quota.h: present but cannot be compiled
configure: WARNING: linux/quota.h:     check for missing prerequisite  
headers?
configure: WARNING: linux/quota.h: see the Autoconf documentation
configure: WARNING: linux/quota.h:     section "Present But Cannot Be  
Compiled"
configure: WARNING: linux/quota.h: proceeding with the preprocessor's  
result
configure: WARNING: linux/quota.h: in the future, the compiler will  
take precedence
configure: WARNING:     ## -------------------------------------------  
##
configure: WARNING:     ## Report this to https://bugzilla.lustre.org/  
##
configure: WARNING:     ## -------------------------------------------  
##
checking for linux/quota.h... yes
checking xtio.h usability... no

which already appears to be in bugzilla, the configure command  
finishes without a problem.

Now I'm ready to run make rpms, so I have a go with that.

That produces a fair amount of output (available on request), but ends  
with this:

/bin/sh /mnt/space/user1/lustre-1.6.4.3/missing --run tar chof -  
lustre-1.6.4.3 | GZIP=--best gzip -c >lustre-1.6.4.3.tar.gz
{ test ! -d lustre-1.6.4.3 || { find lustre-1.6.4.3 -type d ! -perm  
-200 -exec chmod u+w {} ';' && rm -fr lustre-1.6.4.3; }; }
rpmbuild -ta lustre-1.6.4.3.tar.gz
error: File /mnt/space/user1/lustre-1.6.4.3/lustre- 
ldiskfs-3.0.4.tar.gz: No such file or directory
make[1]: *** [rpms-real] Error 1
make[1]: Leaving directory `/mnt/space/user1/lustre-1.6.4.3'
make: *** [rpms] Error 2

If I link in the file it is looking for (but in the wrong place):

ln -s /mnt/space/user1/lustre-1.6.4.3/ldiskfs/lustre- 
ldiskfs-3.0.4.tar.gz \
       /mnt/space/user1/lustre-1.6.4.3/lustre-ldiskfs-3.0.4.tar.gz

and run make rpms again, I can get a bit farther.  Again, a fair  
amount of output is available upon request, but this is the end of the  
run:

Processing files: lustre-ldiskfs- 
debuginfo-3.0.4-2.6.18_prep_200804161020
Checking for unpackaged file(s): /usr/lib/rpm/check-files /var/tmp/ 
lustre-ldiskfs-3.0.4-root
Wrote: /var/tmp/rpm/SRPMS/lustre- 
ldiskfs-3.0.4-2.6.18_prep_200804161020.src.rpm
Wrote: /var/tmp/rpm/RPMS/i386/lustre- 
ldiskfs-3.0.4-2.6.18_prep_200804161020.i386.rpm
Wrote: /var/tmp/rpm/RPMS/i386/lustre-ldiskfs- 
debuginfo-3.0.4-2.6.18_prep_200804161020.i386.rpm
Executing(%clean): /bin/sh -e /var/tmp/rpm-tmp.4196
+ umask 022
+ cd /var/tmp/rpm/BUILD
+ cd lustre-ldiskfs-3.0.4
+ rm -rf /var/tmp/lustre-ldiskfs-3.0.4-root
+ exit 0
make[1]: Leaving directory `/mnt/space/user1/lustre-1.6.4.3'

So that produced the lustre-ldiskfs RPMS, and that's it.  Running make  
rpms-real doesn't change anything as I end up with just the lustre- 
ldiskfs RPM.

If I simply make the source, without producing the RPM, it seems to  
work just fine, so this seems like a packaging problem, more than a  
problem with the source itself.

What am I doing wrong?  I'm open to suggestions that run the gamut  
from the obvious to the obscure.

David

--
David Bogen :: (608) 263-0168
Unix SysAdmin :: IceCube Project
david.bogen at icecube.wisc.edu