Our nodes could use a kernel update to work around some issues we''re having, and of course we''d like to get Lustre going. Currently attempting to use lustre-1.4.6.4. Unfortunately the set of Lustre patches doesn''t seem to be updated to work with the latest kernels. I tried quilt-ing the set of patches for 2.6.12-vanilla into 2.6.17.8, in hopes they would work anyway, but it didn''t work out very well [1]. Does anybody have (even perhaps experimental -- I''m just testing with a node or two for now) updated patches that might apply to later kernels? Based on some *very* quick ChangeLog-skimming, we might be able to go with 2.6.14. Has anyone tried applying the 2.6.12 patches to 2.6.14? -tom [1] tfogal@n130:/usr/src/linux-2.6.17.8> quilt push -av Applying lustre_version patching file include/linux/lustre_version.h Applying vfs_intent-2.6.12 patching file fs/exec.c Hunk #1 FAILED at 122. Hunk #2 FAILED at 137. Hunk #3 FAILED at 493. Hunk #4 FAILED at 508. 4 out of 4 hunks FAILED -- rejects in file fs/exec.c patching file fs/namei.c Hunk #1 succeeded at 337 (offset 36 lines). Hunk #2 FAILED at 439. Hunk #3 FAILED at 477. Hunk #4 succeeded at 526 with fuzz 1 (offset 48 lines). Hunk #5 succeeded at 538 (offset 48 lines). <snip -- more of the same; some FAILED, some succeeded>
From: tom fogal <tfogal@apollo.sr.unh.edu>
Date: Wed, 09 Aug 2006 14:25:27 -0400
Does anybody have (even perhaps experimental -- I''m just testing
with a node or
two for now) updated patches that might apply to later kernels?
I have a not-too-badly-modified lustre 1.6b4 tree which I''ve
demonstrated to
my own satisfaction on vanilla (kernel.org) 2.6.15. If you would find it
useful I could generate a unified diff of my tree against the stock one for
you.
Based on some
*very* quick ChangeLog-skimming, we might be able to go with 2.6.14. Has
anyone tried applying the 2.6.12 patches to 2.6.14?
Based on my experience, I''d say you should expect to have trouble with
that
approach. Things in the kernel diverged fairly substantially after 2.6.13,
such that the 2.6.12 patches start failing in all kinds of entertaining ways.
To generate my lustre-for-2.6.15-vanilla tree, I started with the RHEL patches
and essentially filtered out the redhat-ness. That approach is not without
problems either, but seemed to require less hand-hacking than bringing the
2.6.12 patches forward.
Please let me know if you''d like me to generate that patch set for you.
I, and probably others, would also be very interested in this. Perhaps you could pop it up on a webpage or put it in bugzilla.clusterfs.com? Maybe Clusterfs will take notice and add it to their patch set... Thanks, Brent Nelson Director of Computing Dept. of Physics University of Florida On Wed, 9 Aug 2006, jrd@jrd.org wrote:> From: tom fogal <tfogal@apollo.sr.unh.edu> > Date: Wed, 09 Aug 2006 14:25:27 -0400 > > Does anybody have (even perhaps experimental -- I''m just testing with a node or > two for now) updated patches that might apply to later kernels? > > I have a not-too-badly-modified lustre 1.6b4 tree which I''ve demonstrated to > my own satisfaction on vanilla (kernel.org) 2.6.15. If you would find it > useful I could generate a unified diff of my tree against the stock one for > you. > > Based on some > *very* quick ChangeLog-skimming, we might be able to go with 2.6.14. Has > anyone tried applying the 2.6.12 patches to 2.6.14? > > Based on my experience, I''d say you should expect to have trouble with that > approach. Things in the kernel diverged fairly substantially after 2.6.13, > such that the 2.6.12 patches start failing in all kinds of entertaining ways. > > To generate my lustre-for-2.6.15-vanilla tree, I started with the RHEL patches > and essentially filtered out the redhat-ness. That approach is not without > problems either, but seemed to require less hand-hacking than bringing the > 2.6.12 patches forward. > > Please let me know if you''d like me to generate that patch set for you. > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss@clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss > >
Kernel patches can always be added - but the problem is that gcc4 is kicking up more problems for which we haven''t had enough time. - Peter - Brent A Nelson wrote:> I, and probably others, would also be very interested in this. > Perhaps you could pop it up on a webpage or put it in > bugzilla.clusterfs.com? Maybe Clusterfs will take notice and add it to > their patch set... > > Thanks, > > Brent Nelson > Director of Computing > Dept. of Physics > University of Florida > > On Wed, 9 Aug 2006, jrd@jrd.org wrote: > >> From: tom fogal <tfogal@apollo.sr.unh.edu> >> Date: Wed, 09 Aug 2006 14:25:27 -0400 >> >> Does anybody have (even perhaps experimental -- I''m just testing >> with a node or >> two for now) updated patches that might apply to later kernels? >> >> I have a not-too-badly-modified lustre 1.6b4 tree which I''ve >> demonstrated to >> my own satisfaction on vanilla (kernel.org) 2.6.15. If you would >> find it >> useful I could generate a unified diff of my tree against the stock >> one for >> you. >> >> Based on some >> *very* quick ChangeLog-skimming, we might be able to go with >> 2.6.14. Has >> anyone tried applying the 2.6.12 patches to 2.6.14? >> >> Based on my experience, I''d say you should expect to have trouble >> with that >> approach. Things in the kernel diverged fairly substantially after >> 2.6.13, >> such that the 2.6.12 patches start failing in all kinds of >> entertaining ways. >> >> To generate my lustre-for-2.6.15-vanilla tree, I started with the >> RHEL patches >> and essentially filtered out the redhat-ness. That approach is not >> without >> problems either, but seemed to require less hand-hacking than >> bringing the >> 2.6.12 patches forward. >> >> Please let me know if you''d like me to generate that patch set for you. >> >> _______________________________________________ >> Lustre-discuss mailing list >> Lustre-discuss@clusterfs.com >> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss >> >> > > _______________________________________________ > Lustre-discuss mailing list > Lustre-discuss@clusterfs.com > https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
From: Brent A Nelson <brent@phys.ufl.edu>
Date: Wed, 9 Aug 2006 15:44:15 -0400 (EDT)
I, and probably others, would also be very interested in this. Perhaps
you could pop it up on a webpage or put it in bugzilla.clusterfs.com?
Ok, give me a little time to construct something that I have some
level of confidence in. The total amount of patch isn''t all that
large, but it''s still in a bit of a state of flux on my end. Once I
get it stabilized I''ll post it someplace or stuff it in cfs''s
bugzilla
or wiki or something if they''d like to have it.
tom fogal <tfogal@apollo.sr.unh.edu> writes:> Our nodes could use a kernel update to work around some issues we''re having, > and of course we''d like to get Lustre going. Currently attempting to use > lustre-1.4.6.4.I have working patches for 2.6.15 and 2.6.16 and untested ones for 2.6.17/2.6.18-rc3. 2.6.17 introduces a change in the ext3 code. In the next few days, as time permits, I will upload the patch sets to the lustre project on alioth where debian packaging of lustre will be coordinated: http://alioth.debian.org/tracker/?atid=410831&group_id=30342&func=browse MfG Goswin
On Aug 09, 2006 16:50 -0400, John R. Dunning wrote:> From: Brent A Nelson <brent@phys.ufl.edu> > Date: Wed, 9 Aug 2006 15:44:15 -0400 (EDT) > > I, and probably others, would also be very interested in this. Perhaps > you could pop it up on a webpage or put it in bugzilla.clusterfs.com? > > Ok, give me a little time to construct something that I have some > level of confidence in. The total amount of patch isn''t all that > large, but it''s still in a bit of a state of flux on my end. Once I > get it stabilized I''ll post it someplace or stuff it in cfs''s bugzilla > or wiki or something if they''d like to have it.There is https://mail.clusterfs.com/wikis/lustre/LustreStatusonLinux26 wiki which is probably the best place for user-contributed kernel patches. Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.
From: "John R. Dunning" <jrd@jrd.org>
Date: Wed, 9 Aug 2006 16:50:19 -0400
From: Brent A Nelson <brent@phys.ufl.edu>
Date: Wed, 9 Aug 2006 15:44:15 -0400 (EDT)
I, and probably others, would also be very interested in this. Perhaps
you could pop it up on a webpage or put it in bugzilla.clusterfs.com?
Ok, give me a little time to construct something that I have some
level of confidence in. The total amount of patch isn''t all that
large, but it''s still in a bit of a state of flux on my end. Once
I
get it stabilized I''ll post it someplace or stuff it in
cfs''s bugzilla
or wiki or something if they''d like to have it.
I''ve uploaded a tarball with the latest spin of my patchset for 2.6.15
kernels. See https://mail.clusterfs.com/wikis/lustre/LustreStatusonLinux26
Please let me know if it''s useful, or if I blew something and it
doesn''t work. And if anybody finds bugs in it, please let me know,
perhaps
you''ll save me some effort debugging :-} TIA...