[Lustre-discuss] Announce: e2fsprogs 1.41.5.sun2

[Jim Garlick]

On Fri, May 15, 2009 at 05:19:55PM +0200, Ramiro Alba Queipo wrote:
> Hi Guy,
> 
> On Fri, 2009-05-15 at 09:11 +0100, Guy Coates wrote:
> 
> > Hi,
> > 
> > Is the source available for the sun patches? The I can find the pristine
> > upstream sources, but the lustre download site only has the
> > e2fsprogs-1.41.5.sun1-patches tarball.
> > 
> > Cheers,
> 
> I've downloaded
> http:// downloads.lustre.org/public/tools/e2fsprogs/1.41.5.sun2/e2fsprogs-1.41.5.sun2-0redhat.src.rpm
> and then made a Debian package for Ubuntu 8.04 adm64 (Thanks for your
> documentation at lutre-discuss-list with the subject 'lustre + debian').
> This source is already patched.
> 
> It seems to work for my installation:
> 
> I am using Lustre 1.6.7 from
> http:// packages.debian.org/search?keywords=lustre&searchon=names&suite=testing&section=all
> and OFED-1.3.1 compiled for kernel 2.6.22.14.
> The OSTs are 4 DDN 9900 tiers with Infiniband ports
> 
> I had some problems when using e2fsprogs-1.41.5.sun1-patches tarball and
> trying to activate 'mmp' by using 'tune2fs -O mmp <device file>' until I
> discovered that formating OSTs with the --failnode parameter, the MMP
> feature was automatically enabled. Did you have the same problem?
> Now, with the latest patches I deactivated (tune2fs -O ^mmp <device
> file>) and activated back with no apparent problems, except much more
> delay when mounting the first time.
> 
> Nowadays I am not yet in production stage

Note the latest lustre tune2fs can more clear the MMP state without
mucking with the superblock:

  tune2fs -f -Eclear-mmp <device>

No need to deactivate and reactivate if you get stuck after an aborted
fsck or whatever.

Jim