[Lustre-discuss] Setting up lustre for backup

Andreas Dilger adilger at sun.com
Mon Jul 13 10:55:52 PDT 2009


On Jul 13, 2009  12:07 +0300, Onane wrote:
> In my configuration, OST1 has 80GB & OST2 has 80GB. If I gave them same
> fsname & mgsnode, I would access 160GB lustre filesystem on client. I want
> to use these two volumes like raid 1. That is, how can I configure lustre
> filesystem so that OST2 back up OST1 (say. every hour) ?

One possible way to do this, as Brian suggested, is to use DRBD and do the
mirroring below the level that Lustre sees, at the block device level.

Another possibility is to create a top-level directory in the filesystem
with "lfs setstripe -c 1 -i 0 /mnt/lustre/primary" and create all of the
files in that directory (it will inherit the striping that forces files
to be created on OST index 0), and then rsync all of the files to a
separate directory with "lfs setstripe -c 1 -i 1 /mnt/lustre/backup".

That would work on 1.6.x for 2 OSTs, but the file creation is not strictly
contained to the OSTs that are specified by default striping.

In 1.8.x it would be more flexible to create an OST pool that contains
one or more "primary" OSTs, and then a second OST pool that contains one
or more "backup" OSTs.  The file creation will be strictly enforced to
be within the respective pools.

It isn't an ideal solution, and I _wish_ we had RAID-1 at the OST level,
but that still hasn't made it to the top of the "we need this for Lustre"
feature list.

> On Fri, Jul 10, 2009 at 8:07 PM, Andreas Dilger <adilger at sun.com> wrote:
> > On Jul 10, 2009  15:34 +0300, Onane wrote:
> > > I installed lustre on three machines (mc0, mc1, mc2)
> > >
> > > I configured lustre,
> > > on mc0 as MDT & MGS pair AND OST1,
> > > on mc1 as OST2
> > > on mc2 as client
> > >
> > > I want OST2 to be a mirror of OST1. That is, if I copy a file to OST1, it
> > > will also be copied to OST2. OST2 should be a backup for OST1.
> > >
> > > Can I do this configuration using default lustre options ? Can it be
> > > accomplished by --failnode directive ? If so, how can I do it
> >
> > No, this isn't yet supported by Lustre.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.




More information about the lustre-discuss mailing list