[lustre-discuss] Backup software for Lustre

Dilger, Andreas andreas.dilger at intel.com
Tue Mar 14 14:13:39 PDT 2017

To reply to this old thread, there are two different kinds of Lustre backup solutions:
- file level backups that traverse the client POSIX filesystem, for which any number of
  commercial solutions exist.  Making these solutions "capable of saving Lustre metadata"
  boils down to two simple things - save the "lustre.lov" xattr during backup (at a minimum,
  other xattrs also should be backed up), and then using mknod(2) + setxattr() to restore
  the "lustre.lov" xattr before opening the file and restoring the data.

- device level backups (e.g. "dd" for ldiskfs, and "zfs send/recv" for ZFS).

Using the file level backups allows backup/restore of subsets of the filesystem, since many
HPC sites have Lustre filesystems that are too large to backup completely.  I typically do
not recommend to use device-level backups for the OSTs, unless doing an OST hardware migration,
and even then it is probably less disruptive to do Lustre-level file migration off the OST
before swapping it out.

Whether file level backups are used or not, I would recommend sites always make periodic
device level backups of the MDT(s).  The amount of space needed for an MDT backup is small
compared to the rest of the filesystem (e.g. a few TB at most), and can avoid the need for
a full filesystem restore (e.g. multi-PB of data, if a full backup exists at all) even
though all the data is still available on the OSTs.

The MDT device-level backup can use relatively slow SATA drives, since they will mostly be
used for linear writes (or occasionally linear reads for restore), so a few multi-TB SATA III
drives is sufficient for storing a rotating set of MDT device backups.  At 150MB/s for even
a single SATA drive, this is about 2h/TB, which is reasonable to do once a week (or more often
for smaller MDTs).

While using an LVM snapshot of the ldiskfs MDT for the backup source is desirable for consistency
reasons, having even an MDT backup from a mounted and in-use MDT is better than nothing at
all when a problem is hit, since e2fsck can repair the in-use inconsistencies fairly easily,
and Lustre can deal with inconsistencies between the MDT and OST reasonably (at most returning
an -ENOENT error to the client for files that were deleted).

Cheers, Andreas

On Feb 7, 2017, at 12:32, Andrew Holway <andrew.holway at gmail.com> wrote:
> Would it be difficult to suspend IO and snapshot all the nodes (assuming ZFS). Could you be sure that your MDS and OSS are synchronised?
> On 7 February 2017 at 19:52, Mike Selway <mselway at cray.com> wrote:
>> Hello Brett,
>>                Actually, looking for someone who uses a commercialized approach (that retains user metadata and Lustre extended metadata) and not specifically the manual approaches of Chapter 17.
>> Thanks!
>> Mike
>> Mike Selway | Sr. Tiered Storage Architect | Cray Inc.
>> Work +1-301-332-4116 | mselway at cray.com
>> 146 Castlemaine Ct,   Castle Rock,  CO  80104 | www.cray.com
>>> From: Brett Lee [mailto:brettlee.lustre at gmail.com] 
>>> Sent: Monday, February 06, 2017 11:45 AM
>>> To: Mike Selway <mselway at cray.com>
>>> Cc: lustre-discuss at lists.lustre.org
>>> Subject: Re: [lustre-discuss] Backup software for Lustre
>>> Hey Mike,
>>> "Chapter 17" and
>>> http://www.intel.com/content/www/us/en/lustre/backup-and-restore-training.html
>>> both contain methods to backup & restore the entire Lustre file system.
>>> Are you looking for a solution that backs up only the (user) data files and their associated metadata (e.g. xattrs)?
>>> Brett
>>> --
>>> Protect Yourself From Cybercrime
>>> PDS Software Solutions LLC
>>> https://www.TrustPDS.com
>>>> On Mon, Feb 6, 2017 at 11:12 AM, Mike Selway <mselway at cray.com> wrote:
>>>> Hello,
>>>>          Anyone aware of and/or using a Backup software package to protect their LFS environment (not referring to the tools/scripts suggested in Chapter 17). 
>>>> Regards,
>>>> Mike

Cheers, Andreas
Andreas Dilger
Lustre Principal Architect
Intel Corporation

More information about the lustre-discuss mailing list