[Lustre-discuss] Lustre ZFS Snapshot

[Indivar Nair]

Thanks Andreas.
I have a couple of queries on your suggestion.
Please find them inline *(blue color)* in the mail below.

Regards,


Indivar Nair

On Mon, Jun 30, 2014 at 2:12 PM, Dilger, Andreas <andreas.dilger at intel.com>
wrote:

> On 2014/06/29, 1:24 PM, "Indivar Nair" <indivar.nair at techterra.in<mailto:
> indivar.nair at techterra.in>> wrote:
> Referring to https://jira.hpdd.intel.com/browse/LUDOC-161
>
> The idea is to be able to backup and restore individual files /
> directories to another non-Lustre storage OR tape drive...
>
> Well, the idea of LUDOC-161 is not to backup and restore individual files,
> but backup and restore of the whole MDT or OST.
>
> So, can one mount the MDT and OST snapshots to form a parallel Lustre
> Volume (on the same cluster) and take a point-in-time backup / rsync of the
> complete Lustre filesystem?
>
> In theory yes, but this isn't yet supported.  The problem is that the
> Lustre filesystem name is the same between the snapshots, so the snapshot
> filesystem cannot be mounted on the same clients as the original filesystem.
>
> The device-level backup/zfs send/recv, especially of the MDT, is mostly
> intended for disaster recovery, though with ZFS it may also be practical to
> use it in the manner you are proposing.
>
> If not, can one mount a copy of MDT and OST snapshots on another Lustre
> cluster to recreate the complete Lustre Volume?
>
> It would be OK to mount it on separate clients.
>

The hostname and IP of the MGS, MDS and OSS nodes on the other cluster
would be different. Would this be ok?

Also, On the backup cluster, can I have multiple OSTs (dataset) in a single
zpool?
For example, I may have 4 zpools with 1 OST (dataset) each on the
production cluster, but on the backup cluster I may have 1 zpool with 4
OSTs (dataset). This will allow me to have a backup cluster using lesser
number of cheaper 4TB SATA disks instead of the 1TB SAS disks on the
production cluster.

>
> If you want to do file-level restore it may be better to do file-level
> backup from the mounted filesystem.
>
> But his won't give a point-in-time backup, right?


> Cheers, Andreas
> --
> Andreas Dilger
> Lustre Software Architect
> Intel High Performance Data Division
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lustre.org/pipermail/lustre-discuss-lustre.org/attachments/20140702/a6440a12/attachment.htm>