[Gluster-users] restoring a volume: best practise for backup and restore

Paul Bickerstaff [DATACOM] PaulBi at datacom.co.nz
Mon Dec 5 07:16:03 UTC 2016


A search reveals little definitive in the way of best practise for backing up and restoring glusterfs volumes.

I'm particularly interested in best practise for replicated volumes. Is there any recommended way that is both efficient and proven?

It seems that geo-replication, snapshotting and glusterfind all provide options. But what about a simple backup and restore of a brick using standard backup tools?

The question needs to be asked because our naïve use of a backup tool has created problems. These may be specific to our environment but our environment is unlikely to be ultra-special. We are using gluster 3.7 on CentOS 7.2 and a standard package install.

An issue may be that we do not have native fuse installed. It seems that this is a recognised feature with Red Hat: https://bugzilla.redhat.com/show_bug.cgi?id=947830 .

Use though only of glusterfs-fuse seems to result in replication only when the volume is accessed via gluster mechanisms. E.g. writing to an NFS mounted volume does result in the file being replicated across all replica nodes.

However, creating a file directly in a brick (as would be normal in a backup restore) does not result in replication as claimed in http://blog.gluster.org/category/volumes/ where it is supposedly demonstrated using touch. We also find that restarting the volume does not result in replication. Our volumes are indeed of type cluster/replicate. My guess is that the missing fuse package is the reason why.

But it seems from tools like bitrot (which can seemingly detect and scrub data resulting from '"backend" tinkering of bricks') that restoring to a brick directly may not be best practise.

Nevertheless, we require a means to restore a replicated volume from a backup copy and we need to be able to go back to a point in time and not rely on the last saved state (as in geo-replication). We also want to be able to restore single volumes and leave others as is. And we would like to avoid having to go through an intermediate backup/restore server where the volumes are mounted.

I am keen to know what others have found works well.

Thanks
Paul
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20161205/61dc3b27/attachment.html>


More information about the Gluster-users mailing list