I'm new to Ceph, and considering using it to store a bunch of static
files in the RADOS Gateway. My files are all versioned, so we never
modify files. We only add new files, and delete unused files.
I'm trying to figure out how to back everything up, to protect against
administrative and application errors.
I'm thinking about building one Ceph cluster that spans my primary and
backup datacenters, with CRUSH rules that would store 2 replicas in each
datacenter. I want to use BtrFS snapshots, like
http://blog.rot13.org/2010/02/using_btrfs_snapshots_for_incremental_backup.html,
but automated and with cleanup. I'm doing something similar now, on my
NFS servers with ZFS and a tool called zfs-snapshot-mgmt.
I read that only XFS is recommended for production clusters, since BtrFS
itself is still beta. Any idea how long until BtrFS is usable in
production?
I'd prefer to run Ceph on ZFS, but I see there are some outstanding
issues in tracker. Is anybody doing Ceph on ZFS in production? ZFS
itself seems to be father along than BtrFS. Are there plans to make ZFS
a first class supported filesystem for Ceph?
Assuming that BtrFS and ZFS are not recommended for production, I'm
thinking about XFS in the primary datacenter, and BtrFS + snapshots in
the backup datacenter. Once BtrFS or ZFS is production ready, I'd
slowly migrate all partitions off XFS.
Once the backups are made, using them is a bit tricky.
In the event of an operator or code error, I would mount the correct
BtrFS snapshot on all nodes in the backup datacenter, someplace like
/var/lib/ceph.restore/. Then I'd make a copy of ceph.conf, and start
building a temporary cluster that runs on a non-standard port, made up
of only the backup datacenter machines. The normal cluster would stay
up and running. Once the temporary cluster is up, I'd manually restore
the RADOS Gateway objects that needed to be restored.
If there was ever a full cluster problem, like I did something stupid
like rados rmpool metadata. I'd shut down the whole cluster, and revert
all of the BtrFS partitions to the last known good snapshot, and
re-format all of the XFS partitions. Start the cluster up again, and
let Ceph replicate everything back to freshly formatted partitions. I'd
lose recent data, but it's better than losing all of the data.
Obviously, both of these scenarios would need a lot of testing and many
practice runs before they're viable. Has anybody tried this before? If
not, do you see any problems with the theory?
Thanks for the help.
--
*Craig Lewis*
Senior Systems Engineer
Office +1.714.602.1309
Email cle...@centraldesktop.com <mailto:cle...@centraldesktop.com>
*Central Desktop. Work together in ways you never thought possible.*
Connect with us Website <http://www.centraldesktop.com/> | Twitter
<http://www.twitter.com/centraldesktop> | Facebook
<http://www.facebook.com/CentralDesktop> | LinkedIn
<http://www.linkedin.com/groups?gid=147417> | Blog
<http://cdblog.centraldesktop.com/>
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
