Hello all,
We are working on an upgrade plan that touches a number of things, one of them
being our Gluster setup. I wanted to throw some of this out and see if anyone
sees any glaring problems with it. One of our constraints is that our current
Gluster installation is production and cannot fully go down. (Remounts, etc. are
fine.)
We currently are running 3.6.2, and have approximately 48TB (available) across
10 volumes on six hosts. Most volumes are small distribute-replicate 2x2; one is
about half of the available storage and is replicate, 1x2. We use Gluster
clients and NFS on some volumes. The storage network is all 10G, the rest of the
hardware is relatively young Dell commodity kit.
We will be adding a couple NAS devices (in different locations) with about 100TB
(raw) additional each. We geo-replicate for DR. We?ll be physically installing
those first, so that there?s slack space for us to rearrange/reconfigure as
needed during this process. This will run ZFS on ubuntu.
The plan, in loose terms, will be to add machines (an arbiter, NASes and
machines to front the NASes) to the cluster, restructure the volumes/mountpoints
to be somewhat more aligned with what we currently do, upgrade/perhaps switch
distros, upgrade Gluster, and revamp monitoring. Not in that order.
Questions:
- Is anyone currently running Gluster on Debian or Ubuntu in production? We
would prefer to get off RHEL-flavored hosts (currently Centos 7), thus I?m
hoping I?m wrong, but I?ve seen little evidence that trying this for production
would be a safe endeavor.[1]
- Alongside of this (meaning, as part of the overall plan, but not necessarily
as an integral part of upgrading Gluster), we?re installing oVirt to take the
place of multiple xen hosts. I know the two projects have been adding all sorts
of pretty integration between the two projects, but am unclear on what exactly
the payoff is for managing Gluster via oVirt for people happy with command line
management. Anyone have any thoughts there?
- Since these are all 2x2 or 1x2, I know we need to add at least arbiter nodes.
Is there anything amiss with the idea of adding a single machine to arbitrate
all of those?
In general, does anyone see any problems with this plan?
Thanks in advance for any thoughts.
-j
[1] I?m aware that oVirt is also RHEL-oriented.