Finally I got some time playing with GlusterFS and libvirt.
I’m not going to step into GlusterFS setup as it’s really easy and straight forward, I did CentOS kickstart (non-attendance) installation as well since it may help me in future deployment. Note that I assume you have glusterd running the host node, other all localhost below should be replaced by the hostname that is running glusterd. Continue reading »
Learnt from the mailing list that I should not export mounting point as a brick directly, instead I should create a sub-directory under the mounting point and take it as brick, this will make life much easier later on whenever you have to do migration or re-organization (like changing replica).
Developers of GlusterFS are thinking of retiring replace-brick, from the mailing list I can see they have good point to do so (especially existing codes for replace-brick is too bad to be improved :P), however, as a (sort of) ops guy, I still prefer they can keep it since it makes ops’ life much more easier, but obviously they don’t quite agree with all requests/challenges to keep it. Anyway, it seems life will not end, it will just become a little bit troublesome, and if this is the trade-off for stability and/or performance, I’m fine with it.
Here’s my plan.
I have 4 servers named them as gfs[11-14], each machine has 4 disks, with sda allocated for root, swap, etc, and sdb-sdd as data in XFS.
I’m going to run distributed-replicated volume with replication 2, and I want to keep highest availability, as much as I can. So the layout will be something like this: Continue reading »