On Tue, Mar 23, 2021 at 6:42 PM Arman Khalatyan <arm2arm(a)gmail.com> wrote:
Hello everybody,
I would like to deploy HCI with our 2 buildings each with 8 compute nodes.
Each host has a mirrored OS disks and 1 slot for the SSD. So I will use SSD for the
glusterfs.
my question is what is the best type of the glusterfs volume?
I can leave with 8way mirror but what happened if the connection between buildings will
go down?
where will my ovirt-engine start?
I'd start by asking a simple question: Is there any solid reason to do
an 8-way cross building gluster cluster?
Beyond the unbelievable waste of space (8 way mirror) that can only be
solved by creating a complex setup of alternating normal and arbiter
gluster bricks, you are also literally opening yourself to countless
gluster stability and performance problems due to semi-reliable
cross-building network link(s).
In-short, unless you really, really, really need an 8 way mirror,
simply don't do it.
If I were you, I'd go with the much simpler and better solution of
creating two separate clusters, one in every building, making sure
each VM has an active version in one cluster and a recent backup in
the other (and vice versa).
- Gilboa