2:47 p.m.
Hey there,
I tried using oVirt for some time now and like to convert my main Proxmox Cluster (2 Nodes
+ RPI for Quorum) to oVirt. But I obvioulsy need three servers to do HA. Now I am asking
myself: can't I just use the PI as an arbiter node for the Selfhosted-Engine on
gluster? I tried running the playbook already, but it's always failing and even if the
gluster deployment would go through, it would most likely fail aftwards, right?
Before I waste too much energy and time into that, does anyone of you guys use a 2 node
cluster? And no I can't add another node (energy + space + money)!
Also I know with doing that I'm totally on my own etcetera
Thanks!
7:04 a.m.
Yes That was my goal! But isn't it using this:
https://github.com/gluster/gluster-ansible playbook? And it is supposed to work on
debian/ubuntu based oses. So if all prerequesites are met, shouldn't it work?
10:13 a.m.
On Mon, Mar 1, 2021 at 3:05 PM <justin.lamp96(a)gmail.com> wrote:
Yes That was my goal! But isn't it using this:
https://github.com/gluster/gluster-ansible playbook? And it is supposed to work on
debian/ubuntu based oses. So if all prerequesites are met, shouldn't it work?
Never tried using a playbook to setup a Gluster cluster. Can't comment on that.
BTW, which distribution is used on the RPI? are you certain the RPI
and oVirt hosts have the same Gluster versions?
Is the hosted-engine HA then? How about updates, is it retaining the
config?
The Gluster takes care of the storage HA. The two oVirt nodes, if
setup correctly, will handle the oVirt HA.
I do have a PI 4. Wouldn't that be alright for an arbiter node?
Since it's just saving metadata. My nodes are connected via 10Gig btw
In general, In my experience Gluster is very sensitive to network
latency. The RPI, even the RPI4, has a lousy CPU and lousy network.
Given the fact that Gluster depends on the performance of the weakest
link in the chain, the 10GbE links between the nodes may go unused,
while the RPI will start throttling down...
If I were you, I'd get a mATX with a low end i3 or AMD CPU (even
second hand will do), connect a 10GbE link (ebay is your friend...),
and use it as an arbiter. (As an added bonus it can be part of the
oVirt HA cluster and even run a VM or two).
- Gilboa
3:49 p.m.
Thanks!
I'm running 20.04.02 on the pi, which is using glusterfs 7.2. Guess I have to compile
from source then...
That's good to know.
Yeah latency, right... I already checked that and the latency is pretty terrible as you
said. The things is that can't really afford to put in another machine. Might be able
to do a NUC at somepoint at most (space heat energy). The gluster share should also only
be used for the hosted-engine (right now) as I already have a freeNAS box handling the
rest.
7:32 a.m.
On Wed, Feb 24, 2021 at 9:08 AM <justin.lamp96(a)gmail.com> wrote:
I usually create the gluster shares by hand and then deploy oVirt on
each node using the pre-existing gluster shares.
There's no reason it won't work for you.
That said, I never had a situation in which one of the gluster
members isn't also a member of the oVirt cluster.
Oh, which RPI is it? Are you sure its networking is up to the task?
(RPI3 has a terrible NIC, RPI4 isn't much better).
- Gilboa
7:07 a.m.
Is the hosted-engine HA then? How about updates, is it retaining the config?
I do have a PI 4. Wouldn't that be alright for an arbiter node? Since it's just
saving metadata. My nodes are connected via 10Gig btw
Thanks for your answers!
1475
days inactive
1481
days old
6 comments
3 participants
participants (3)
-
Gilboa Davara -
Jayme -
justin.lamp96＠gmail.com