[ClusterLabs] Hyperconverged 3 Node Cluster

[Ken Gaillot]

Hi,

The "Clusters from Scratch" document is a good place to start:

https://clusterlabs.org/pacemaker/doc/2.1/Clusters_from_Scratch/singlehtml/

It gives a walk-through of creating a two-node cluster with DRBD shared
storage and a web server, but the intent is simply to get familiar with
the cluster tools. Once you're comfortable with that, it should be
reasonably straightforward to do what you want.

On Tue, 2023-05-16 at 22:18 +1000, adam-clusterlabs at blueraccoon.com.au
wrote:
> Hi ClusterLabs Community,
> 
> I am rebuilding my homelab and looking to further my learning with
> some 
> new technologies being KVM and pacemaker. Is anyone able to offer
> some 
> general guidance on how I can setup a hyperconverged 3 node cluster
> with 
> KVM without shared storage. Based on my research the best stack for
> my 
> use case appears to be;
> 
> * Rocky 9 - I prefer rpm based distributions
> 
> * KVM for the hypervisor
> 
> * ZFS -  To be able to snapshot my VM's regularly and replicate 
> snapshots to an external ZFS pool
> 
> * Gluster (on top of  ZFS) 3 node replica - Used to sync the storage 
> across all 3 nodes
> 
> 
> I assume the missing piece of the puzzle to enable high availability
> of 
> the virtual machines is pacemaker and corosync?
> 
> Will pacemaker be able to automatically move virtual machines based
> on 
> resource (CPU and Memory) across my 3 nodes?
> 
> How will I be able to migrate virtual machines to each of nodes?
> 
>      Preference would be to be able to use either Cockpit or the
> command 
> line based on what I am doing.
> 
> 
> I am also aware of oVirt that could achieve my goals but the project 
> doesnt seem all that active and I am looking to simplify my solution
> for 
> reliability and ability to troubleshoot.
> 
> 
> Appreciate any guidance to point me the right direction.
> 
> Thanks,
> 
> Adam
-- 
Ken Gaillot <kgaillot at redhat.com>