Hello!

On Fri, Apr 13, 2018 at 7:02 PM, Joe DiTommasso <jdito@domeyard.com> wrote:

Hi, I'm currently running an out-of-date (3.6) 3-node oVirt cluster, using NFS storage. I'd like to upgrade to the latest release and move to a hyperconverged setup, but I've only got the three hosts to play with. I'm currently thinking of pulling one node from my current cluster, rebuilding it to start a new cluster, and then migrating VMs/physical hosts over to the new cluster. Is that something that would seem feasible? I've read the documentation on setting up a hyperconverged cluster, but didn't see anything about adding new hosts after cluster deployment.Thanks!

Well, first of all - it is very tricky and risky. We have a good reason to require 3 hosts in HC cluster - data redundancy and service reliability. If you have luxury of turning everything down and upgrading using backup/restore or something similar - please, go this way.

But in case you really really really need to upgrade your deployment in your way, you can try following:

* Backup first :-)

* Upgrade your OS to the maximum version supported by 3.6 and any of the next ovirt releases. You can also try to upgrade ovirt first.

* Remove one of the nodes from your cluster

* Manually install gluster servers on all 3 nodes and create a replicated volumes for your future engine and vms

* Deploy HE on the free node, using existing gluster volumes

* Migrate VMs to that engine.

* One by one remove hosts from your old cluster and add them to your new cluster

* When you will finish that, backup again :) and try to continue your upgrade.

The whole idea of the process, described above, is to provide you with gluster replica 3 volume from the very beginning. Technically you can use a single brick distributed volume to install new engine and migrate VMs to your new cluster, but problems with that single brick distributed volume will affect all your VMs.

Another one option for you is to make replicated volume in a single host. If you have enough space for keeping data thee times, you can:

* Backup (!)

* Remove one of the hosts from service, upgrade it

* And after that create replica 3 volumes for the engine and VMs, using 6 _local_ bricks

* Migrate VMs from one of your hosts

* Upgrade that second hosts, add it to your new cluster and manually migrate two bricks (one brick per volume) from the first host to the second host. 

* Repeat for remaining host

Both approaches are VERY risky and highly THEORETICAL. I don't think anyone ever did that, so think twice before doing that. Following any of those scenario requires you to deeply understand, what are you doing and involves a lot of work in console. Seriously, thnk one more time before following them.