[SOLVED] Configuring a non-High Availability cluster possible?

[polarox]

Hi everyone. I've been reading the docs about clusters and I have a question. Is it possible to have a PVE cluster without making it high availability? I have two servers running proxmox and I want a non-high availability cluster so I can leave one of the servers off most of the time to save power and only turn it on when I need extra computing resources. There are places in the documentation that talk about the cluster and high-availability as though they are separate, but I don't see any explanation of how to disable HA.

If it's not possible to have a cluster without HA, is there any configuration that allows me to migrate VMs to the other host when needed?

 

[polarox]

OK, so I finally found the documentation I was looking for. It seems what I want is the following configuration:

 In /etc/pve/corosync.conf

quorum {
           provider: corosync_votequorum
           expected_votes: 2
           two_node: 1
           wait_for_all: 0
       }

This would give me the two node configuration referenced in man votequorum and I would only need to boot one node to get the initial quorum. I can understand why this configuration is not desirable, but since I know which server is my main node, this one would always get booted first in the rare event that both machines are down.

I'm totally open to different ways of doing this if somebody has a better suggestion or recommendation.

 

[dietmar]

Instead, I would simply assign 2 votes to your main server..,

 

[polarox]

Instead, I would simply assign 2 votes to your main server..,

Just curious, why is that preferable? It seems at first glance to be less flexible than using the two_node option, but I'm no expert.

 

[dietmar]

I just said that I would prefer it (pernonal opinion). But it seems to fulfill all your requirements. The two-node config is too much magic for me, and even dangerous if you disable wait_for_all and do not use active fencing.

 

[polarox]

I just said that I would prefer it (pernonal opinion). But it seems to fulfill all your requirements. The two-node config is too much magic for me, and even dangerous if you disable wait_for_all and do not use active fencing.

Thanks for the info!