Possible to have mixed cluster of HA and Non-HA nodes?

[victorhooi]

Hi,

I currently have a 3-node Promox HA cluster, with a converged Ceph cluster also across those three nodes.

I have an additional two machines I'd like to join the cluster, but purely for management - that is, they wouldn't be part of the HA cluster, or the Ceph cluster (They lack the Intel Optane drives for OSDs, and they also have slower networking).

Is this possible? If so, how do I add them to the cluster, without impacting on the HA or Ceph?

Thanks,
Victor

 

[wolfgang]

Hi,

yes you can do this if you use a Qdevice[1]
Without the QDevice you will lose management capability if one node is down.

1.) https://pve.proxmox.com/pve-docs/pvecm.1.html

 

[victorhooi]

Thanks for replying!

I'm a bit confused though.

Currently - I have a 3-node HA cluster (with Ceph across the same 3 nodes). If I lose one node - I still have management capabilities at the IP addresses for the remaining two nodes

Are you saying that if I add two additional nodes, for a 5-node cluster, I will lose that ability? As in, if any of the 5 nodes goes down, I will no longer be able to manage any of the nodes from their IP address? I'm not challenging this point - just curious why it is?

I've reading the "QDevice Technical Overview" section in the pvecm manpage - it seems it simply provides an additional quorom vote. It mentions that it's not recommended for an odd number of nodes?

My desired end-state is a 5-node Proxmox cluster, with HA running across nodes 1-3, and Ceph running across nodes 1-3, and the ability to spin-up VMs on nodes 4-5 in non-HA.

 

[wolfgang]

Sorry, my fault I guess you have just two nodes separate.
So, in the end, an extra two nodes cluster.

If you add this two nodes to the 3 nodes cluster it will work without problems.

 

[victorhooi]

Yup, correct, just the extra two nodes.

I just wanted to make sure that adding them didn't affect the existing HA or Ceph between nodes 1-3.

But you're saying it shouldn't, right? Do I need to do something to ensure that VMs don't get failed-over to them? (I believe they would run - but they only have 1Gb network connectivity - rather than 10Gb back to the Ceph storage cluster, so I assume that would cause issues).

 

[wolfgang]

You have to create ha-groups which are restricted to the three nodes.
So the guests are not allowed to migrate to the new two nodes.
see the section GROUPS for the needed information

https://pve.proxmox.com/wiki/High_Availability#_configuration