Transfer VM within Cluster : IP Addressing Questions

[Mat0]

Hey Community,

I have been working with Proxmox for some time now and want to make the next step towards Cluster.

We are hosting our 3 Server at Hetzner.
Every Server has it own dedicated IP from different Subnets. The ones Hetzner gives you when you order a server.
I also understand that I can use vSwitch to establish a private network for the cluster sync processes.

Until now I ordered single IPs for our VMs on our single Proxmox host.

For a cluster, I am uncertain how Proxmox handles VM IPs, if the Guest is transferred to another node.

- Does it stick to it's IP and Proxmox cluster handles somehow the routing?
- Should I order 1 dedicated subnet like /29 or /28 and assign my nodes and VMs to this?
- Do I need 3 dedicated subnets, one for each node?
- Should I order 1 dedicated subnet and assign it to a vSwitch Profile plus place all nodes and guests within this vSwitch network?


You cluster experienced people probably think what an obvious and easy answer.
For me, I am just stuck on this last issue and need an answer for my upcoming Hetzner ressource/order planning.

Your help is very much appreciated,

Mat

 

[sb-jw]

For a cluster, I am uncertain how Proxmox handles VM IPs, if the Guest is transferred to another node.

This has nothing to do with Proxmox, it has something to do with network design.
The IP addresses must be usable from all nodes. I mean that Hetzner binds MAC addresses to the IPs, so you would probably have to enter the virtual MAC of your VMs here, if Hetzner allows this.

In a normal setup in the data center, where you get at least one VLAN, you usually also get complete subnets - anything else makes no sense for a cluster in my opinion. You can use these IPs in the entire VLAN.

From my point of view, this is much more about the specifics of the Hetzner network, about which I can't say anything, at least in detail. Either you're lucky and there's someone here in the forum with the relevant information, you can find posts about it using the search, or you can contact Hetzner support and document your answers here to help others.

 

[Mat0]

Thanks for your reply!
And you are correct, in a Hetzner environment you use bridges and the MAC addresses for networking.
And you have the choice to order single IPs or Subnets and add them to Server or what they call vSwitch (an internal) VLAN you can set up between Hetzner hosted server.

I asked Hetzner about this issue and they send me some "Hetzner" links I had a look at, but none answered this question.
They also suggested this forum, so I hope some people with Hetzner cluster experience will see this and and hopefully reply.

Thanks again for your hints.
I will also try the search.

 

[sb-jw]

I asked Hetzner about this issue and they send me some "Hetzner" links I had a look at, but none answered this question.

Can you share the links mentioned? Maybe something can be interpreted from this after all.

 

[Mat0]

Of course ...

https://docs.hetzner.com/robot/dedicated-server/network/vswitch/

https://docs.hetzner.com/de/robot/dedicated-server/ip/failover/

About the second link, I am not sure if a failover-IP is really needed.
I have the feeling that bridging to a vlan (vSwitch) with a public subnet (i.e. /28) might be the best bet to be able to migrate VM between nodes without loosing connectivity.

 

[sb-jw]

From my point of view, these articles are quite clear and Hetzner refers to the principle that I have already stated. Instead of a real VLAN, there is a virtual network on the physical one, which behaves similarly to a VLAN. However, requirements such as MTU 1400 are necessary for its use.

I therefore interpret the reference to the article to mean that you will only achieve your goal with a vSwitch. The article also points out that you can use all private areas and you can also book subnets, which are then available in the virtual network. There is no mention of the fact that you can do this with individual IPs.

I don't think a failover IP is necessary for you, it only switches between two destinations, for example if Server A fails then the IP of server B can be taken over. In principle it can also be compared with keeplalived, where you do it directly on the server level.

 

[Mat0]

Thank you very much for your hints and suggestions!
I will continue this way.