HA for 1 VM on site

[imadam]

Hi,
I have customer for whom I started to research financially viable solution. This customer has 1 x Windows 2019 Server VM per site where HW need to provide failover in case that one node goes offline. Vmware Ess+ with some SDS is to expensive. Lookig at proxmox as 2-node (if possible) or 3-node solution. HW specs are really low. i3/8GB RAM/128GB SSD all we need for this VM.

Any recommendation on HW and complete setup using proxmox?

Br, Imadam

 

[Astraea]

For Proxmox and HA, you will need a minimum of 3 nodes. You should also have HA shared storage for this Proxmox cluster as well this can be either through a separate storage deployment or something like CEPH.

Given the VM specs in my opinion I would build 3 identical nodes as follows:
Intel i5 or simular AMD CPU
16GB RAM
2 x 25GB SSDs configured in a ZFS mirror for Proxmox and ISO images
2 (ideally 4) x SSDs configured using CEPH / Though these could be removed if you are using a separate storage system
1 x 1GB NIC for Proxmox and cluster traffic
1 x 1GB NIC for CEPH or storage traffic (10GB is better if available or able to be implemented)
1 x 1GB NIC for VM traffic (10GB is better if available or able to be implemented)

Ideally, you would have 2 NICs for each type of network connection running to redundant switches. When designing an HA setup you need to keep in mind any single point of failure be it a total node failure or failure within a single node (such as 1 of 2 PSU failing) and also any support or connected infrastructure/devices that could bring down the HA implementation.

 

[imadam]

Thank you for detailed specs. If one node fails, cluster will run?

 

[Astraea]

In a three-node cluster if one node fails, yes the cluster will still have a quorum and will still run. However, depending on the load on those other two nodes you may or may not be able to have all services running using them. You will need to spec your system so that two nodes are able to handle the load. Then adding the third is what is giving you fault tolerance in the cluster so to speak.

In my cluster, I will have 7 nodes (currently only have 5 and not using automatic HA failover as the current layout was not designed for it) so I will have the ability to have 3 nodes fail before I lose quorum however I will not be able to support all the VMs and containers that are running while only using 4 nodes depending on which ones have failed and so I will be looking as using HA groups and other Proxmox HA methods to prioritize critical systems over not critical systems. I will also be leveraging docker running in VMs to allow for a swarm to self-manage docker containers without the need to migrate the VM around the cluster.

 

[imadam]

We need this HA for very small VM which works fine on i3, 8GB, 128GB SSD. Is Intel NUC good fit for this?

 

[imadam]

Related to this config:
Intel i5 or simular AMD CPU
16GB RAM
2 x 25GB SSDs configured in a ZFS mirror for Proxmox and ISO images
2 (ideally 4) x SSDs configured using CEPH / Though these could be removed if you are using a separate storage system
1 x 1GB NIC for Proxmox and cluster traffic
1 x 1GB NIC for CEPH or storage traffic (10GB is better if available or able to be implemented)
1 x 1GB NIC for VM traffic (10GB is better if available or able to be implemented)

is it possible for:
2 x 25GB SSDs configured in a ZFS mirror for Proxmox and ISO images
2 (ideally 4) x SSDs configured using CEPH / Though these could be removed if you are using a separate storage system

to be 2 driver M.2 and carved out like 25GB and rest for VM

 

[imadam]

or can we just have node with 1 x 32GB m.2 and 1x128GB since we don't care about redundancy within node itself?