[PVE-User] Create proxmox cluster / storage question.

[Ronny Aasen]

On 28.02.2020 15:25, Leandro Roggerone wrote:
> Dear Eneko:
> When you talk about redundacy and availability, do you want...
> - HA? (automatic restart of VMs in the other node in case one server fails)
> - Be able to move VMs from one server to the other "fast" (without
> copying the disks)?
> Yes , this is what I want.
> So regarding my original layout from my 5.5Tb  storage:
> Im using 1T for LVM , 1TB for LVM-thin and  3.5 TB unassigned space, is it
> ok to use this unassigned space for a ceph ?
> Can I set it later ?  with server on production ?
> 
> Other:
> Using an NFS system, means to have an external server running a file sistem?
> So you should have at least two servers for the cluster and one for the
> file system?
> It seems to me that using ceph has a better redundancy plan an it is easier
> to deploy since I only need two servers. (am i right?).
> Thanks!
> 
> El vie., 28 feb. 2020 a las 11:06, Eneko Lacunza (<elacunza at binovo.es>)
> escribió:
> 
>> Hola Leandro,
>>


for NFS you have just moved your SPOF from proxmox to NFS. so unless you 
have a HA NFS solution like a SAN with controller failover i would not 
think about NFS.

for ceph you need 4 server to a have a proper fault tolerant HA system
you have 3 copies of data. and a node as a failuredomain. you want the 
disk layout to be fairly identical on the servers.

you can in theory run ceph on 3 nodes, but you will have a degraded 
cluster when you loose a node. and basically neuter the best features of 
ceph. so use 4 or more nodes. ceph really shines when you get to 8-10 
nodes. also most defaults assume a 8 node cluster, so you may need to 
tweak some warning levels when you want to run with fewer.

Also proxmox HA requires quorum. so proxmox also need 3 or more nodes. 
if you only have 2 nodes you can not do a reliable HA solution. But you 
can do a zfs replication system, where the running vm is replicated to 
the other server periodically. and you can start that backup vm when the 
main one dies.

Good luck
Ronny Aasen