2 nodes failover/replicated cluster?

[michal_jot]

Hi, i need help with this topic, i have a 2 same servers HP DL360G8, on this machines i have a Proxmox installed, one of them working fine in our datacenter and it hosting 8 critical VMs actually. We need a replication and failover on second machine. My plan is:

1. Machine "A" hosting 8vms and replicating it to another machine, someday the server is broke..
2. ..And machine "B" takes over his roles and hosting this vms on site, people can connect to all vms and all comapny is working fine.

How i can do this? I cant migrate to ZFS because i cant stop this machines now. On both machines we have 4 disks - 2 SSD with RAID 1 and 2 HDDs with RAID 1 on PERC H400 controller.

If it necessarily i can involve a third server but it is old and slow DELL R210 with 32GB RAM and HDD's only.



Can you help me guys?

 

[VictorSTS]

If you don't have ZFS you can't replicate among nodes [1]. Also for a cluster to sustain a single node failure or maintenance you need at least 3 nodes: quorum requires majority and in a 2 node cluster having just 1 working isn't majority but just 50%.

What I would do is getting backups of every VM in a different server/storage. Test the backups are recoverable and VMs are operational after a restore.

Then, setup ZFS on that second DL360G8 (it has no production loads at the moment, right?). Then, create a cluster with your 2 DL360G8 + that old R210 starting the cluster configuration on the DL360G8 that has your critical VMs. Plan for your cluster setup, networking, IPs etc *before* implementing the cluster, there are things hard to change afterwards.

Once your 3 node cluster is up and running, live migrate your critical VM to the other DL360G8 to the ZFS storage using CLI (WebUI still has no option to migrate changing the destination storage but AFAIK it's in the works):

qm migrate VM_ID DESTINATION_NODE_NAME -targetstorage local-zfs --online

Shouldn't be any downtime if done properly, BUT,

Once all VM's are moved, setup ZFS on your first DL360G8. Then you'll be able to live migrate them from the gui and set up replication.

As your R210 will be acting just as a witness node and no VM's should be run on it because is too old, I suggest to install Proxmox Backup Server on it and setup backups of your VMs to this PBS in the R210. It is always a good idea to have backups in a different server than the one they are running in.

NOTE: this may get quite complex, so practice in labs or get professional support if you are not sure about what you are doing.

---

[1] https://pve.proxmox.com/wiki/Storage_Replication

 

[michal_jot]

Thank you for such a comprehensive response!
But i have one issue - one of the DL360G8 is 200kilometers or 125 miles away from first datacenter, we have pernament connection betwin locations via P2P Forti VPN. The first location has a 192.168.10.x/24 addressation and 2nd location have a 192.168.11.x/24 addressation, does this somehow interfere with the way you suggested?

 

[VictorSTS]

Proxmox cluster is not "wan-aware", as corosync requires latencies around 2ms among all nodes. Your latency will probably higher than that, besides the implications related to the fact of dealing wih failures related to WAN conectivity + VPN. So no, you probably can't do what I explained in my previous post.

In such scenario just install Promox Backup Server in your second, remote DL360G8 and get backups there from your first, local DL360G8. When this one breaks, use Live Restore of PBS to restore your VMs. Obviously this will imply data loss (data not backed up since the last backup and the moment the node breaks) and downtime as VMs will have to be restored.