Prxomox HA

[manues50]

Hello

I was trying to create a Proxmox HA infrastructure from last few days. I tried with glusterfs ,ceph and CIFS but I am not able to get a complete solution.

The intention is if one of my Proxmox nodes went down another node should serve services ( VMs and containers) without any downtime.

And I have only two Proxmox nodes and one samba share is available now.

 

[aaron]

And I have only two Proxmox nodes and one samba share is available now.

Well, for HA you need at least three nodes in the cluster. But the third node doesn't need to be a full Proxmox VE installation. You can add a so called QDevice to the cluster. Basically a service that needs to be installed on the Proxmox nodes and the counterpart on a third device. Check our documentation for this [0].

As for the storage in that situation you can forget using Ceph because that needs at least three nodes for it's cluster. One option is to use the Samba share as storage. The other would be to use ZFS with replication between the two nodes. Depending on how often you replicate the important VMs you will have a small to not so small data loss if a node fails and HA on the other boots up the VM. AFAIK you can replicate down to each minute.

No idea about GlusterFS as I do not have any experience with it.

[0]: https://pve.proxmox.com/pve-docs/pve-admin-guide.html#_corosync_external_vote_support

 

[t.lamprecht]

Well, for HA you need at least three nodes in the cluster. But the third node doesn't need to be a full Proxmox VE installation. You can add a so called QDevice to the cluster. Basically a service that needs to be installed on the Proxmox nodes and the counterpart on a third device. Check our documentation for this [0].

It's still recommended that you use more nodes. Remeber that once a node fails the remaining need to take over the HA Services (VMs/CTs) from the failed node. If you only have two nodes + QDevice a single node must be able to host all of the HA services at the same time. Thus it's most of the time better to have more nodes to be able to handle more failures and spread the load out more evenly in case of such failures. Also you can then use Ceph, which fits really well for a HA setup.

without any downtime.

No downtime is not possible, in case of a full hardware failure the HA services will be recovered and started on a working node after 60 seconds (best case) to 120 seconds (normal case).

Further, this is the sub-forum about Mailgateway HA, thus I moved it over to the PVE one.