If I want to temporarily add a node to a cluster to migrate VMs then remove the node; will it affect the VMs?
Will removing a node from a cluster affect the VMs?
- Thread starter FrankB
- Start date
So, you have a cluster running, want to add a new node, move the VM's on the cluster to the new node and remove the new node from the cluster? After that you have a empty cluster and one server running the VM's. Right?
First: If you remove a node from a cluster, it must never come back online again. If it will, your cluster will be screwed.
So, if you take care the current cluster and the new node will never see eachother again, this should be possible, I think (be carefull, make sure you have backups and do this at your own risk!)
1. Add the node to the cluster (pvecm add IP-ADDRESS-CLUSTER)
2. Transfer the VM's to the new node
3. Take care HA is disabled for all VM's
4. Change expected votes to 1
5. Take care the cluster and the single node doesn't see eachother
6. On the single node, remove the cluster nodes (pvecm delnode nodename)
7. On a cluster node, remove the single node
8. On the cluster, change expected votes back to original value (before adding the new node).
First: If you remove a node from a cluster, it must never come back online again. If it will, your cluster will be screwed.
So, if you take care the current cluster and the new node will never see eachother again, this should be possible, I think (be carefull, make sure you have backups and do this at your own risk!)
1. Add the node to the cluster (pvecm add IP-ADDRESS-CLUSTER)
2. Transfer the VM's to the new node
3. Take care HA is disabled for all VM's
4. Change expected votes to 1
5. Take care the cluster and the single node doesn't see eachother
6. On the single node, remove the cluster nodes (pvecm delnode nodename)
7. On a cluster node, remove the single node
8. On the cluster, change expected votes back to original value (before adding the new node).
The idea is to be able to shrink or grow the nodes as necessary. For example. Node-1 is a small weak node as this is required to make a profit. Eventually, that node will be too small one enough customers come online. We will be able to afford a DUAL XEON BOX with 32 gigs of RAM and 4TB of drive space. So the Idea was to add beefy NODE-2 to the cluster, transfer the COntainers/VMs and remove Node-1, leaving only NODE-2. Then when enough clients are gained, purchase another Beefy NODE-3 and create a cluster between NODE-3 and NODE-2 for HA, and other services, etc. I understand that when you add nodes they must not have anything running on them to prevent conflict with the numbering. I hope they will add renumbering of containers in the future, so that you can easily renumber containers before you add the server as a node in a cluster.
Wow sounds like a bad divorce. But the node that was removed from Cluster-A can be part of a new Cluster-B without any problems, right? Since they have never seen each other before.
Okay, first of all, a 1- or 2-node setup is not a good idea for a (HA) production environment, for a serious HA-cluster setup you need at least 3 nodes, so you always keep majority when one node fails. If you have a 2-node HA-cluster you are always at risk of a split-brain situation. When you grow you can upgrade to a 5-node cluster so you may lose 2 nodes in the cluster, but for a production HA environment you really need 3 nodes, at least. Charges always go for the benefits, there ain't no such thing as a free lunch.
Hmm, don't know if it will work, but I would never try to remove a node from Cluster-A and then add this node to Cluster-B without a re-install. If this change is needed, why not just do a clean install on the node, this is done in 10-15 mins....
So that would defeat the purpose of taking a unit that used to be a node in cluster become the master node in a new cluster. The idea was to transfer all my VMs to a beefy node, break it off and make it the master node in new cluster. Obviously you can't format a machine running VM's.
I understand there is no free lunch, but that is the idea of the kinda flexibility that would be great to include in a product like this. I currently use rented dedicated servers. So I'm always upgrading or downgrading servers depending on the needs of the business. Sometimes you get clients that are with you for only 6 months or so and you need to eliminate the most expensive servers to save what you were spending on their service(del a node). Sometimes you get clients that leave you, and go to godaddy hosting, then when they realize that they are getting rate and process limited they comeback and you need to expand service again(add new nodes or upgrade/downgrade to a new server).
It would be great if proxmox made it easy and simple to add/remove/scale/migrate at will. For example, one of my current low end servers needs to be migrated to a beefier server to handle a new clients needs. The rent of a new server costs an additional 30 dollars a month. Upgrading the low end server I currently have to a beefier server is only 5 more bucks a month. So it makes financial sense to upgrade my current service. That being the case I need to migrate to the new server. In that scenario, you are saying that adding the new server to the node and transfer the vms and deleting the original node will put me at some sort of disadvantage if I want to repeat the process in the future?