In-place upgrade of cluster 1.9 to 2.3

[jinjer]

Hi,
I know this is a no-no, but I'd like to give it a try.
I upgraded a test custer from 1.9 to 2.3 in my office, but now I need to do it on real servers in a web farm (ouch!).

I think that a node cannot join a cluster in 2.3, if the node contains any virtual machines. So the correct procedure would be to upgrade nodes and then import old machines in this sequence:

1. Upgrade the master node
2. Create the cluster
3. Import VM running on master in the cluster.

Then on each node repeat:

1. Upgrade node
2. Join cluster
3. Import VM from that node to cluster.

Is that correct? Can I import machines from a secondary node in the cluster with the "import" script?
I'm trying to minimize downtime so that I would like to import VM on each node as soon as it's running the new proxmox.

jinjer

 

[tom]

the in place upgrade uses not not the pve repository, it uses a special one http://download.proxmox.com/debian/dists/squeeze/pve1upgrade/

after this, you need to follow this howto:
http://pve.proxmox.com/wiki/Downloads#Update_a_running_Proxmox_Virtual_Environment_2.x_to_latest_2.3
(adapt your sources.list)

for the inplace upgrade, see http://pve.proxmox.com/wiki/Upgrade_from_1.9_to_2.0

 

[jinjer]

Tom,
thank you for the quick response.

Can I upgrade nodes in the cluster one at a time and import the VM/CT in the cluster one at a time?
I'm sure I cannot join a cluster with a node that has VM/CT provisioned on it.
Can I join new nodes and import their VM/CT after the join, but one at a time?

That is I don't have the luxury of stopping all the cluster at once, and upgrade all nodes at one and then join them together before importing their VM/CTs

Thank you

 

[tom]

I cannot tell the best strategy for you. If you do an inplace-upgrade from all nodes the same time and you got an issue, all your boxes are offline.

Looks like a high risk strategy. You need to find the best mix, minimum downtime, low risk.

 

[jinjer]

Would you suggest to upgrade the boxes one at a time and move VMs from an old box to a new box?

If so, I have a question about the KVM machines running on shared storage (nfs). These are very big (up to 1TB of storage) which is not backed up by proxmox. Instead I am taking snapshots on the NFS server.

If i make a backup of such a machine from proxmox 1.9, how do I restore it to 2.3? It could take hours to complete the restore over an iscsi network at gigabit speeds.

It there a "shortcut" like importing the configuration file from 1.9 to 2.3 and then copying the disks at the storage server level (i.e. on the san) ?

Thank you.

 

[jinjer]

At the end I took the chance and upgraded my old 1.9 cluster to 2.3

The catch is that the nodes must be upgraded one at a time and must join the cluster before having their machines imported using the pve-upgrade script here: http://pve.proxmox.com/wiki/Upgrade_from_1.9_to_2.0

So, what I did was:

First node:

1. Upgrade from 1.9 to 2.0 using the upgrade script
2. Reboot with new kernel and pve environment.
3. Change sources.list to use the "pve" repository in place of the "pve1update" one
4. Upgrade from 2.0 to 2.3
5. reboot with new kernel from 2.3
6. create the new cluster using the latest 2.3
7. import machines from 1.9 using the upgrade script.

On the other nodes, the procedure is similar, but instead point 6 becomes "join the new cluster".

Now... there's an issue on all KVM hosts running Windows derivative systems. Apparently the new KVM in 2.3 messes up the hardware and I had to reinstall all virtio drivers (baloon/ethernet/disk).

Other than that I have not encountered other issues (yet).

Thank you for the help.

jinjer