scenario:
- 4 proxmox 3.4 nodes with AOE shared lvm storage backend. No containers, just KVM vms.
- upgrade cluster to 5.3
- migrate all vms off of one node
- install new 5.3 from iso on unoccupied node. (boot off of iso. I used a burned dvd)
- reconnect (now) 5.3 node to all shared storage. use same name for storage as before. (I've got one aoe shared storage and two nfs)
- shutdown some vms running on a remaining 3.4 node for manual migration
- copy vm XXX.conf files of the downed vms from /etc/pve/nodes/<3.4 NODENAME>/qemu-server/ to /etc/pve/nodes/<5.3 NODENAME>/qemu-server/ (I 'cat'ed the file on one terminal and copy/pasted into an editor on the other)
- start the vms on the new server.
- boot the 5.3 install iso on the next unoccupied node
- reconnect it to all shared storage
- create cluster (use the first node. clusters object to joining nodes with vms already on them. I didn't. I had to shutdown vms, backup the XXX.conf files on the first node and copy them back after joining the cluster)
- manually migrate more vms off the next node to be upgraded.
- proceed till whole cluster is upgraded
- I found this way easier than upgrading the OS with apt-get. I was able to jump two major versions ahead without much fuss
- Obviously containers would have made this much more complicated
- Maybe lvm storage was a factor in how easy this was. Maybe not.
- obviously there is risk that some bozo will restart vms on the un-upgraded nodes. Thankfully we are a very small shop and nobody will do that.
Last edited: