NOT_SUPPORTED_DURING_UPGRADE()

paco

I am running into the same issue as in this post. But I'm confused as to how one upgrades a cluster of hosts from 8.2.1 to 8.3 without massive downtime. I have 3 hosts, A, B, and C. A is the master. I moved all the workloads off of A, and then upgraded it to 8.3.

I'd like to move workloads off one of the slaves, so the slave can take as long as necessary to upgrade. The upgrade is not quick.

The only way to upgrade from 8.2.1 to 8.3 is to boot from the ISO, which is fine. But once a node is upgraded, I can't migrate workloads to it from the non-upgraded nodes. How do I roll this upgrade through the cluster without just taking an entire host and all its workloads offline for 45 minutes while it upgrades?

I have been able to move workloads from old to new by shutting down a VM on an old node, using the copy function in Xen Orchestra to copy it to the upgraded master, and then booting the new copy. But that takes the VM offline for the duration of the copy. A few of my VMs can tolerate that, but not many.

What am I missing?

olivierlambert

The upgrade process doesn't involve downtime.

You start with the master: evacuate all VMs to slaves
Upgrade the master to 8.3 via the iSO
When done, target one slave: move all VMs elsewhere (including the up to date master), and repeat the process.

In short, you always migrate VMs BEFORE upgrading a node, and always starting with the master. The order of slaves doesn't matter.