Veeam CBT and Rolling Cluster Upgrades

[chris.jones]

I was planning to perform an upgrade to a Hyper-V cluster at our primary site until I read that Veeam CBT would be disabled during the process. I was expecting the upgrade to take about 2 weeks because it takes time for me to re-image all the physical hosts with Server 2016 and then get maintenance windows scheduled to notify application owners that VMs will need to be shutdown temporarily to upgrade the VM configuration version.

Is it true that I just have to live with the fact that my backup windows will be totally ruined for that 2 week period because of this? Is there a better way to go about this upgrade while still having CBT enabled? Thanks.

[nmdange]

If you can take two physical hosts out of your cluster at once, then you can create a separate 2016 cluster. You can then live migrate individual VMs, just remember to shut them down as soon as you have a maintenance window to update the VM configuration version otherwise Veeam won't be able to use the new Hyper-V 2016 RCT (Resilient Change Tracking). This is how I did my upgrade and it worked pretty well. It also lets you test 2016 thoroughly with non-production VMs before moving your most critical VMs.

[chris.jones]

Thanks for the reply, but this seems kind of ridiculous. I shouldn't have to build a separate cluster just to prevent issues with my round-the-clock backup windows. How is this being handled in larger enterprises with hundreds or even thousands of VMs?

[csydas]

They're on VMware

Sorry, couldn't help it.

I appreciate the difficulty you have, but you have to understand that pre-2016 there was no CBT for Hyper V. Mixing and matching levels of Hyper-V just isn't kosher in general within a cluster (homogenous clusters are always preferable), and a home-rolled CBT solution is just as fragile as the problem its built for.

The truth is for larger enterprises, you have enough spare hosts that you just dump the VMs elsewhere and call it a day; there's enough redundancy built into the system that you can find space for everything, even if it's a crunch for awhile.

It's not a great answer, but it's about the best option you can get.

P.S. - Run all your updates on your VMs before migrating to save yourself the headache of the 37xxx errors on Hyper V 2016. Most of them seem to be fixed now, but it's all dependent on patch version.

[nmdange]

My environment has 600 VMs, is that big enough? Given the choice between having to quickly upgrade with rolling upgrade, or run clusters side-by-side and do a controlled migration, I chose to do the side-by-side setup. However, I do use scale-out file servers for storage. That makes running multiple clusters easier since you don't need separate CSVs for each cluster. Though in my opinion a large environment should have multiple clusters anyway for independent fault domains.