There is nothing more frustrating that doing all possible due diligence, raising the relevant Change Requests, completing testing, just to have a routine Windows Update fail.
Unfortunately, within increasing demand on software through mechanisms such as SDN (Software Defined Networks), SDS (Software Defined Storage) and HCI (Hyper-converged Infrastructure) a failure can have an exponential impact on workloads.
Introducing the Cumulative Update May 9th 2018 update
If you have this update installed, and are running Storage Spaces Direct.
EXCERCISE CAUTION WHEN COMPLETING MAINTENANCE
This update introduced a SMB resiliency mechanism and this mechanism can cause clusters under heavy load to experience node/cluster failures during node reboots. Systems that aren’t under heavy load are unaffected by this current bug and I’ve been able to successfully manage windows updates on numerous clusters.
However, if you do have a heavily loaded cluster, Microsoft have an article on how you can perform update more safely. In my experience this improved resiliency during maintenance, but didn’t resolve the issue (We only had 1 node failure instead of 3 - which resulted in the cluster staying alive, so Yay, I guess?)
This is still unresolved (As of Oct 5th), and my Microsoft Partner case is making very little progress. If you can afford the outage, I’d suggest that you organise complete cluster outages to install updates.
These high level steps may help you:
- Shutdown all VM’s on the Cluster
Get-VM -ComputerName (Get-ClusterNode) | Stop-VM - Detach all virtual disks
Get-VirtualDisk | Disconnect-VirtualDisk - Shutdown the Cluster
Stop-Cluster <<CLUSTERNAME>> - Install all updates and reboot nodes
- Start the Cluster
Start-Cluster <<CLUSTERNAME>> - Attach all virtual disks
Get-VirtualDisk | Connect-VirtualDisk - Monitor storage jobs - You may also want to invoke storage jobs
Get-VirtualDisk | Repair-VirtualDisk -AsJob Get-StoragePool S2D* | Optimize-StoragePool - Once complete power up all your VM’s
Its worth noting that the update window will be considerably shorter than using SCVMM or Cluster Aware updating as there are no storage rebuilds between node reboots.