Hi all,
In the past when I've needed to apply windows updates to my 3 Hyper-V cluster nodes I used to make a note of which VM's were running on each node, then I'd live migrate them to one of the other cluster nodes before pausing the node I need to work on and
carry out the updates, once I finished installing the updates I'd then simply resume the node and live migrate the VM's back to their original node.
Having recently upgraded my nodes to Windows 2012 R2 I decided to use the new functionality in Failover Cluster Manager where you can pause & drain a node of its roles, perform the updates/maintenance, and then resume & fail roles back to the node,
unfortunately this didn't go as smoothly as I'd hoped, for some reason it seems like the drain/fail back decided to be cumulative rather than one off jobs per-node ... hard to explain, hopefully the following will be clear enough if the formatting survives:
1. Beginning State:
Hyper1 Hyper2 Hyper3
VM01 VM04 VM07
VM02 VM05 VM08
VM03 VM06 VM09
2. Drain Hyper1:
Hyper1 Hyper2 Hyper3
VM04 VM01
VM05 VM02
VM06 VM03
VM07
VM08
VM09
3. Fail Roles Back:
Hyper1 Hyper2 Hyper3
VM01 VM04 VM07
VM02 VM05 VM08
VM03 VM06 VM09
4. Drain Hyper2:
Hyper1 Hyper2 Hyper3
VM01 VM04
VM02 VM05
VM03 VM06
VM07
VM08
VM09
5. Fail Roles Back:
Hyper1 Hyper2 Hyper3
VM01 VM07
VM02 VM08
VM03 VM09
VM04
VM05
VM06
6. Manually Live Migrate VM's back to correct location:
Hyper1 Hyper2 Hyper3
VM01 VM04 VM07
VM02 VM05 VM08
VM03 VM06 VM09
7. Drain Hyper3:
Hyper1 Hyper2 Hyper3
VM01 VM04
VM02 VM05
VM03 VM06
VM07
VM08
VM09
8. Fail Roles Back:
Hyper1 Hyper2 Hyper3
VM01
VM02
VM03
VM04
VM05
VM06
VM07
VM08
VM09
9. Manually Live Migrate VM's back to correct location:
Hyper1 Hyper2 Hyper3
VM01 VM04 VM07
VM02 VM05 VM08
VM03 VM06 VM09
Step 8 was a rather hairy moment, although I was pleased to see my cluster hardware capacity planning rubber stamped, good to know that if I were ever to loose 2 out of 3 nodes everything would keep ticking over!
So, I'm back to the old ways of doing things for now, has anyone else experienced this strange behaviour?
Thanks in advance,
Ben