Downsizing a SimpliVity cluster

[CadenLange]

Hi

I'm drafting a plan to downsize a 4 node SimpliVity cluster to a two node SimpliVity cluster. This is being down due to a reduction in virtual machines and the closure of a branch office. The current configuration is a stretched 4 node cluster  (running 5.3) with 2 nodes per office and using availability zones between the two offices. Ideally I want to end up with a two node cluster running in the head office without availability zones. Due to the reduction in VMs on the entire cluster there is enough capacity for all primary and secondary replicas to fit onto only two nodes.

Is the following plan workable:

• vMotion all workloads to the Head office nodes (All primary replicas now running in head office and secondaries in branch office)
• Remove the Availability zones (to allow secondary replicas to be rebuilt without zoning constraints at a later stage)
• Shutdown one of the nodes in the branch office (All VMs with secondaries on this node will now show a redundancy warning)
• Evict the other node in the Branch office (All secondaries on this node will be rebuilt onto the Head office nodes as the availability zones have been removed and the other node on the branch office is shutdown so cannot be a used to secondary rebuilds)
• Once this secondary rebuild operation is complete, power back on the other branch office node
• Once booted, evict the remaining branch office node (All secondaries on this node will be rebuilt onto the Head office nodes as the availability zone has been removed)
• End result should be a functional 2 node cluster in the head office with both primary and secondary replicas split acorss both these two nodes

Many thanks

C

[ghç9]

To mee the above plan looks ok and the the overall objective will be achievable. However, I would modify the procedure to avoid powering off one branch node before beginning eviction.

A sequential eviction with full HA rebuild completion between each removal is typically the safer and more supportable SimpliVity approach because it minimizes the time spent in a degraded redundancy state. 

•  vMotion workloads to Head Office hosts
• Remove Availability Zone configuration (svt-zone-unassign / svt-zone-delete followed by realize)
• Verify cluster is now operating as a normal non-stretched cluster
• Evict first Branch node
• Wait until Storage HA is fully compliant
• Evict second Branch node

During the above steps make sure that you follow the space capacity and hive replica distribution / situation on the nodes.

• Verify all VMs are in HA on the two remaining Head Office nodes

[CadenLange]

That’s great, many thanks. Yes, I agree that evicting the first branch node without shutting down the other first would probably be a safer option. This would result in some replicas being recreated on the second branch node but would mean that all VMs would spend less time without a replica. Thanks