Host Failure and ‘Write-Back One Replica’ Caching Policy
‘Write-Back One Replica’ Caching Policy ensures that a single host failure in a cluster doesn’t result in data loss.
Now if you have a 2-host cluster and one host fails, then the VMs that are in ‘Write Back’ operational state on the host that is still up and running will move to ‘Write Through’ operational state. And VMs that are on the failed host that now restart on the solitary host will also move to ‘Write Through’ operational state. This is because there is no other host available to mirror the writes and hence the caching policy will be throttled to ‘Write Through’. Once you restore the cluster back to its original 2-host state, then the VMs will go back to their original ‘Write Back’ operational state.