HA is fully contained within the managed storage platform. If you are running a customer VM on HV A, and it has a virtual disk attached which has replicas residing on HV B and HV C. If HV C were to crash for example, the data path will experience a temporary delay while the block path times out the failed path, and then transparently failover to just the good set of paths. i.e. the VM doesn't experience any block IO errors, or loss of data, just a degraded IO queue response for a short period of time.