For 3rd party HPCC data access (e.g. Spark), we need to expose the data via a dafilesrv service, in much the same way it is exposed in bare-metal today.

Unlike a bare-metal system, there will only need to be one dafilesrv that can serve up the data for cloud storage, however, that can be 1 end point service, that could horizontally scale the number of dafilesrv replicas based on demand/load.