Abstract: Data sets are synchronized between two or more clusters of nodes that support different versions of files (e.g., stub files) within a distributed file storage system. Moreover, the distributed file storage system employs a tiered cloud storage architecture. In one aspect, for stub files having versions that are not commonly supported by the two or more clusters, an application protocol interface (API) is utilized that employs a deep-copy process wherein cloud-backed data referenced by a stub file is retrieved from a cloud storage (e.g., public cloud) and sent by a primary cluster to one or more secondary clusters. Moreover, the API can determine an optimal synchronization process on a per-file basis.