Finding Consistency in an Inconsistent World: Towards Deep Semantic Understanding of Scale-out Distributed Databases

We present a new problem in data storage: how to build efficient backup and restore tools for increasingly popular Next-generation Eventually Consistent STorage systems (NECST). We show that the lack of a concise, consistent, logical view of data at a point-in-time is the key underlying problem; we suggest a deep semantic understanding of the data stored within the system of interest as a solution. We discuss research and productization challenges in this new domain, and present the status of our platform, Datos CODR (Consistent Orchestrated Distributed Recovery), which can extract consistent and deduplicated backups from NECST systems such as Cassandra, MongoDB, and many others.