We address a practical challenge in agile web development against NoSQL data stores: Upon a new release of the web application, entities already persisted in production no longer match the application code. Rather than migrating all legacy entities {\em eagerly}/ (prior to the release) and at the cost of application downtime, {\em lazy}/ data migration is a popular alternative: When a legacy entity is loaded by the application, all pending structural changes are applied. Yet correctly migrating legacy data from several releases back, involving more than one entity at-a-time, is not trivial. In this paper, we propose a holistic $\textit{Datalog}_\textit{non-rec}^\neg$ model for reading, writing, and migrating data. In implementing our model, we may blend established Datalog evaluation algorithms, such as incremental and bottom-up, as well as ideas from the magic set algorithm. Our systematic approach guarantees that from the viewpoint of the application, it remains transparent whether data is migrated eagerly or lazily.