Shannon Lowder walks us through a multi-zone approach to storing data in a data lake:
Our first zone is the raw zone. This zone will serve as the landing point for source files. Like the extract (or stage) schema in our data warehouse, we want these files to match the source system as close as possible.In the data lake, we actually go one step beyond saying we want the schema of our raw files to match the source system, we also want these files to be immutable.
Immutable means once they are written to the raw folder we shouldn’t be able to modify or delete them. That way, we can always reconstruct different states from these files without having to retrieve them from the source system.
Worth reading the whole thing.