Data Platform concepts
Four ideas run underneath every feature of the Data Platform. Understand these and the rest of the topic reads as consequences of them.
- Work happens inside a boundary. Every resource belongs to a workspace, or to a clean room when more than one trust domain is in the room. Isolation is between those boundaries, not between customers.
- You provision resources, not databases. Catalogs, tables, structures, storage namespaces, streams, functions, and data products are the things you create. The cell's own internal stores are plumbing you never touch.
- The read is governed. Row filters, column masks, classification, lineage, and cost attach to the query itself, compiled into the plan rather than wrapped around it.
- Many access methods, one policy. A dozen engine families speak the protocols your clients already know, and every one of them re-enters the same authorizer. The wire is not a bypass.
In this area
| Page | What it covers |
|---|---|
| Workspaces and clean rooms | The universal ownership boundary, and the mutual-distrust overlay for parties who must collaborate without trusting each other |
| Resources you provision | The full inventory of what you create, and the five properties every one of them carries |
| Governed reads | How classification, filters, masks, lineage, and cost compile into every query |
| Engines and access methods | The engine families, the protocols each is reached over, and why a wide protocol surface is not a wide attack surface |
Looking for the internal databases?
A cell partitions its own state into nine logical databases with architecturally enforced boundaries. That is real and worth understanding, but it is not a surface you manage, so it lives in Platform Architecture → Inside a cell.
Related topics
- Data governance, the enforcement plane the governed read rests on.
- Lakehouse & catalog, the open formats and the catalog over the substrate.
- Platform Architecture, the cell model these concepts assume.