Data Engineering › Data Governance & Privacy
Data Fabric
An integration layer that connects data across systems using metadata.
Also known as: data fabric architecture, fabric data integration, active metadata fabric, federated data layer
A data fabric is a metadata-driven layer that connects data across many systems, so people can find, access and combine it without copying everything into one place first. The systems might be several warehouses, a lake, operational databases and SaaS tools, on premises and in different clouds. The fabric uses metadata, lineage, semantics and policy to route a request to the right source, apply the right controls, and return a consistent result.
It is a response to a real constraint: centralising all data is slow, expensive, and sometimes forbidden by residency or contractual rules. A fabric tries to leave data where it lives and make it usable in place.
What it is usually built from
- A metadata catalog covering technical, business and operational metadata (metadata types).
- Lineage and a graph of how datasets relate, used to find sources and assess impact.
- A semantic layer that maps business terms to physical columns, so “revenue” means one thing (semantic layer).
- Federation or virtualisation, which queries remote sources on demand rather than loading them.
- A policy engine that enforces classification and access wherever the data is (data classification, data governance).
The “active metadata” idea is central: the metadata is not just documentation, it drives routing, recommendations and policy automatically.
How it relates to data mesh
They are often mentioned together but are different in kind. Data mesh is mostly organisational: domain teams own their data as products. A data fabric is mostly technical: an integration and metadata layer. A mesh can be built on a fabric, and a fabric can support a mesh’s self-serve platform.
Trade-offs
Federation is not free. Querying across systems at runtime is harder to optimise than a single warehouse and can be slow or fragile; join across sources with care, and cache or materialise where it matters.
It needs good metadata to work. A fabric over poor, stale metadata just routes people to the wrong place faster.
The term is used loosely by vendors. The substance is metadata management plus integration plus governance; ask what a product actually does rather than trusting the label.
It does not replace modelling or ownership. A fabric connects data; it does not decide what a good table looks like or who is accountable for it.
Treat a data fabric as an architecture you grow into as your metadata and governance mature, not a product you install and finish.