Federation gives a query engine access to multiple sources without first making one complete centralized copy. Trino’s documentation describes connectors to source systems and catalogs that configure them, allowing a query to refer to data through catalog and schema names (Trino, Concepts). This is a way to access data across systems; it does not automatically supply business definitions or decide who should be allowed to read each dataset.
For example, an analyst could join customer records from one source with subscription events from another. Before giving an agent that query surface, test the join keys, the freshness of each source and the behavior when one source is unavailable. Compare the result with a known reference query. A query that executes successfully can still use the wrong interpretation of customer or subscription.
Describing his preferred response to fragmented enterprise data, Jitender Aswani says, “put a virtual compute layer on top through data federation.” (episode 66, 15:11).
For builders, evaluate federation against the actual workload rather than assuming it replaces every warehouse or pipeline. Record source-specific permissions and the evidence needed to explain a result. Context graph covers business interpretation; conversational analytics covers a natural-language interface to the data. Do you need to centralize data for AI? develops Aswani’s argument and its practical limits.
Sources
- Trino, Concepts — Describes Trino connectors, catalogs and distributed query execution. This implementation example does not establish that federation removes every copy, transfer or governance task.