Analysts need to query a CRM database, an ERP database and files in cloud object storage together through a single SQL interface. A compliance rule forbids copying the data into a central repository, so each source must remain the system of record. Which approach meets this need?

A virtual data warehouse, often implemented through data virtualization or federated query, presents multiple sources as one logical database. The virtualization layer holds metadata about each source, translates a user's SQL into source-specific requests, pushes filters and joins down where possible, and combines the results, without persisting a copy of the underlying data. This satisfies both requirements: analysts get a single SQL interface, and every source remains the system of record because nothing is copied into a central store. An ETL pipeline transforms data and loads it into a physical warehouse, which is exactly the central copy the compliance rule prohibits. An ELT pipeline also loads data into a warehouse, simply deferring transformation, so it violates the rule in the same way. Robotic process automation drives application user interfaces to automate repetitive tasks; exporting reports would create copies, would not provide an ad hoc SQL interface and would be fragile. Exam caveat: virtualization trades storage and latency of copies for query-time load on the sources, so performance and source availability must be considered, and some products cache results, which may count as copying under a strict rule. Operational check: run a representative cross-source query, review the execution plan to confirm filters are pushed to each source, and verify that no result data is persisted in the virtualization layer.