A vendor proposes replacing your ingestion pipelines with data virtualisation — query everything in place, no copies. Assess it.
Show the full answer Hide the answer
What the interviewer is testing
Whether you can evaluate an attractive proposition against physics rather than against its marketing.
What is genuinely true
No copies, no pipelines, no duplicated storage, and one governed access point enforcing permissions consistently across heterogeneous stores. For discovery, for joining a small reference table to a large one, and for avoiding a pipeline built purely to satisfy one query, it is a legitimate and useful tool.
What the physics constrain
A federated join must move data across the network between systems, and the optimiser has poor statistics on the remote side — so plans are frequently bad in ways that are hard to diagnose.
Performance is bounded by the slowest source. A query joining the warehouse to a legacy database runs at the legacy database's speed.
It puts analytical load on operational systems, which is precisely where it does least good. The transactional database now serves scans that compete with customer transactions. This alone disqualifies it as a wholesale replacement.
No historical snapshot. Querying in place gives current state. Point-in-time reporting, reproducibility of a prior figure, and slowly changing history all require materialisation.
No transformation persistence. Complex conformance and business logic recomputed per query is both slow and expensive.
The assessment
Complement, not replacement. Virtualisation is the right tool for the queries not worth a pipeline; consolidation is the right tool for the ones that are — high-volume analytics, anything needing predictable latency, anything requiring history, and anything that would pressure an operational system.
What a strong answer adds
Proposing the concrete test: take the three heaviest existing pipelines, implement them as federated queries, and measure latency, source system impact and cost at real volume. That converts a vendor claim into evidence in about two weeks.
Common weak answers
Rejecting it outright, which discards a genuinely useful capability. Accepting the replacement framing.