A pipeline's source system adds two columns and renames a third. The pipeline continues to run successfully but the renamed column's data disappears from the output. What practice would have caught this before consumers noticed?
Select an answer to reveal the explanation.
Short Explanation
An upstream rename is silent by design—the old column simply isn’t there anymore, and a tolerant pipeline shrugs and carries on. Assert the expected schema on the way in, and the run fails loudly instead of quietly emitting nothing.
Full Explanation
The correct practice is schema validation on ingest, which compares the incoming structure against what the pipeline expects and raises an alert or fails the run when they diverge. Without it, a rename presents as an absent column, tolerant logic proceeds, and the loss surfaces only when a consumer notices missing data.
Increasing the engine size so the pipeline processes the wider source faster is incorrect because compute has no bearing on structural change detection. Faster processing produces the same empty column sooner.
Reducing query history retention so recent runs are easier to find is incorrect because history retention affects how long you can review past runs. Shortening it makes investigation harder, not detection earlier.
Switching the source connection from ingestion to federation is incorrect because federation exposes the same renamed column. The change is upstream and reaches you either way.
For the exam, connect upstream schema drift to validation at the ingest boundary.