A design places lakehouse data on object storage and states that access will be controlled entirely through bucket permissions granted to each analyst. What is the flaw?
Select an answer to reveal the explanation.
Short Explanation
Bucket permissions work at the wrong granularity. If an analyst needs three columns of one table, you hand over the whole set of objects, and direct file reads skip the governance policies your engines would apply. Keep entitlement in governance, not storage.
Full Explanation
The flaw is that bucket permissions operate on objects and prefixes. They cannot distinguish a table from the files beneath it, let alone restrict columns or filter rows, so granting an analyst bucket access necessarily grants far more than the requirement. Worse, it creates a path to the data that bypasses the engines where governance policy is enforced. The claim that bucket permissions cannot be granted to individual users, only to service accounts, is incorrect as a factual claim; individual grants are possible. The problem is granularity and bypass, not who can hold the grant. The claim that direct bucket grants are fine but must be duplicated in each engine's configuration is incorrect because duplicating storage grants into engine configuration would not add expressiveness, and the platform's model is that engines hold the storage credentials while user entitlement is evaluated by policy. The claim that there is no flaw provided each analyst is given read-only access is incorrect because read-only is precisely the concern. The risk being described is unauthorised reading, which read-only access does nothing to constrain. For the exam, keep the layers straight: storage credentials belong to the platform and its engines, user entitlement belongs to governance.