When an interviewer asks for insert, delete, and random access in a single data structure, the answer is combining structures: a hash map for lookups and an array for random indexing. The trick is the swap-to-end deletion strategy that keeps the array compact.
Amazon's distributed systems need random sampling for load balancing and A/B testing. This is also a design-flavored problem that shows up when the interviewer wants to see if you think about data structure trade-offs rather than just writing loops.