Document, key-value, wide-column, and graph stores, with honest selection criteria.
4 questions in Databases & SQL.
It is decided by everything besides fetching a profile by user id. If every read is by id and every write replaces the whole profile, a key-value store is enough. Login by email, unique emails and field-level updates push towards a document database, or towards index items you maintain yourself.
Database selection in a design round rests on three questions: what the hot path reads and writes, by which key; which facts must never be wrong; and how much data and traffic arrive at peak, derived as numbers. Then commit to one store in a sentence, with the reason and the fact that would change it.
Compare key-value, document, wide-column and graph databases by access pattern: known-key lookups, aggregate documents, partition-local range reads or unbounded relationship traversal. The right NoSQL choice follows the query shape, not a generic scalability claim. It also connects cassandra to the point an interviewer is testing.
An acknowledged write is only guaranteed on the replicas that acknowledged it, so a read served elsewhere can legitimately return the previous version. Locate the staleness - replica, secondary index or cache - then buy read-your-writes for that operation only, and price what it costs during a partition.