Thanks for sharing. While this solves the review CRUD and user path on a point to point product, how would you solve the problem space for a set of related products with high affinity. E.g, reviews on all Samsung phones with 24 MP cameras, or review all dishwashers from Bosch? This is probably a goldmine of information for analysts and for users looking to get an aggregated set of reviews across a brand/aspect.

While the reviews are in moderation queue, apart from assigning a unique integer within the product scope, they should be tagged to an analytical vertical/store path as well. This can be done by exploiting the globally sortable, portable and collision free properties of a "ksuid". Other approaches may also work.

However I feel that with the massive data and the "obvious duplication" in index, it's better to utilize a review knowledge graph, do offline processing during ingestion and then show richer reviews not limited to the original product to the users. Maybe a "See Reviews from Related products" section where you need to rank reviews on "related products" in O(1)