Question

You design a Cosmos DB for NoSQL solution for an e-commerce site. Product documents are stored in a 'products' container, and customer reviews are stored in a separate 'reviews' container partitioned by /productId. The most common query on the reviews page displays each review along with the product's current display name. Product names are edited rarely (a few times per year), but review pages are read millions of times per day. Which data modeling approach minimizes RU cost for the high-volume read?

Accepted Answer

Denormalize (embed a copy of) the product display name into each review document, and update the copies with a background process when a product name changes.

Answer

Store only productId in each review and perform a separate point read against the products container for each review displayed.

Answer

Use a cross-partition JOIN query across the products and reviews containers to combine the data at read time.

Answer

Move reviews into the products container as an embedded array so every review is retrieved with the product.