Code Judging System: Data Storage and Indexing

The diagram splits the data into three parts, and each one wants a different kind of store.

Problem data. Problems live in a database tuned for reads, since they are read constantly and edited almost never. A NoSQL document store like DynamoDB or MongoDB fits well. Each problem document holds the statement together with an array of test cases. So one read returns everything needed to judge that problem. Its fields include id, title, description, difficulty, and the allowed languages