...
Cassandra
Pros
Can support faster key based queries. For ex: a query on course_id and user_id
Need to scale only one DB to scale the entire courses infra
Cons
Limited query capability. Performance is guaranteed only when queried via the partition key.
Filtering either by user properties or course properties needs to be done in memory of the API after fetching the data from db.
Data joins to be done in memory.
Druid
Pros
Faster and easier to scale.
Supports joins from 0.18 version onwards
Can query on any dimension
Cons
Append only DB. The Samza/Flink job needs to take care of idempotency
Can query only by date field (as segments are created by date). Need to do custom data source design to be able to support the courses reporting needs which can become extremely complex
...
Edge Caching
...