I think it’s an issue with NoSQL.
No joints, not ACID, duplication of data, constraints on read rather on write.
Some NoSQL tries to solve this “lack” of features.
I said lack with quotes because usually that’s what you’re seeking for when choosing NoSQL.
The biggest pain point for me is querying, especially nested stuff. This drives me crazy.
Nevertheless, I go with PG as default, if I need to store JSON, I’ll go for JSONB in PG. Same for GeoJSON, I’ll use PostGIS, etc.
Thought it doesn’t support sharding, but there’s other way to get this mechanism in SQL.
I’ve used DynamoDB to act like an event store, same goes for Firebase.
At work we’ve a MongoDB database, with Rails backend, the thight coupling beteween MongoDB and the models make the migration task almost impossible, or we’ll have to cut the development of any feature during 6 months or more…