For me the most difficult part are the cloud databases and the Big Data variants of the Databases that use a SQL layer for processing queries. Trying to fit a Relational paradigm to a Columnar or even cloud DB is very difficult. Many times they say its auto scale and auto optimised but many a times we do not have a control over the performance of the Query like we have with Index and DB hints.

Hence I feel relational databases with optimisations available to DBAs and Database developers is a much better option at least we know how to tame a beast then these hybrid solutions.

Another issue is the lack of standardisation while implementing SQL standards and how they implement stored procedures and vanilla data types like Json, xml or variant. Once you go past simple queries to more advanced Data Analytics stuff I feel SQL is not the best option and yet you find many BI teams trying to cram their Data Analytics initiatives more towards Database queries and Window functions rather using a more efficient options.