Any resource to learn SQL will do, there are minor syntax differences between databases but they are not important. And there is not difference to learn SQL for big data, machine learning or any other purpose
PostgreSQL does have a lot of options, and this includes indexing on binary serialized JSON data, that can help a lot depending on what you want to do. PLV8 is also incredibly interesting.
As an aside, I tend to find that if I can do something in a Docker container (locally), I will do it in a container. This lets you spin everything up, and down and cleanup and retry without leaving remnants of stuff behind on your host environment (desktop). The only gotcha is don't use volume mounts for your db data if you're using docker desktop for windows/mac, it does not perform well.
AFAIK ML doesn't really care about the source of data so DB is irrelevant. The conjunction of DB and ML in your question makes me wonder if you're sort of confusing the two. AFAIK they are totally independent, though some DBs come with some data mining tools, which is ML eg. MSSQL (you can prob download an eval version for nothing. It's time limited though, 6 months or so).
But to repeat AFAIK ML and DBs are entirely different things.
If you want to grab a DB, well I've no experience with postgres but I've heard a lot that's good, so perhaps start here.
I don't understand about your configs being in json, nor why that makes it more suitable for mongo. If you need to store semistructured blobs of data, relational DBs can do that fine (in BLOB fields), but IIRC postgres supports json, including indexing into individual json subfields. <http://www.postgresqltutorial.com/postgresql-json/> but to repeat, I've never used postgres so speak to someone else first who has.
If you can do ML as it seems you are, install postgres and just get started on this tutorial stuff here - best of luck!