Some is a little rough around the edges but it has extremely quickly replaced much more complicated setups for me. It's fast and very simple and pairs beautifully with having a bunch of data just in parquet files.
Every group by boils down to be columnar. When doing analysis you normally do a lot of grouping/slicing/dicing. True, traditional DBs can do this and more modern one will support alternative storage engines and index modes to support that kind of queries.