"good performance" from an RDB compared to a special purpose KV store interface is a very fuzzy concept indeed
nothing is ever as fast as what is written specific for a task, unless it has a good compiler and the language is simple enough it doesn't take so much resources it destroys the benefit for JIT (you could make JIT compiled search functions out of sql with go from a reasonably indexed data source that probably would beat any SQL engine)
i find it sad that i have to explain these simple physical laws of computer language processing to people who think they understand how computers work and think that sql is *always* better
no, it's not, it's lazy better, it's not faster better