Somehow Spark also ended up converging towards yet another SQL engine. But I‘ve seen that you need to optimize the physical layout quite a bit (partitions etc) for your use case. Other approaches are much more flexible for that.