that said, I agree that approximate functions should be part of a modern database system. Redshift has approximate count distinct (based on hyperloglog) and approximate percentiles (based on quantile summaries)