> a way of merging multiple HyperLogLogs together and still have a reasonable cardinality estimate? HLLs can be merged together with the same sort of result a

> a way of merging multiple HyperLogLogs together and still have a reasonable cardinality estimate? HLLs can be merged together with the same sort of result as if you ran the whole set over the same HLL. The merge can't do better than that, but it can do that even if you mess around a bit with the HLL sizes (like, HIVE-18079 is fixing p=14 bits to be p=10 accuracy by default, while retaining old HLLs as-is).