1. 2

I wasn’t going to submit this originally (the cardinality aggregation landed a few weeks ago), but since the Redis article was submitted, thought this might deserve a mention too.

The ES implementation is basically the Google HLL++ so it uses linear counting unlike Redis. Some things were tweaked, such as using a hash table instead of a sorted list for linear counting (e.g. higher memory overhead, so it switches to HLL faster, but gives a non-negligible speed boost).