Apache Druid

first thing i would consider is: do you need to actually store the theta sketches? if you have the input stored as its own column, you can compute the sketches at query time.

you only need to store sketches if you're using rollup and are also not retaining the field that would be the input to the sketch

if you do determine that you need to store them, i'm not sure if you'd be able to squish it below an avg of 8 bytes. i think that's about as small as one can get given the preamble

Well, we do not to store the fields used as input to sketches, so we definitely need to store the sketches.