Thanks for your replay.
So to resume the best way how I understand get accurate results not approximate is to increasing "shard_size".
And the last things:
- From your experience what is the best way to calculate "shard_size" size. For example instance is 32 Memory (GiB).
- Now for example we have about 25 millions records in one index and use "SUM" query like wrote on first post. If we will split to the smaller index (daily index) will it be a big impact to get the better accuracy of the results?
Thanks.