My team has encountered this known issue with rollup_search terms aggregations which means that results can exclude top terms when sorting by document count. This can lead to confusing and inaccurate results. For example we have found when aggregating on a certain field the correct 2nd and 3rd terms are missing (verified by comparing this to our primary data store).
Increasing the size attribute helps increase the accuracy of the results, but knowing what value to set it to is difficult. For example in one case, we found to get an accurate 8th result for a top 10, we needed to increase the aggregation size from
We are currently using Elasticsearch 7.16.2 and based on the changelog this issue doesn't appear to have been fixed in 7. It would be useful to know if this issue has been fixed in ES 8 and if not is there a plan to address it?