Closed
Description
Description
We have had this longstanding issue that running a top-level filter aggregation is often (not always, e.g. not if you need the total hit count too) a performance bug, as Elasticsearch would iterate over docs that match the query and would then check each of these docs against the filter of the filter aggregation. It's usually a better idea to put the filter directly into the query, where it can be used to drive iteration if appropriate.
We could do this automatically by implementing the LeafCollector#competitiveIterator API that was introduced for dynamic pruning, in order to automatically start intersecting the query with the filter of the filter aggregator once enough hits have been counted.