The Lucene metric is segmented. In short, only a subset of the index is read from each segment. And sub-editors are real readers who work directly on a segment (one segment => one segment reader). And the IndexReader that the client uses is an aggregated implementation that uses IndexReader to do the actual work.
source share