As for now, we mimicked the Elasticsearch non blocking infrastructure but we don't have a non blocking Lucene.

Having a non blocking layer on top of Lucene would probably be something we should do at some point. If we decide against it, we can probably remove some of the existing additional complexity.