The parent Sort must only use fields from the parent documents; sorting by field in the child documents is not supported.
You should only use this collector if one or more of the clauses in the query is a ToParentBlockJoinQuery. This collector will find those query clauses and record the matching child documents for the top scoring parent documents.
Multiple joins (star join) and nested joins and a mix of the two are allowed, as long as in all cases the documents corresponding to a single row of each joined parent table were indexed as a doc block.
For the simple star join you can retrieve the TopGroups instance containing each ToParentBlockJoinQuery's matching child documents for the top parent groups, using #getTopGroups. Ie, a single query, which will contain two or more ToParentBlockJoinQuery's as clauses representing the star join, can then retrieve two or more TopGroups instances.
For nested joins, the query will run correctly (ie, match the right parent and child documents), however, because TopGroups is currently unable to support nesting (each group is not able to hold another TopGroups), you are only able to retrieve the TopGroups of the first join. The TopGroups of the nested joins will not be correct. See org.apache.lucene.search.join for a code sample. @lucene.experimental