When we're building the Document to be indexed, the FacetConfig initialization and it's org.apache.lucene.facet.FacetsConfig.build(Document) require quite some memory.
We should investigate how to avoid creating a new FacetConfig instance for each document, avoid it unless faceting is used, and see if it can be used more efficiently when faceting is being used.

It looks like the FacetConfig implementation was designed to be threadsafe and configured with index specific options, so it seems the developer's intention was for it to be created only once per index.