I think facet counts at scale are probably best handled by Elasticsearch or Solr, but I'd be delighted if a PostgreSQL expert would step in and convince me otherwise based on some of the amazing work that went into PostgreSQL 10.