Improving query performance in ElasticSearch at scale