Based on lag hunting samples using Spark, I've found this:  We need to make searching async so it's not holding up the main thread.