I am using Dataflow to process a Shapefile with about 4 million features (about 2GB total) and load the geometries into BigQuery, so before my pipeline starts, I extract the