Why are new columns added to parquet tables not available from glue pyspark ETL jobs?

We've been exploring using Glue to transform some JSON data to parquet. One scenario we tried was adding a column to the parquet table. So partition 1 has columns [A] and partition 2 has columns [A...