Spark re-write without clearing wrong output when there is an error during writing Parquet

Question

When I use Spark to write Parquet data to the output path, I found the following things from Spark UI:

1. There are two save() jobs. First one is fail