Details
-
Improvement
-
Status: Resolved
-
Minor
-
Resolution: Fixed
-
2.4.0
-
In Spark 3.0, when empty partitions are written to a CSV, JSON or text data source, they no longer produce an empty file, but instead produce no file at all.
Description
Text based datasources like CSV, JSON and Text produces empty files for empty partitions. This introduces additional overhead while opening and reading such files back. In current implementation of OutputWriter, the output stream are created eagerly even no records are written to the stream. So, creation can be postponed up to the first write.