[SPARK-23395] Add an option to return an empty DataFrame from an RDD generated by a Hadoop file when there are no usable paths

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Incomplete
• Affects Version/s: 2.2.0, 2.2.1
• Component/s: Input/Output
• Labels: DataFrame HadoopInputFormat RDD bulk-closed
• Assignee: Unassigned
• Reporter: Jens Rabe
• Created: 12/Feb/18 12:48
• Updated: 21/May/19 04:12
• Resolved: 21/May/19 04:12

Description

When using file-based data from custom formats, Spark's ability to use Hadoop's FileInputFormats is very handy. However, when the path they are pointed at contains no usable data, they throw an IOException saying "No input paths specified in job".

It would be a nice feature if the DataFrame API somehow could capture this and return an empty DataFrame instead of failing the job.