The current approach has two problems. First, it forces the stream to buffer one which breaks cache coherency. Second, it breaks the scanner as it is today. An empty file would emit zero batches and the resequencer would see it as a skip in the fragment index and fail.
The current approach has two problems. First, it forces the stream to buffer one which breaks cache coherency. Second, it breaks the scanner as it is today. An empty file would emit zero batches and the resequencer would see it as a skip in the fragment index and fail.
Reporter: Weston Pace / @westonpace
Note: This issue was originally created as ARROW-12920. Please see the migration documentation for further details.