[BEAM-3484] HadoopInputFormatIO reads big datasets invalid

• Type: Bug
• Status: Resolved
• Priority: P3
• Resolution: Fixed
• Affects Version/s: 2.3.0, 2.4.0
• Fix Version/s: 2.5.0
• Component/s: io-java-hadoop-format
• Assignee: Alexey Romanenko
• Reporter: Lukasz Gajowy
• Created: 16/Jan/18 14:22
• Updated: 16/May/20 13:22
• Resolved: 19/Apr/18 20:05

Description

For big datasets HadoopInputFormat sometimes skips/duplicates elements from database in resulting PCollection. This gives incorrect read result.

Occurred to me while developing HadoopInputFormatIOIT and running it on dataflow. For datasets smaller or equal to 600 000 database rows I wasn't able to reproduce the issue. Bug appeared only for bigger sets, eg. 700 000, 1 000 000. 

Attachments:
 - text file with sorted HadoopInputFormat.read() result saved using TextIO.write().to().withoutSharding(). If you look carefully you'll notice duplicates or missing values that should not happen

 - same text file for 600 000 records not having any duplicates and missing elements

• link to a PR with HadoopInputFormatIO integration test that allows to reproduce this issue. At the moment of writing, this code is not merged yet.

Issue Links

• links to: GitHub Pull Request #5166
• links to: PR with HadoopInputFormatIOIT to reproduce the issue