[BEAM-13010] Delete orphaned files

• Type: Bug
• Status: Open
• Priority: P1
• Resolution: Unresolved
• Affects Version/s: 2.34.0
• Component/s: io-py-files
• Assignee: Unassigned
• Reporter: David
• Created: 06/Oct/21 14:23
• Updated: 04/Jun/22 22:33

Description

Until version 2.33.0 of Apache Beam, (tested with a Python streaming pipeline consuming events from PubSub and writing them into GCS), some files were being deleted from the temporary folder before being moved to the destination. This was the original issue: 

https://issues.apache.org/jira/browse/BEAM-12950

In version 2.34.0 we applied a temporary workaround to be sure that no data is dropped. Instead of deleting the orphaned files, we just log them:

https://github.com/apache/beam/pull/15576

Most probably the root cause of the missing event was that we were removing files at an erroneous time. We need to delete orphaned files in a subsequent step (after we're sure that there won't be retries). 

Once the original issue is fixed and the orphaned files are deleted at the correct time, we should remove the decorator of the unit test skipped in the Pull Request above.  

Issue Links

• is caused by: BEAM-12950 Missing events when using Python WriteToFiles in streaming pipeline