Improve cleanup of zombie temporary files in WriteFiles
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 2d 2h
- Merged PRs (30d)
- 205
Description
See user issue in https://stackoverflow.com/questions/47113773/dataflow-2-1-0-streaming-application-is-not-cleaning-temp-folders
For windowed writes, the proper solution is probably to put temp files into finer-grained directories, e.g. sharded by date or hour, and somehow clean them up by globbing and deleting the entire directory when the watermark goes past said date. It's complicated by late data and multiple trigger firings, of course.
Imported from Jira [BEAM-3145](https://issues.apache.org/jira/browse/BEAM-3145). Original Jira may contain additional context.
Reported by: jkff.
Contributor guide
Research direction
Start by reading the WriteFiles implementation and the linked Stack Overflow report, then consult the imported Jira issue BEAM-3145 for additional context. The issue is done when windowed writes no longer leave zombie temporary files while accounting for late data and multiple trigger firings.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- java
- Domain
- data-engineering, stream-processing
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100