apache / apache/datafusion-comet
Consider using single spill file for multiple partitions
Open
area:shuffle
enhancement
performance
- Dominant language
- Scala
- Stars
- 1.3k
- Forks
- 373
- Avg merge
- 2d 4h
- Merged PRs (30d)
- 198
Description
### What is the problem the feature request solves?
When spilling occurs in native shuffle, Comet writes one spill file per partition. These files are then merged into the final file when shuffle completed.
Gluten combines spill data from multiple partitions into a single spill file. The data is ordered by partition. This results in fewer file handles and fewer metadata ops, which could be a benefit when using EBS for shuffle data.
### Describe the potential solution
_No response_
### Additional context
_No response_
Contributor guide
Assessment
This issue has not been assessed yet.