lakeFS icon indicating copy to clipboard operation
lakeFS copied to clipboard

Use bulk-delete API for recursive delete in lakeFSFS

Open arielshaqed opened this issue 3 years ago • 2 comments

Measurements indicate that this can speed up the cleanupJob operation on FileOutputCommitter (which uses recursive delete) by 7x.

arielshaqed avatar Sep 19 '22 09:09 arielshaqed

Plan

Be more like (modern) S3AFileSystem, which uses a queue + separate thread for deletion.

arielshaqed avatar Sep 19 '22 09:09 arielshaqed

Implementation

Bulk-delete makes it really hard to report errors accurately. Instead, we shall report the first bulk of errors and stop. S3AFileSystem appears to do something similar.

Additionally, should add a config flag to turn off bulk deletion, for when things really start confusing the user! 👻

arielshaqed avatar Sep 19 '22 09:09 arielshaqed