optimise deletion #6353
Description
What would you like to be added:
Optimise the deletion.
Ex.: slices where refs < 0 may need to be stored separately, since under large-scale filesystem, a global scan performs poorly. The database previously optimized the refs field with an index (#3842), but that doesn’t help in the KV engine.
Why is this needed:
The complete deletion workflow involves many states and processes, and in large-scale file system, every single stage can potentially become a bottleneck.
https://juicefs.com/docs/community/faq/#how-does-juicefs-asynchronous-deletion-work