Clone (not copy) data chunks when extracting from uncompressed chunks

[HaleTom]

I made my chunk store uncompressed via:

  "store-options": {
    "/btrbk/ssh-in/desync-store.home-var/": {
      "uncompressed": true
    }

My hope was to save disk space (the btrfs volume itself uses compression): that when extracting, data, the extracted file would be cloned from BOTH any uncompressed chunks in the store AND other seeds, taking up a trivially small amount of space.

However, this indicates that only seed file data is cloned, uncompressed chunks are ignored:
https://github.com/copilot/share/0073513a-01a4-8002-8813-260884952024

There may be file system fragmentation implications in a store with very small chunks -- so cloning chunks would best determined by the user.

Would you consider an option --clone-chunks for cloning any uncompressed chunks?

The the user could decide that the average chunk size is sufficiently large to clone the underlying file system extents instead of copying the data in them. This would be especially useful for chunks in the megabyte size range.

[folbricht]

Just to make sure I understand it correctly. The ask is that given a local+uncompressed chunk store, running extract would reflink the chunks to the appropriate position in the file directly from chunks in the store, rather than copy, with the goal of saving on I/O and space?

desync does use reflinking on filesystems that support it, but not directly from a store at this point. It's a narrow use-case given the number of conditions that need to be in place but it's valid. Not quite sure how complex that'd be to implement.