I make backups to a Rsync daemon.
Each backup is made to an empty directory, using 20 --link-dest parameters, so that already uploaded files are hard-linked.
I made some tests adding --fuzzy --fuzzy so that non-existing files can use a base-file, when possible.
It works perfectly, really good thing to find a base-file within link-dest directories.
However, it really slows down the backup process, even when it's not needed.
For example, it terribly impacts performance even for a backup with no new files at all, for which all files will be hard-linked.
Here are some backup durations for a backup with no new files :
- without -yy : 9 seconds
- with -y : 9 seconds
- with --y and 1 --link-dest : 11 seconds
- with --y and 8 --link-dest : 22 seconds
- with --y and 20 --link-dest : 37 seconds
My question is then, could it be possible for --fuzzy --fuzzy (-yy) to only work for files which really need to be uploaded ?
So to avoid working for files which will simply be hard-linked.
Would such a modification only impact daemon side ? (would be great)
Thank you very much !
Created attachment 13748 [details]
Do fuzzy only when needed
Here's a patch for this issue. Thx !
This patch still works as expected with rsync 3.1.3.
Patch moved : https://github.com/WayneD/rsync/pull/1