I’m experimenting kopia with rclone to try to backup data to 1fichier, a very cheap file hosting provider that rclone can talk with: https://rclone.org/fichier/
here is the result:
kopia repository create rclone --remote-path 1fichier:/din/kopia/test
Enter password to create new repository:
Re-enter password for verification:
Initializing repository with:
block hash: BLAKE2B-256-128
encryption: AES256-GCM-HMAC-SHA256
splitter: DYNAMIC-4M-BUZHASH
kopia: error: cannot initialize repository: unable to write format blob: unable to write format blob: MOVE kopia.repository.f-5577006791947779410: 403, try --help
I checked on 1fichier console that the file was created in the right place.
For what it’s worth, I used the same remote to do the same thing with restic and it worked.
I tried to run kopia with rclone to b2 backblaze and it works fine. So the issue seems to be storage provider specific. Maybe I should talk to the rclone guys. But can you think of a specific feature that kopia needs and that 1fichier could not provide ?
I use kopia 0.7.3 with rclone 1.53.3-1+b2 on debian bullseye
In order to bypass the MOVE, I created a local repo that I synched to its final online destination (1fichier) with rclone.
I could then get a connection with:
So the lack of MOVE would be a problem with current implementation, because move is used for atomic upload - Kopia will first upload a file named something.tmp and when done it will rename/move it to something.f. Only files ending with .f are treated as fully uploaded for safety reasons (in case the client died half-way through we don’t want to rely on partial contents).
Somebody could try to add an option in PutBlobInPath in webdav_storage.go to do direct upload to an .f file bypassing MOVE, but that would probably be unsafe - I’m not sure how rclone would handle atomicity of webdav uploads. If rclone webdav (which kopia uses as its abstract API) is translated as atomic upload to the 1fichier backend which is also atomic, this will in fact work and be safe.