Found by the acceptance test (#418), testing rclone destinations on next at d747dde.
Steps to reproduce
Use an rclone destination whose remote has a server-side move (local and WebDAV both do).
Kill vaultik snapshot create with kill -9 while a blob uploads, then run it again to completion.
Run vaultik prune --force, then vaultik remote info.
What happened
The killed upload leaves blobs/xx/yy/HASH-RANDOM.partial on the destination; on the local remote it was 21 MB of a 35 MB blob. The next backup ignores it and completes correctly. prune never deletes it, however old: only remote nuke calls DeletePartialUploads (internal/vaultik/prune.go). remote info reports no orphaned data, because its listing skips .partial names. So each killed upload can leave up to blob_size_limit (default 10 GB) that the user is never shown. The README says such a file "is ignored and can be deleted", but on an rclone remote that takes the rclone tool and the hashed path. A file:// destination behaves the same, by code trace.
What should happen
prune deletes a .partial object once it is older than prune_grace_period, as it does an unreferenced blob. remote info counts such objects.
Not blocking: nothing breaks, but storage leaks unseen. Worth fixing before 1.0.
Model: opus-5-5
Found by the acceptance test (https://git.eeqj.de/sneak/vaultik/issues/418), testing rclone destinations on `next` at `d747dde`.
**Steps to reproduce**
- Use an rclone destination whose remote has a server-side move (local and WebDAV both do).
- Kill `vaultik snapshot create` with `kill -9` while a blob uploads, then run it again to completion.
- Run `vaultik prune --force`, then `vaultik remote info`.
**What happened**
The killed upload leaves `blobs/xx/yy/HASH-RANDOM.partial` on the destination; on the local remote it was 21 MB of a 35 MB blob. The next backup ignores it and completes correctly. `prune` never deletes it, however old: only `remote nuke` calls `DeletePartialUploads` (`internal/vaultik/prune.go`). `remote info` reports no orphaned data, because its listing skips `.partial` names. So each killed upload can leave up to `blob_size_limit` (default 10 GB) that the user is never shown. The README says such a file "is ignored and can be deleted", but on an rclone remote that takes the `rclone` tool and the hashed path. A `file://` destination behaves the same, by code trace.
**What should happen**
`prune` deletes a `.partial` object once it is older than `prune_grace_period`, as it does an unreferenced blob. `remote info` counts such objects.
Not blocking: nothing breaks, but storage leaks unseen. Worth fixing before 1.0.
Model: opus-5-5
clawbot
self-assigned this 2026-11-19 08:03:26 +01:00
prune deletes .partial objects under metadata/ and blobs/ once older than prune_grace_period; remote info counts them on their own line and in --json. The storage interface's DeletePartialUploads became ListPartialUploads, so remote nuke now also names any .partial object it could not delete.
Model: opus-5-5
PR: https://git.eeqj.de/sneak/vaultik/pulls/474 (branch `issue-466-prune-partial-uploads`).
`prune` deletes `.partial` objects under `metadata/` and `blobs/` once older than `prune_grace_period`; `remote info` counts them on their own line and in `--json`. The storage interface's `DeletePartialUploads` became `ListPartialUploads`, so `remote nuke` now also names any `.partial` object it could not delete.
Model: opus-5-5
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Found by the acceptance test (#418), testing rclone destinations on
nextatd747dde.Steps to reproduce
vaultik snapshot createwithkill -9while a blob uploads, then run it again to completion.vaultik prune --force, thenvaultik remote info.What happened
The killed upload leaves
blobs/xx/yy/HASH-RANDOM.partialon the destination; on the local remote it was 21 MB of a 35 MB blob. The next backup ignores it and completes correctly.prunenever deletes it, however old: onlyremote nukecallsDeletePartialUploads(internal/vaultik/prune.go).remote inforeports no orphaned data, because its listing skips.partialnames. So each killed upload can leave up toblob_size_limit(default 10 GB) that the user is never shown. The README says such a file "is ignored and can be deleted", but on an rclone remote that takes therclonetool and the hashed path. Afile://destination behaves the same, by code trace.What should happen
prunedeletes a.partialobject once it is older thanprune_grace_period, as it does an unreferenced blob.remote infocounts such objects.Not blocking: nothing breaks, but storage leaks unseen. Worth fixing before 1.0.
Model: opus-5-5
PR: #474 (branch
issue-466-prune-partial-uploads).prunedeletes.partialobjects undermetadata/andblobs/once older thanprune_grace_period;remote infocounts them on their own line and in--json. The storage interface'sDeletePartialUploadsbecameListPartialUploads, soremote nukenow also names any.partialobject it could not delete.Model: opus-5-5