Refuse a backup destination an unplugged volume left empty (closes #239)

Every backup now writes vaultik-destination.txt at the destination
root. While the local index holds snapshots, a destination without it
is not present: snapshot create refuses before writing, prune and
snapshot purge fail, and snapshot list and snapshot remove warn.
remote nuke deletes the marker last. The file:// backend no longer
creates its destination directory and checks for it on every write;
that protects a directory on a removable volume, not the mount point
itself.

Tests back up into existing directories; fixtures standing for a
written destination carry the marker.

Judgement call: the marker is rewritten on every backup.
Deviation: for the complexity limit, snapshot name selection left CreateSnapshot and the --force/--json test moved into confirmRemoveSnapshot, both unchanged.

Model: opus-5-5

ARCHITECTURE.md

@@ -54,10 +54,12 @@ The database tracks five primary entities and their relationships:
 
 #### File (`database.File`)
 Represents a file, directory, or symlink in the backup system. Stores metadata needed for restoration:
-- Path, source_path (for restore path stripping), mtime
+- Path, mtime
 - Size, mode, ownership (uid, gid)
 - Symlink target (if applicable)
 
+It also stores `source_path`, the source directory the scan found it under, made absolute and with symlinks resolved. Restore does not read it.
+
 #### Chunk (`database.Chunk`)
 A content-addressed unit of data. Files are split into variable-size chunks using the FastCDC algorithm:
 - `ChunkHash`: SHA256 hash of chunk content (primary key)
@@ -82,9 +84,11 @@ The final storage unit uploaded to S3. Contains many compressed and encrypted ch
 Blob creation process:
 1. Chunks are accumulated (up to MaxBlobSize, typically 10GB)
 2. As each chunk is added, its uncompressed bytes are fed to a running SHA-256
-3. Concurrently, the same bytes are compressed with zstd, then encrypted with age (recipients configured in config), and streamed to storage
+3. Concurrently, the same bytes are compressed with zstd, then encrypted with age (recipients configured in config), and written to a temporary file
 4. On finalize, the blob's name is the double SHA-256 of the uncompressed contents — `hex(SHA256(SHA256(...)))` — not a hash of the compressed, encrypted bytes
-5. Uploaded to `blobs/{hash[0:2]}/{hash[2:4]}/{hash}`
+5. The finished file is uploaded to `blobs/{hash[0:2]}/{hash[2:4]}/{hash}` and then deleted
+
+A backup needs free temporary space, because each blob is written whole to a temporary file before it is uploaded (up to about `blob_size_limit`; an rclone destination that cannot stream uploads needs about twice that) and the metadata export writes copies of the local index. Temporary files go to `$TMPDIR` (default `/tmp`); with `TMPDIR` unset, SQLite writes one of those copies to `/var/tmp`.
 
 #### BlobChunk (`database.BlobChunk`)
 Maps chunks to their position within blobs:
@@ -307,6 +311,8 @@ Repository interfaces:
 
 ```
 CreateSnapshot(opts)
+    │
+    ├─► markDestination()   // Refuse a destination without the marker, then write it
     │
     ├─► PruneDatabase()   // Critical: avoid dedup errors
     │
@@ -363,6 +369,8 @@ CreateSnapshot(opts)
 
 ```
 bucket/
+├── vaultik-destination.txt          # Marker written by every backup
+│
 ├── blobs/
 │   └── {hash[0:2]}/
 │       └── {hash[2:4]}/
@@ -374,6 +382,15 @@ bucket/
         └── manifest.json.zst        # Blob list (for pruning/verification)
 ```
 
+While the local index holds snapshots, a destination without
+`vaultik-destination.txt` is treated as not present
+(`checkDestinationMarker` in `internal/vaultik/destination_marker.go`):
+`snapshot create` refuses to write to it, and listing it fails as for a
+destination that cannot be listed. An unplugged volume's empty mount point
+is therefore never taken for a destination whose snapshots are all gone.
+The `file://` backend never creates its destination directory, so a write
+to a missing one fails.
+
 The `{remote-key}` directory name is a one-way double SHA-256 hash of the human
 snapshot ID, so the human ID (hostname, snapshot name, timestamp) is never
 written to the store as a directory name. See

README.md

@@ -24,6 +24,10 @@ grep 'public key' vaultik_backup_private_key.txt
 vaultik config set age_recipients.0 age1YOUR_PUBLIC_KEY_HERE
 vaultik config set storage_url "file:///Volumes/usbstick/mybackup"
 
+# create the destination directory; vaultik never creates it, so a backup
+# to a volume that is not plugged in fails instead of filling the disk
+mkdir /Volumes/usbstick/mybackup
+
 # macOS only: grant your terminal app Full Disk Access first
 # (System Settings → Privacy & Security → Full Disk Access), otherwise
 # the backup will abort with a permission error on protected directories
@@ -200,8 +204,10 @@ local index or the destination store. `config`, `database delete`,
 ### stdout and stderr
 
 Log output — everything from `--verbose` and `--debug`, and every
-warning and error the logger emits — goes to **stderr**. stdout carries
+warning and error the logger emits — goes to **stderr**, and so does the
-the output you asked for: tables, and the documents produced by `--json`.
+startup banner. stdout carries the output you asked for: tables, the
+documents produced by `--json`, `config get` values, and completion
+scripts.
 
 This means `vaultik snapshot list --verbose > out.txt` captures the
 listing and leaves the diagnostics on your terminal. To capture both,
@@ -284,13 +290,19 @@ needed.
   remove orphaned blobs from remote storage. By default keeps only the latest
   snapshot per name; use `--keep-newer-than` for a rolling window.
 * `--keep-newer-than <duration>`: With `--prune`, keep snapshots newer than
-  this duration instead of only the latest (e.g. `4w`, `30d`, `6mo`, `1y`)
+  this duration instead of only the latest (e.g. `4w`, `30d`, `6mo`, `1y`).
+  Without `--prune`, or with a value that is not a valid duration, the
+  command fails before the backup starts.
+* Fails before it writes anything when a `file://` destination directory
+  does not exist, or when the local index holds snapshots and the
+  destination has no `vaultik-destination.txt` (see
+  [remote storage layout](#remote-storage-layout))
 
 **`snapshot list`**: Show every snapshot known to this host — the union
 of the local index and the backup destination store — with timestamps
 and three sizes per snapshot (compressed remote size; total
-uncompressed chunk size; size of chunks newly referenced by that
+uncompressed chunk size; size of the chunks that snapshot stored that
-snapshot).
+were not stored before, recorded when its backup ran).
 
 Listing the destination store does **not** require the age secret key,
 so it works in vaultik's intended configuration, where the backed-up
@@ -303,8 +315,8 @@ only the hash, so hostname and snapshot name exist solely in the local
 index and in the encrypted per-snapshot database. Snapshots found only
 on the destination store are therefore listed as
 `<remote only:<abbreviated remote key>>` and show `<remote only>` in
-the uncompressed and "new chunk" columns, which can only be computed
+the uncompressed and "new chunk" columns, which come only from the
-from the local index. Their timestamp and compressed size are real,
+local index. Their timestamp and compressed size are real,
 read from the manifest.
 
 Snapshots in the local index with no counterpart on the destination
@@ -313,7 +325,10 @@ invocation that reconciles them.
 
 If the destination store cannot be listed (unmounted volume,
 permission denied, network down), the command warns, falls back to the
-local index alone, and still exits zero.
+local index alone, and still exits zero. It does the same when the
+local index holds snapshots and the destination has no
+`vaultik-destination.txt`, as when a volume mounted at the destination
+is unplugged and leaves its empty mount point behind.
 * `--json`: Output in JSON format. Each entry carries `locally_tracked`
   (whether the snapshot is in the local index), `remote_key` (the full
   64-character storage key), and `remote_present` (whether it was seen
@@ -347,16 +362,22 @@ the backup destination store. Blobs are NOT touched — deleting blobs
 requires reading every remaining remote manifest (the destination store
 may hold snapshots this host doesn't know about), which is what
 `vaultik prune` does. On success the command prints the exact `vaultik
-prune` invocation to run as a follow-up. Local row cleanup (files,
+prune` invocation to run as a follow-up. An ID that names no snapshot in
-chunks, blobs the snapshot was the last referrer for) runs
+the local index or on the destination store is an error; with
-automatically. If the destination store is unreachable, the local-DB
+`--local-only`, the snapshot must be in the local index. Local row
-removal still completes and a warning is emitted; rerun `vaultik prune`
+cleanup (files, chunks, blobs the snapshot was the last referrer for)
-once the store is reachable to finish remote cleanup. To wipe everything
+runs automatically. If the destination store is unreachable, or the
-on the destination in one go, use `vaultik remote nuke --force`.
+local index holds snapshots and the destination has no
+`vaultik-destination.txt`, the local-DB removal still completes and a
+warning is emitted; run `vaultik snapshot remove <snapshot-id>` again
+once the store is reachable to remove the snapshot's metadata from it
+(`vaultik prune` does not). To wipe everything on the destination in one
+go, use `vaultik remote nuke --force`.
 * `--local-only`: Skip remote cleanup; only touch the local index
 * `--dry-run`: Show what would be deleted without deleting
 * `--force`: Skip confirmation prompt
-* `--json`: Output result as JSON
+* `--json`: Output result as JSON. Also skips the confirmation prompt, as
+  `--force` does.
 
 **`snapshot restore`**: Restore files from a backup snapshot.
 * Requires `VAULTIK_AGE_SECRET_KEY` environment variable
@@ -378,29 +399,48 @@ manifest on the destination store to compute the still-referenced blob
 set and delete any blob not in that set. Step (3) reads all remote
 manifests — network cost scales with the number of snapshots. `snapshot
 create --prune` runs the same cleanup automatically; this is the
-manual entry point for the same work.
+manual entry point for the same work. If the destination store cannot
+be listed, or the local index holds snapshots and the destination has
+no `vaultik-destination.txt` (an unplugged volume's empty mount point),
+prune fails and deletes nothing. `snapshot purge` does the same. If a
+blob cannot be deleted, the command (and `snapshot create --prune`)
+exits non-zero.
 * `--force`: Skip confirmation prompt
-* `--json`: Output stats as JSON
+* `--json`: Output stats as JSON. Also skips the confirmation prompt, as
+  `--force` does. A blob that cannot be deleted is counted in
+  `blobs_failed`, and the document is printed before the non-zero exit.
 
 **`info`**: Display system configuration, storage settings, encryption
 recipients, and local database statistics.
 
 **`remote info`**: Show storage backend type and location plus detailed
 remote storage inventory: per-snapshot metadata sizes, blob counts, and
-orphaned blob detection.
+orphaned blob detection. A name under `metadata/` that is not a remote
+key is skipped with a warning and is not printed. If a listed
+`manifest.json.zst` cannot be read, or sits under a skipped name, the
+orphaned blob figures are reported as unknown; `--json` gives them as
+`null`, lists the remote key of each unreadable manifest in
+`unreadable_manifests` and counts the manifests under skipped names in
+`skipped_manifest_count`. An unreadable manifest also leaves its
+snapshot's blob count and blob size unknown, `null` in `--json`. A file
+under `blobs/` whose name is not a blob hash is not counted as a blob,
+since `prune` never deletes it.
 * `--json`: Output as JSON
 
-**`remote nuke`**: Delete every snapshot's metadata and every blob from the
+**`remote nuke`**: Delete every snapshot's metadata, every blob and
-backup destination store, leaving the bucket prefix empty. Destructive and
+`vaultik-destination.txt` from the backup destination store, leaving the
-irreversible. This is the single supported way to wipe the entire
+bucket prefix empty. Destructive and irreversible. This is the single
-destination store.
+supported way to wipe the entire destination store. If anything cannot
+be deleted, the command exits non-zero; running it again deletes what is
+left.
 * `--force`: Required to confirm destruction.
 
 **`database delete`**: Delete the local SQLite state database file
 entirely. Remote storage is unaffected; the next backup will do a full
 scan and re-deduplicate against existing remote blobs, and the local
 index will re-bind to the currently configured storage destination.
-Use this after changing `storage_url` to a different destination.
+Use this after changing `storage_url`, or `s3.endpoint`, `s3.bucket` or
+`s3.prefix`, to a different destination.
 * `--force`: Skip confirmation prompt
 
 ---
@@ -414,11 +454,29 @@ object store. Credentials are read from `s3.access_key_id` and
 `s3.secret_access_key` in the config file.
 
 **Local filesystem** (`file:///path/to/backup`): Stores blobs and metadata on
-a local or mounted filesystem. Useful for testing or backing up to a NAS.
+a local or mounted filesystem. Useful for testing or backing up to a NAS. The
+path must be absolute; a relative one is refused when the config loads.
+Percent-escapes in the path are decoded, so `my%20stick` and `my stick` name
+the same directory. The directory must exist before the first backup: vaultik
+never creates it. A directory on a removable volume is gone while the volume is
+not mounted, so a backup then fails instead of writing to the disk under the
+mount point. The mount point itself stays behind when the volume is unplugged:
+there only `vaultik-destination.txt` stops such a backup, and only while the
+local index holds snapshots.
 
 **Rclone** (`rclone://remote/path`): Uses rclone's 70+ supported cloud
 providers. Requires rclone to be configured separately (`rclone config`).
 
+An upload cut off part-way leaves nothing under the object's name on S3, which
+shows an object only once its upload has completed, and on the local filesystem
+backend, which writes a temporary file and renames it into place. Rclone
+remotes with a server-side move (local and sftp among them) are written under a
+temporary name ending in `.partial` and moved into place. Rclone remotes without
+one are written in place: where such a remote shows a file while it is still
+being written, a killed upload can leave a truncated object under its name,
+which a later backup takes for complete. A leftover `.partial` file is ignored
+and can be deleted.
+
 Legacy S3 configuration via `s3.*` fields (endpoint, bucket, prefix, etc.) is
 still supported for backward compatibility. `storage_url` takes precedence if
 both are set.
@@ -435,6 +493,7 @@ For an implementation-level view of the internals — the data model, the
 
 ```
 <bucket>/<prefix>/
+├── vaultik-destination.txt     # Marker written by every backup
 ├── blobs/
 │   └── <aa>/<bb>/<full_blob_hash>
 └── metadata/
@@ -443,6 +502,15 @@ For an implementation-level view of the internals — the data model, the
         └── manifest.json.zst   # Unencrypted blob list (for pruning)
 ```
 
+* `vaultik-destination.txt` holds one fixed line of text, with no host
+  name, path or time. While the local index holds snapshots, a
+  destination without it is treated as not present: `snapshot create`
+  refuses to write to it, `prune` and `snapshot purge` fail, and
+  `snapshot list` and `snapshot remove` warn, as for a destination that
+  cannot be listed. This keeps an unplugged volume's empty mount point
+  from being taken for a destination whose snapshots are all gone. Mount
+  the volume if it is removable; if the destination was emptied on
+  purpose, run `vaultik database delete` and back up again.
 * Blobs are two-level directory sharded using the first 4 hex chars of the blob hash
 * `db.zst.age` is a binary SQLite database (zstd compressed, age encrypted)
   containing all file metadata, chunk mappings, and relationships for the snapshot
@@ -537,7 +605,7 @@ complete annotated example also lives in
 | `s3.*` | | Legacy S3 configuration (endpoint, bucket, credentials) |
 | `exclude` | | Global exclude patterns (applied to all snapshots) |
 | `chunk_size` | `10MB` | Average chunk size for content-defined chunking |
-| `blob_size_limit` | `10GB` | Maximum blob size before splitting. Must be at least four times `chunk_size` (the largest chunk the chunker can emit), otherwise a single-chunk blob could exceed the limit |
+| `blob_size_limit` | `10GB` | Maximum blob size before splitting. Must be at least four times `chunk_size` (the largest chunk the chunker can emit), otherwise a single-chunk blob could exceed the limit. A backup needs free temporary space, because each blob is written whole to a temporary file before it is uploaded (up to about `blob_size_limit`; an rclone destination that cannot stream uploads needs about twice that) and the metadata export writes copies of the local index. Temporary files go to `$TMPDIR` (default `/tmp`); with `TMPDIR` unset, SQLite writes one of those copies to `/var/tmp` |
 | `compression_level` | `3` | zstd compression level (1-19) |
 | `hostname` | system hostname | Hostname used in snapshot IDs |
 | `index_path` | platform data dir | Local SQLite index path |
@@ -645,8 +713,8 @@ Work planned after 1.0. Loosely ordered by priority.
 ## output style
 
 Every command's user-facing output is governed by `internal/ui`, in one
-of two ways. Color is enabled when stdout is a TTY and the `NO_COLOR`
+of two ways. Color is enabled when the stream written to is a TTY and
-environment variable is unset (https://no-color.org/).
+the `NO_COLOR` environment variable is unset (https://no-color.org/).
 
 * **Status, progress, warnings, and errors** go through the `internal/ui`
   message methods below: marker-prefixed, colored on a TTY, and — except
@@ -656,24 +724,26 @@ environment variable is unset (https://no-color.org/).
   `config init`, `config set`, and `database delete`.
 * **The data a command exists to produce** is written plain, with no
   marker and no color, because a marker would corrupt a table or a
-  parsed document. This covers the `version`, `info`, and `remote info`
+  parsed document. This covers the `version`, `info`, `remote info` and
-  reports, the `snapshot list` table, `config get` values, and every
+  `snapshot verify` reports, the `snapshot list` table, `config get`
-  `--json` document. `--quiet` silences the human reports and tables
+  values, and every `--json` document. `--quiet` silences the human
-  (`version`, `info`, `remote info`, `snapshot list`) but never the
+  reports and tables (`version`, `info`, `remote info`, `snapshot
-  machine-consumed `config get` value or the `--json` documents, which a
+  verify`, `snapshot list`) but never the machine-consumed `config get`
-  script depends on. The `database delete` confirmation prompt is also
+  value or the `--json` documents, which a script depends on. The
-  written this way and always shown: it is an interactive exchange the
+  `database delete` confirmation prompt is also written this way and
-  operator must see.
+  always shown: it is an interactive exchange the operator must see.
 
-`internal/ui` writes to stdout; it is the output the user asked for.
+`internal/ui` writes to stdout; it is the output the user asked for. The
-Structured log records are a different thing and go through
+exceptions are the startup banner and the error a failed command ends
-`internal/log`, which writes to stderr (see "stdout and stderr" above).
+with, which go to stderr. Structured log records are a different thing
+and go through `internal/log`, which writes to stderr (see "stdout and
+stderr" above).
 
 Message classes:
 
 | Class | Marker | Alignment | Use for |
 |-------|--------|-----------|---------|
-| Banner | none | column 0 | The startup line printed once per invocation |
+| Banner | none | column 0 | The startup line printed once per invocation, on stderr |
 | Begin | `》` (white) | column 0 | An operation is about to start (present-continuous verb) |
 | Complete | `》` (green) | column 0 | An operation just finished (past-tense verb) |
 | Info | `》` (white) | column 0 | Neutral status update |
@@ -794,7 +864,8 @@ them. We provide:
   in the `Dockerfile` together.
 * `script/release` — cross-compile and publish the release artifacts
   with the pinned `goreleaser`. Refuses a `goreleaser` on `PATH` whose
-  version is not the pinned one, on the same reasoning as `script/lint`.
+  version is not the pinned one, because a different version would build
+  a different release from the same tag.
 * `script/release-snapshot` — the same build with no publishing and no
   tagging, into `./dist`
 * `script/test` — run the test suite by building the `test` phase of
@@ -904,14 +975,14 @@ It is passed to `goreleaser` as `GITEA_TOKEN`. The runner's automatic
 token is deliberately not used: it is not guaranteed to carry release
 write access.
 
-The Go toolchain that compiles the released binaries comes from an
+The Go toolchain that compiles the released binaries is installed by
-`actions/setup-go` step pinned by commit sha, reading its version from
+`script/install-go`, which downloads the version named by `go.mod`
-`go.mod` (currently `1.26.1`, the same version the `Dockerfile` builder
+(currently `1.26.1`, the same version the `Dockerfile` builder stage
-stage pins by digest). `goreleaser` shells out to `go` for every
+pins by digest) and refuses the archive unless its sha256 matches the
+value committed in the script. `goreleaser` shells out to `go` for every
 cross-compile, so without that step the release would either fail
 outright or ship binaries built by whatever unpinned toolchain the
-runner happened to carry — the one unpinned thing in an otherwise
+runner happened to carry.
-hash-pinned release path.
 
 To rehearse the whole build without publishing or tagging anything:
 

TODO.md

@@ -22,6 +22,252 @@ the tag exists and is exercised; what is left is merging `next` to
 
 # Completed Steps
 
+- 2026-10-13: Stopped `remote info` counting a file under `blobs/` that
+  is not a blob as an orphaned blob
+  ([issue #295](https://git.eeqj.de/sneak/vaultik/issues/295)). A stray
+  file such as `.DS_Store` was reported as an orphaned blob and the
+  report advised running `prune`, which skips such a file and deleted
+  nothing. `remote info` now takes its blob list from the same listing
+  `prune` uses, in the table and in `--json`.
+
+- 2026-10-13: Made `snapshot remove` fail for a snapshot that does not
+  exist ([issue #293](https://git.eeqj.de/sneak/vaultik/issues/293)).
+  The command never checked the ID: it reported the snapshot removed
+  from the local database and from the destination store and exited 0,
+  under `--json`, `--dry-run` and `--local-only` too. An ID that names
+  no snapshot in the local index or on the destination store is now an
+  error; with `--local-only`, the snapshot must be in the local index.
+  The command now reports a removal from either side only when one
+  happened, and `remote_removed` is true only when metadata was deleted.
+
+- 2026-10-12: Bound the local index to the destination an `s3.*` config
+  names ([issue #298](https://git.eeqj.de/sneak/vaultik/issues/298)).
+  Only `storage_url` was checked, so after `s3.bucket` or `s3.prefix`
+  changed, the next backup treated every chunk as already stored,
+  uploaded nothing, and completed a snapshot that could not be restored.
+  Without `storage_url`, the local index is now bound to `s3.endpoint`,
+  `s3.bucket` and `s3.prefix`, and a change is refused as a changed
+  `storage_url` is.
+
+- 2026-10-12: Rejected a retention duration longer than about 292 years
+  ([issue #289](https://git.eeqj.de/sneak/vaultik/issues/289)). The
+  number given to `--older-than` or `--keep-newer-than` was multiplied
+  by the length of its unit with no limit, so from `293y`, `15251w`,
+  `3559mo` or `106752d`, or a sum of parts such as `200y200y`, the
+  result wrapped to a negative duration. The cutoff then lay in the
+  future, every snapshot of the name was deleted, the one just made
+  included, and the command exited 0. Such a duration is now rejected
+  as invalid and nothing is deleted.
+
+- 2026-10-08: Made a backup cancelled under `--skip-errors` stop at once
+  ([issue #286](https://git.eeqj.de/sneak/vaultik/issues/286)). After
+  Ctrl-C or SIGTERM, phase 2 treated the cancellation error from each
+  remaining file like an unreadable file: it opened the file, printed an
+  error line for it, counted it as failed and went on to the next one.
+  The run now stops at the first file after the cancellation, and no
+  file is counted as failed because of it.
+
+- 2026-10-08: Made `remote nuke` delete the `.partial` files that
+  uploads cut off part-way leave on the destination store
+  ([issue #281](https://git.eeqj.de/sneak/vaultik/issues/281)). The
+  file and rclone listings skip such a file, so the command left it in
+  place and still reported the store empty. It now removes them under
+  `metadata/` and `blobs/` as its last step.
+
+- 2026-10-08: Made a local index error while recording a directory or
+  symlink stop a backup under `--skip-errors`
+  ([issue #284](https://git.eeqj.de/sneak/vaultik/issues/284)). Phase 2
+  only records such an entry in the local index, since it has no data
+  to open or read, but an error doing so was skipped like an unreadable
+  file. The snapshot then completed without the entry, and a restore
+  did not recreate it. The error now stops the backup, as it does
+  without the flag.
+
+- 2026-10-08: Counted a file that a backup could not store as failed
+  ([issue #280](https://git.eeqj.de/sneak/vaultik/issues/280)). A file
+  that phase 1 counted and phase 2 could not open, because it was
+  unreadable under `--skip-errors` or removed in between, was reported
+  in the summary as unchanged with its bytes as backed up, and the
+  `snapshots` row's `file_count` and `total_size` included it. The
+  summary now counts it as failed, and its data total and the row leave
+  it out.
+
+- 2026-10-08: Made `remote info` report a snapshot's blob count and
+  blob size as unknown when its manifest cannot be read
+  ([issue #272](https://git.eeqj.de/sneak/vaultik/issues/272)). The
+  orphan figures were already unknown in that case, but the snapshot's
+  row still gave 0 blobs and 0 B, in the table and in `--json`. The row
+  now reads `unknown` and `--json` gives `null`. A directory with no
+  manifest still shows 0, since the orphan figures count its blobs as
+  orphaned.
+
+- 2026-10-08: Kept the progress line of a snapshot with more than one
+  path within 100%
+  ([issue #271](https://git.eeqj.de/sneak/vaultik/issues/271)). The
+  bytes and files processed counted every path, but the totals they
+  were divided by held only the current path's, so a second path
+  smaller than the first showed more than 100% and an ETA of `unknown`.
+  The totals now add up over the paths scanned so far. The rate behind
+  the ETA restarts when each path's scan phase ends, so that phase does
+  not lower the rate while the path is processed; until then the rate
+  is the previous path's.
+
+- 2026-10-08: Made a second `snapshot create` of one name succeed when
+  it starts in the same second as the first
+  ([issue #270](https://git.eeqj.de/sneak/vaultik/issues/270)). The
+  timestamp in a snapshot ID is in whole seconds, so the second run got
+  the first run's ID and failed with `UNIQUE constraint failed:
+  snapshots.id`. When the local index already has a snapshot with the
+  ID, the create now waits a second and takes a new timestamp.
+
+- 2026-10-07: Documented that `--json` skips the confirmation prompt of
+  `snapshot remove` and `prune`
+  ([issue #268](https://git.eeqj.de/sneak/vaultik/issues/268)). Both
+  commands delete without asking under `--json`, since a prompt on stdout
+  would break the JSON document, but the help and the README described
+  only `--force` as skipping it. The `--json` help of both commands and
+  their README entries now say so.
+
+- 2026-10-07: Made a symlink whose target cannot be read stop the backup
+  ([issue #269](https://git.eeqj.de/sneak/vaultik/issues/269)). It was
+  left out of the snapshot with only a debug log line, even without
+  `--skip-errors`. It now aborts the run, or with `--skip-errors` is
+  skipped with the `Failed to access` error line that any other entry
+  the scan cannot read gets.
+
+- 2026-10-07: Stopped a killed rclone upload from leaving a truncated
+  object under its key
+  ([issue #266](https://git.eeqj.de/sneak/vaultik/issues/266)). The
+  rclone backend wrote each object straight to its key, so killing an
+  upload to a local or sftp remote left a truncated object there; the
+  next backup found the key with `Stat`, skipped the upload and recorded
+  a snapshot that could not be restored. On a remote with a server-side
+  move, an object is now written under a temporary name ending in
+  `.partial` and moved into place, and listings skip such names. Remotes
+  without one are still written in place.
+
+- 2026-10-07: Made an interrupted command exit 130 and say so
+  ([issue #267](https://git.eeqj.de/sneak/vaultik/issues/267)). Ctrl-C
+  or SIGTERM during `snapshot create`, `snapshot restore` or `snapshot
+  verify` exited 0 with no error line (`snapshot verify --json` exited
+  1, also without one), so a `--cron` run that never finished looked
+  like a success. A command stopped by either signal now exits 130 and
+  prints `interrupted before the command finished` on stderr, under
+  `--cron` and `--json` too.
+
+- 2026-10-07: Corrected documentation, help text and comments that were
+  false about the code
+  ([issue #233](https://git.eeqj.de/sneak/vaultik/issues/233)). A blob
+  is not streamed to storage. The README, `ARCHITECTURE.md` and
+  `config.example.yml` now say a backup needs free temporary space,
+  because each blob is written whole to a temporary file before it is
+  uploaded (up to about `blob_size_limit`; an rclone destination that
+  cannot stream uploads needs about twice that) and the metadata export
+  writes copies of the local index. Temporary files go to `$TMPDIR`
+  (default `/tmp`); with `TMPDIR` unset, SQLite writes one of those
+  copies to `/var/tmp`. Also corrected: the snapshot ID format, what restore
+  reads and how incomplete snapshots are removed in `docs/DATAMODEL.md`,
+  what `source_path` holds, the `index_path` and config file defaults,
+  what `snapshot remove` cleans up, how the release gets its Go
+  toolchain, and the `script/release` and `script/fmt-check` comments.
+
+- 2026-10-07: Cut the time the `internal/vaultik` and `internal/database`
+  tests take ([issue #235](https://git.eeqj.de/sneak/vaultik/issues/235)).
+  Most of the `internal/vaultik` time went to 24 tests that ran one at a
+  time only because they call `log.Initialize`; they now call it before
+  `t.Parallel()`, as the package's other tests do. `TestLargeDatasets`
+  committed each of its 1,500 inserts on its own and now makes them in
+  one transaction. `TestDedupOnlySnapshotRestores` gives its second
+  backup its own snapshot name instead of sleeping past the one-second
+  timestamp in the snapshot ID.
+
+- 2026-10-07: Made two messages say only what is true
+  ([issue #240](https://git.eeqj.de/sneak/vaultik/issues/240)). A config
+  file that others can read was warned about as containing S3
+  credentials even when it set none, as a `file://` config does. The
+  warning now says the file may contain S3 credentials only when
+  `s3.access_key_id` or `s3.secret_access_key` is set, since either may
+  come from a `${...}` reference rather than the file, and otherwise
+  says the file is readable by others. `snapshot purge` against a
+  destination store it could not list gave an error with
+  `listing remote snapshots:` in it twice; the prefix now appears once.
+
+- 2026-10-07: Made `s3.part_size` set the multipart upload part size
+  ([issue #232](https://git.eeqj.de/sneak/vaultik/issues/232)). It was
+  loaded and defaulted but never passed to the S3 client, whose uploader
+  used a fixed 10MiB part. It now reaches the uploader for `storage_url`
+  and for the `s3.*` fields, and a part size S3 refuses, below 5MiB or
+  above 5GiB, `0` included, fails at config load. A blob too large for
+  S3's limit of 10,000 parts at the configured size is uploaded in larger
+  parts. The docs gave the default as `5MB`, which the config file reads
+  as 5,000,000 bytes, below the minimum; they now say `5MiB`.
+
+- 2026-10-07: Made per-name retention work when the hostname contains `_`
+  ([issue #230](https://git.eeqj.de/sneak/vaultik/issues/230)). A
+  snapshot ID is `hostname_name_timestamp`, and the name was read as
+  everything between the first and the last `_`, so with
+  `hostname: my_host` the name `home` came out as `host_home`.
+  `snapshot purge --keep-latest --snapshot home` then printed "No
+  snapshots to delete", and `snapshot create --prune` purged nothing
+  without a message. The name is now read using the hostname the
+  `snapshots` table stores with each snapshot, cut at its first `.` as it
+  is in the ID.
+
+- 2026-10-07: Made `remote info` stop reporting a snapshot's blobs as
+  orphaned when its manifest cannot be read, and stop printing raw
+  names from under `metadata/`
+  ([issue #228](https://git.eeqj.de/sneak/vaultik/issues/228)). A
+  manifest it failed to read was skipped, so that snapshot's blobs were
+  counted as orphaned and the report advised running `vaultik prune`.
+  The orphan figures are now unknown in that case, with no prune
+  advice, and `--json` gives them as `null` with the unreadable remote
+  keys in `unreadable_manifests`. A name under `metadata/` that is not
+  64 lowercase hex characters is now skipped with a warning instead of
+  being printed, control characters included. A manifest under a
+  skipped name is then not read either, so it also leaves the orphan
+  figures unknown, and `--json` counts such manifests in
+  `skipped_manifest_count`. A directory with no manifest in it, as left
+  by an interrupted backup, leaves the figures known.
+
+- 2026-10-07: Made `config set` keep a string that looks like a number
+  ([issue #229](https://git.eeqj.de/sneak/vaultik/issues/229)). It wrote
+  every value unquoted, and `config.Load` reads the file through untyped
+  YAML, so an access key `00112233` loaded as `38043` and a hostname `007`
+  as `7`. A value for a string setting in `config.Config` is now tagged as
+  a YAML string, which the file quotes wherever YAML would read a number or
+  a boolean; other settings are still written unquoted.
+
+- 2026-10-07: Made a backup notice a file rewritten with its size
+  unchanged and a new mtime in the same second as the one in the index
+  ([issue #226](https://git.eeqj.de/sneak/vaultik/issues/226)). The
+  `files` table held mtime in whole seconds and the scanner compared
+  whole seconds, so every later snapshot kept the old content. A new
+  `mtime_nsec` column holds the nanoseconds within the second that
+  `mtime` holds, and the scanner compares the full mtime. A local index
+  created before the change lacks the column and is rebuilt with
+  `vaultik database delete` and a full backup.
+
+- 2026-10-07: Made taking the process-wide lock atomic
+  ([issue #227](https://git.eeqj.de/sneak/vaultik/issues/227)). The lock
+  read `vaultik.pid`, checked whether that PID was alive and then wrote
+  its own, so two writers started together could both pass the check and
+  both run. It is now an `flock` on `vaultik.pid`, held until the run
+  ends; the kernel drops it when the process exits, so a crash leaves no
+  lock behind. A clean exit now empties the file instead of deleting it,
+  because deleting it would let two later runs each lock a different
+  file.
+
+- 2026-10-06: Made `snapshot remove --json` write only its document to
+  stdout when the destination store cannot be reached
+  ([issue #251](https://git.eeqj.de/sneak/vaultik/issues/251)). Its
+  warning that the snapshot's metadata was left on the destination store
+  went to stdout ahead of the document, breaking `| jq` on a command that
+  exited 0. Under `--json` the warning now reaches stderr only, through
+  the logger. The warning, the README and the command's help said
+  `vaultik prune` would finish the cleanup, but `prune` never removes
+  snapshot metadata; they now say to run `vaultik snapshot remove` for the
+  snapshot again once the destination store is reachable.
+
 - 2026-10-06: Made the backup summary and the `snapshots` row count each
   file, byte and upload once
   ([issue #225](https://git.eeqj.de/sneak/vaultik/issues/225)). The
@@ -35,6 +281,23 @@ the tag exists and is exercised; what is left is merging `next` to
   the snapshot references, and `docs/DATAMODEL.md` now says
   `chunk_count` and `blob_count` count what the run added.
 
+- 2026-10-06: Made command output follow the README's stdout and stderr
+  rules ([issue #224](https://git.eeqj.de/sneak/vaultik/issues/224)). The
+  startup banner went to stdout, so a `completion` script or a
+  `config get` value started with it; the banner now goes to stderr. A
+  failing `remote info`, `prune` or `snapshot remove` under `--json`
+  printed nothing on either stream, and now reports its error on stderr.
+  `snapshot verify --quiet` printed its whole report; it now prints
+  none, and a failure still reaches stderr with the same exit status.
+
+- 2026-10-06: Made a backup without `--cron` of a snapshot with two or
+  more `paths` complete instead of panicking with `close of closed
+  channel` ([issue #253](https://git.eeqj.de/sneak/vaultik/issues/253)).
+  `Scan` runs once per path and started and stopped the progress
+  reporter each time, and a second stop panics. The reporter is now
+  started and stopped once per snapshot, around the scans of all its
+  paths.
+
 - 2026-10-06: Made a restore path argument select only that path and
   what is beneath it
   ([issue #223](https://git.eeqj.de/sneak/vaultik/issues/223)). The

config.example.yml

@@ -287,15 +287,18 @@ storage_url: "rclone://myremote/path/to/backups"
 #  #use_ssl: true
 #
 #  # Part size for multipart uploads
-#  # Minimum 5MB, affects memory usage during upload
+#  # Minimum 5MiB, maximum 5GiB; affects memory usage during upload
-#  # Supports: 5MB, 10M, 100MiB, etc.
+#  # A blob too large for 10,000 parts of this size gets larger parts
-#  # Default: 5MB
+#  # Supports: 10MB, 16MiB, 100MiB, etc. (5MB is below the minimum)
-#  #part_size: 5MB
+#  # Default: 5MiB
+#  #part_size: 5MiB
 
 # Path to local SQLite index database
 # This database tracks file state for incremental backups
-# Default: /var/lib/vaultik/index.sqlite
+# Default: the platform data directory, e.g.
-#index_path: /var/lib/vaultik/index.sqlite
+#   macOS: ~/Library/Application Support/vaultik/index.sqlite
+#   Linux: ~/.local/share/vaultik/index.sqlite
+#index_path: /path/to/index.sqlite
 
 # Average chunk size for content-defined chunking
 # Smaller chunks = better deduplication but more metadata
@@ -310,6 +313,12 @@ storage_url: "rclone://myremote/path/to/backups"
 # Chunking uses no secret (the FastCDC parameters are fixed and public). At a
 # large limit a blob holds hundreds of chunks, so individual chunk lengths are
 # not visible in its size; lowering the limit toward chunk_size exposes them.
+# A backup needs free temporary space, because each blob is written whole to
+# a temporary file before it is uploaded (up to about blob_size_limit; an
+# rclone destination that cannot stream uploads needs about twice that) and
+# the metadata export writes copies of the local index. Temporary files go to
+# $TMPDIR (default /tmp); with TMPDIR unset, SQLite writes one of those copies
+# to /var/tmp.
 # Supports: 1GB, 10G, 500MB, 1GiB, etc.
 # Default: 10GB
 #blob_size_limit: 10GB

docs/DATAMODEL.md

@@ -36,7 +36,8 @@ Stores metadata about files in the filesystem being backed up.
 **Columns:**
 - `id` (TEXT PRIMARY KEY) - UUID for the file record
 - `path` (TEXT NOT NULL UNIQUE) - Absolute file path
-- `mtime` (INTEGER NOT NULL) - Modification time as Unix timestamp
+- `mtime` (INTEGER NOT NULL) - Modification time, whole seconds since the Unix epoch
+- `mtime_nsec` (INTEGER NOT NULL) - Nanoseconds within that second, 0 to 999999999
 - `size` (INTEGER NOT NULL) - File size in bytes
 - `mode` (INTEGER NOT NULL) - Unix file permissions and type
 - `uid` (INTEGER NOT NULL) - User ID of file owner
@@ -89,7 +90,7 @@ Stores information about packed, compressed, and encrypted blob files.
 - `created_ts` (INTEGER NOT NULL) - Creation timestamp
 - `finished_ts` (INTEGER) - Finalization timestamp (NULL if in progress)
 - `uncompressed_size` (INTEGER NOT NULL DEFAULT 0) - Total size of chunks before compression
-- `compressed_size` (INTEGER NOT NULL DEFAULT 0) - Size after compression and encryption
+- `compressed_size` (INTEGER NOT NULL DEFAULT 0) - Size after compression and encryption. Once uploaded, the size of the object in storage: when the blob was already stored, under another compression level or other recipients, this can differ from the size of the local copy
 - `uploaded_ts` (INTEGER) - Upload completion timestamp (NULL if not uploaded)
 
 **Purpose:** Tracks blob lifecycle from creation through upload. The UUID primary key allows immediate association of chunks with blobs.
@@ -110,7 +111,7 @@ Maps chunks to the blobs that contain them.
 Tracks backup snapshots.
 
 **Columns:**
-- `id` (TEXT PRIMARY KEY) - Snapshot ID (format: hostname-YYYYMMDD-HHMMSSZ)
+- `id` (TEXT PRIMARY KEY) - Snapshot ID (format: `hostname_name_timestamp`, e.g. `server1_home_2025-06-01T12:00:00Z`: the hostname up to its first `.`, the snapshot name, and an RFC 3339 UTC timestamp)
 - `hostname` (TEXT) - Hostname where backup was created
 - `vaultik_version` (TEXT) - Version of Vaultik used
 - `vaultik_git_revision` (TEXT) - Git revision of Vaultik used
@@ -118,6 +119,7 @@ Tracks backup snapshots.
 - `completed_at` (INTEGER) - Completion timestamp (NULL if in progress)
 - `file_count` (INTEGER) - Number of files in snapshot
 - `chunk_count` (INTEGER) - Number of chunks this snapshot stored that were not stored before
+- `new_chunk_size` (INTEGER) - Total size of those chunks before compression; `snapshot list` shows it as NEW CHUNK SIZE
 - `blob_count` (INTEGER) - Number of blobs this snapshot created
 - `total_size` (INTEGER) - Total size of all files
 - `blob_size` (INTEGER) - Total compressed size of all referenced blobs
@@ -182,7 +184,7 @@ Tracks blob upload metrics.
    - `UPDATE blobs SET blob_hash = ?, finished_ts = ?` - Finalize blob after packing
 
 4. **Upload**
-   - `UPDATE blobs SET uploaded_ts = ?` - Mark blob as uploaded
+   - `UPDATE blobs SET uploaded_ts = ?, compressed_size = ?` - Mark blob as uploaded, with the size of the object in storage (the upload is skipped when the blob is already stored)
    - `INSERT INTO uploads` - Record upload metrics with snapshot_id
    - `INSERT INTO snapshot_blobs` - Associate blob with snapshot
 
@@ -217,8 +219,8 @@ The `{remote-key}` directory name is a one-way hash of the human snapshot ID, so
 ### 4. Restore Process
 
 The restore process doesn't use the local database. Instead:
-1. Downloads snapshot metadata from S3
+1. Downloads and decrypts the snapshot's metadata database (`db.zst.age`) from S3
-2. Downloads required blobs based on manifest
+2. Downloads the blobs holding the chunks of the files being restored, found through that database's `blob_chunks` table; the manifest is not read
 3. Reconstructs files from decrypted and decompressed chunks
 
 ### 5. Pruning
@@ -231,9 +233,8 @@ The restore process doesn't use the local database. Instead:
 
 Before each backup:
 1. Query incomplete snapshots (where `completed_at IS NULL`)
-2. Check if metadata exists in S3
+2. Delete each one and all its associations, without checking S3 for its metadata
-3. If no metadata, delete snapshot and all associations
+3. Clean up orphaned files, chunks, and blobs
-4. Clean up orphaned files, chunks, and blobs
 
 ## Repository Pattern
 

docs/REPOSTRUCTURE.md

@@ -4,7 +4,7 @@ This document describes the structure and organization of data stored in the S3
 
 ## Overview
 
-Vaultik stores all backup data in an S3-compatible object store. The repository consists of two main components:
+Vaultik stores all backup data in an S3-compatible object store. The repository consists of two main components, plus a marker file:
 1. **Blobs** - The actual backup data (content-addressed, encrypted)
 2. **Metadata** - Snapshot information and manifests (partially encrypted)
 
@@ -12,6 +12,7 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
 
 ```
 <bucket>/<prefix>/
+├── vaultik-destination.txt
 ├── blobs/
 │   └── <hash[0:2]>/
 │       └── <hash[2:4]>/
@@ -24,6 +25,10 @@ Vaultik stores all backup data in an S3-compatible object store. The repository
 
 The metadata subdirectory is named with the **remote key**, a one-way hash of the snapshot ID, not with the human-readable snapshot ID itself. See [Remote Key Derivation](#remote-key-derivation).
 
+## Marker File (`vaultik-destination.txt`)
+
+Every backup writes this file at the root before it uploads anything. It holds one fixed line of plain text and names no host, path or time. While the local index holds snapshots, a destination without it is treated as not present: `snapshot create` refuses to write to it, and `prune`, `snapshot purge`, `snapshot remove` and `snapshot list` behave as for a destination that cannot be listed. An unplugged volume leaves an empty directory at its mount point, and without the marker that directory would read as a destination whose snapshots are all gone. `remote nuke` deletes the marker last.
+
 ## Blobs Directory (`blobs/`)
 
 ### Structure

internal/blob/packer.go

@@ -245,6 +245,17 @@ func (p *Packer) Flush(ctx context.Context) error {
 	return nil
 }
 
+// DiscardCurrentBlob closes and removes the temp file of the blob being
+// packed, if any, and drops that blob. A caller that stops before Flush
+// must call it, or the temp file is left behind. Thread-safe.
+func (p *Packer) DiscardCurrentBlob() {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+
+	p.cleanupTempFile()
+	p.currentBlob = nil
+}
+
 // FinalizeBlob finalizes the current blob being assembled.
 // This compresses the accumulated chunks, encrypts the result, and computes
 // the content-addressed hash. The finalized blob is either passed to the

internal/cli/app.go

@@ -200,12 +200,16 @@ func RunApp(ctx context.Context, app *fx.App) error {
 }
 
 // errReported marks a failure the operation has already shown the user
-// (and deliberately withheld under --json). Entry turns it into a
+// (or, under `snapshot verify --json`, put in its document). Entry
-// non-zero exit status without printing anything further, so the error
+// turns it into a non-zero exit status without printing anything
-// line is not doubled. It flows up from RunOperation through cobra to
+// further, so the error line is not doubled. It flows up from
-// Entry.
+// RunOperation through cobra to Entry.
 var errReported = errors.New("operation failed")
 
+// errInterrupted marks an operation that SIGINT or SIGTERM stopped
+// before it finished. Entry shows it and returns exitCodeInterrupted.
+var errInterrupted = errors.New("interrupted before the command finished")
+
 // RunOperation runs op against the Vaultik instance inside the fx app
 // and turns a failure into a returned error rather than an os.Exit from
 // within the goroutine. An os.Exit there skipped main's deferred
@@ -220,17 +224,21 @@ var errReported = errors.New("operation failed")
 // interrupt OnStop cancels op and waits for the goroutine to return, so
 // op's cleanup (removing decrypted scratch files) runs before the
 // process exits; the wait is bounded by shutdownTimeout. report is
-// called with a non-canceled failure so the caller can log it (and
+// called with a failure so the caller can show it to the user before
-// suppress it under --json) before it becomes errReported. A context
+// it becomes errReported.
-// cancellation is the interrupt path, not a failure: it is neither
+//
-// reported nor counted as one.
+// The run counts as interrupted unless op returned, without an
+// interrupt having cancelled it, before RunWithApp returned. An
+// interrupted op is not reported, whatever it returned; RunOperation
+// returns errInterrupted instead.
 func RunOperation(
 	ctx context.Context, opts AppOptions,
 	op func(v *vaultik.Vaultik) error, report func(err error),
 ) error {
 	var (
-		mu     sync.Mutex
+		mu       sync.Mutex
-		failed bool
+		finished bool // op returned before any interrupt cancelled it
+		failed   bool // op finished with an error
 	)
 
 	opts.Invokes = append(opts.Invokes,
@@ -241,11 +249,21 @@ func RunOperation(
 				OnStart: func(_ context.Context) error {
 					stop = v.StartOperation(func() {
 						err := op(v)
-						if err != nil && !errors.Is(err, context.Canceled) {
+
-							report(err)
+						// Only stop, called from OnStop below, cancels the
+						// Vaultik context, so a live context means no
+						// interrupt cancelled op. Check the context, not
+						// err: an interrupted op need not return
+						// context.Canceled (`snapshot verify --json`
+						// returns a verification failure).
+						if v.Context().Err() == nil {
+							if err != nil {
+								report(err)
+							}
 
 							mu.Lock()
-							failed = true
+							finished = true
+							failed = err != nil
 							mu.Unlock()
 						}
 
@@ -278,28 +296,34 @@ func RunOperation(
 		return err
 	}
 
-	// The goroutine sets failed before triggering the shutdown that lets
+	// RunWithApp returns only after the app was asked to stop, either by
-	// RunWithApp return, so the write is in place by the time we read it.
+	// an interrupt or by the goroutine's Shutdown call. When op finished
+	// without being cancelled, the goroutine set finished before that
+	// call. So if finished is unset here, an interrupt stopped the app,
+	// and op either returned after it was cancelled or is still running
+	// because the shutdown timed out.
 	mu.Lock()
 	defer mu.Unlock()
 
-	if failed {
+	switch {
+	case !finished:
+		return errInterrupted
+	case failed:
 		return errReported
+	default:
+		return nil
 	}
-
-	return nil
 }
 
 // runVaultikApp runs the standard single-operation command lifecycle
 // shared by the snapshot list/purge/remove and remote nuke subcommands:
 // resolve the config, then run op against the Vaultik instance through
-// RunOperation, reporting a failure prefixed with failMsg (suppressed
+// RunOperation, reporting a failure prefixed with failMsg on stderr. mode
-// while suppressErrors is true, e.g. under --json). mode says whether the
+// says whether the command takes the PID lock. jsonOutput marks a command
-// command takes the PID lock. jsonOutput marks a command whose stdout is a
+// whose stdout is a JSON document: it quiets the UI but, unlike Quiet,
-// JSON document: it quiets the UI but, unlike Quiet, leaves the stderr log
+// leaves the stderr log level alone.
-// level alone.
 func runVaultikApp(
-	cmd *cobra.Command, mode lockMode, jsonOutput, suppressErrors bool,
+	cmd *cobra.Command, mode lockMode, jsonOutput bool,
 	failMsg string, op func(v *vaultik.Vaultik) error,
 ) error {
 	configPath, err := ResolveConfigPath()
@@ -319,10 +343,6 @@ func runVaultikApp(
 		},
 		Mode: mode,
 	}, op, func(err error) {
-		if suppressErrors {
-			return
-		}
-
 		log.Error(failMsg, "error", err)
 		ReportErrorf("%s: %v", failMsg, err)
 	})

internal/cli/config.go

@@ -7,11 +7,14 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"reflect"
 	"strconv"
 	"strings"
+	"unicode/utf8"
 
 	"github.com/spf13/cobra"
 	"gopkg.in/yaml.v3"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/ui"
 )
 
@@ -31,6 +34,9 @@ const configDirMode = 0o755
 // yaml.Marshal's 4-space default.
 const configYAMLIndent = 2
 
+// yamlStringTag is YAML's tag for a string scalar.
+const yamlStringTag = "!!str"
+
 var (
 	errConfigExists     = errors.New("config file already exists")
 	errEmptyConfig      = errors.New("empty config file")
@@ -199,7 +205,7 @@ storage_url: ""
 #   access_key_id: YOUR_ACCESS_KEY
 #   secret_access_key: YOUR_SECRET_KEY
 #   # region: us-east-1        # Default: us-east-1
-#   # part_size: 5MB           # Multipart upload part size. Default: 5MB
+#   # part_size: 5MiB          # Upload part size, 5MiB to 5GiB. Default: 5MiB
 #   # For the s3:// form, disable TLS with ?ssl=false in the URL, not use_ssl.
 
 # ─── OPTIONAL ────────────────────────────────────────────────────────────────
@@ -245,6 +251,12 @@ func NewConfigCommand() *cobra.Command {
 		Use:   "config",
 		Short: "Manage the configuration file",
 		Long:  "Commands for creating, editing, and querying the vaultik config file.",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(newConfigInitCommand())
@@ -583,9 +595,58 @@ func yamlPathSet(root *yaml.Node, keys []string, value string) error {
 		}
 	}
 
+	// config.Load reads the file through untyped YAML, which turns an
+	// unquoted 00112233 into the number 38043 and 1e5 into 100000. Tagging
+	// a string setting as a string makes the encoder quote such a value.
+	// Other settings stay unquoted, so compression_level 9 is a number.
+	// The encoder refuses to write a value that is not valid UTF-8 as a
+	// string. Left untagged, such a value is written as base64 !!binary and
+	// loads back unchanged.
+	if configKeyIsString(keys) && utf8.ValidString(value) {
+		node.Tag = yamlStringTag
+	}
+
 	return nil
 }
 
+// configKeyIsString reports whether the dotted key names a string in
+// config.Config, following the fields' yaml tags, as s3.access_key_id and
+// snapshots.home.exclude.0 do.
+func configKeyIsString(keys []string) bool {
+	typ := reflect.TypeFor[config.Config]()
+
+	for _, key := range keys {
+		switch {
+		case typ.Kind() == reflect.Map || typ.Kind() == reflect.Slice:
+			// The key is a snapshot name or a list index.
+			typ = typ.Elem()
+		case typ.Kind() == reflect.Struct:
+			field, ok := yamlField(typ, key)
+			if !ok {
+				return false
+			}
+
+			typ = field.Type
+		default:
+			return false
+		}
+	}
+
+	return typ.Kind() == reflect.String
+}
+
+// yamlField returns the field of struct type typ whose yaml tag names key.
+func yamlField(typ reflect.Type, key string) (reflect.StructField, bool) {
+	for field := range typ.Fields() {
+		name, _, _ := strings.Cut(field.Tag.Get("yaml"), ",")
+		if name == key {
+			return field, true
+		}
+	}
+
+	return reflect.StructField{}, false
+}
+
 // yamlSetInMapping resolves (creating if needed) the value node for key
 // within a mapping node, setting it to value when it is the final path
 // element, and returns the node to descend into.

internal/cli/config_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"os"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"testing"
 
@@ -94,6 +95,110 @@ func TestConfigSetRecipientOnFreshConfig(t *testing.T) {
 	}
 }
 
+// TestConfigSetStringLooksLikeNumber sets string settings to values that
+// YAML reads as numbers or booleans when they are unquoted, and checks that
+// config.Load returns each one unchanged.
+func TestConfigSetStringLooksLikeNumber(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		key   string
+		value string
+		field func(cfg *config.Config) string
+	}{
+		{"s3.access_key_id", "00112233",
+			func(cfg *config.Config) string { return cfg.S3.AccessKeyID }},
+		{"s3.secret_access_key", "12345678901234567890123456789012",
+			func(cfg *config.Config) string { return cfg.S3.SecretAccessKey }},
+		{"hostname", "007",
+			func(cfg *config.Config) string { return cfg.Hostname }},
+		{"s3.prefix", "1e5",
+			func(cfg *config.Config) string { return cfg.S3.Prefix }},
+		{"s3.bucket", "true",
+			func(cfg *config.Config) string { return cfg.S3.Bucket }},
+		{"s3.region", "FALSE",
+			func(cfg *config.Config) string { return cfg.S3.Region }},
+		{"snapshots.home.exclude.0", "1.10",
+			func(cfg *config.Config) string { return cfg.Snapshots["home"].Exclude[0] }},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.key+"="+tt.value, func(t *testing.T) {
+			t.Parallel()
+
+			cfg := loadAfterConfigSet(t, tt.key, tt.value)
+
+			got := tt.field(cfg)
+			if got != tt.value {
+				t.Errorf("%s = %q after config set %q", tt.key, got, tt.value)
+			}
+		})
+	}
+}
+
+// TestConfigSetNonUTF8Path checks that config set still accepts a value that
+// is not valid UTF-8, such as a path with a Latin-1 file name, and that
+// config.Load returns it unchanged.
+func TestConfigSetNonUTF8Path(t *testing.T) {
+	t.Parallel()
+
+	const dir = "/srv/caf\xe9"
+
+	cfg := loadAfterConfigSet(t, "snapshots.home.paths.0", dir)
+
+	got := cfg.Snapshots["home"].Paths[0]
+	if got != dir {
+		t.Errorf("snapshots.home.paths.0 = %q, want %q", got, dir)
+	}
+}
+
+// TestConfigSetNumberStaysNumber checks that a number set for an integer
+// setting is still read as a number, not as a quoted string.
+func TestConfigSetNumberStaysNumber(t *testing.T) {
+	t.Parallel()
+
+	const level = 9
+
+	cfg := loadAfterConfigSet(t, "compression_level", strconv.Itoa(level))
+
+	if cfg.CompressionLevel != level {
+		t.Errorf("compression_level = %d, want %d", cfg.CompressionLevel, level)
+	}
+}
+
+// loadAfterConfigSet writes the file `config init` writes, sets storage_url
+// to a local directory so that the file passes validation, applies
+// `config set key value` and returns what config.Load reads back.
+func loadAfterConfigSet(t *testing.T, key, value string) *config.Config {
+	t.Helper()
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(defaultConfigTemplate), configFileMode)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	out := ui.NewWithColor(&bytes.Buffer{}, false)
+
+	err = writeConfigSet(out, path, "storage_url", "file:///mnt/backups")
+	if err != nil {
+		t.Fatalf("config set storage_url: %v", err)
+	}
+
+	err = writeConfigSet(out, path, key, value)
+	if err != nil {
+		t.Fatalf("config set %s: %v", key, err)
+	}
+
+	cfg, err := config.Load(path)
+	if err != nil {
+		t.Fatalf("config.Load: %v", err)
+	}
+
+	return cfg
+}
+
 const testYAML = `# top comment
 compression_level: 3
 age_recipients:

internal/cli/database.go

@@ -15,6 +15,12 @@ func NewDatabaseCommand() *cobra.Command {
 		Use:   "database",
 		Short: "Manage the local state database",
 		Long:  `Commands for managing the local SQLite state database.`,
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	cmd.AddCommand(

internal/cli/entry.go

@@ -15,17 +15,25 @@ import (
 // the startup banner.
 const shortCommitLen = 12
 
+// exitCodeInterrupted is the exit status of a command that SIGINT or
+// SIGTERM stopped. It is 128 plus SIGINT's number, 2, which is what a
+// shell reports for a command stopped by Ctrl-C.
+const exitCodeInterrupted = 130
+
 // Entry is the main entry point for the CLI application.
-// It prints the startup banner to stdout (unless a banner-suppressing
+// It prints the startup banner to stderr (unless a banner-suppressing
 // flag is present in os.Args — see bannerSuppressedInArgs), executes the
 // root cobra command, and routes any returned error through the
 // ui.Writer so the user sees a properly formatted "🛑 ERROR:" line.
+// The banner goes to stderr because stdout carries only the output the
+// user asked for, such as a completion script or a `config get` value.
 //
-// It returns the process exit code (0 on success, 1 on error) rather
+// It returns the process exit code (0 on success, 130 when interrupted,
-// than calling os.Exit, so that main's deferred profile writers run
+// 1 on any other error) rather than calling os.Exit, so that main's
-// before the process ends. See run in cmd/vaultik/main.go.
+// deferred profile writers run before the process ends. See run in
+// cmd/vaultik/main.go.
 func Entry() int {
-	emitStartupBanner(os.Args[1:], os.Stdout)
+	emitStartupBanner(os.Args[1:], os.Stderr)
 
 	rootCmd := NewRootCommand()
 	rootCmd.SilenceErrors = true
@@ -33,14 +41,20 @@ func Entry() int {
 	err := rootCmd.Execute()
 	if err != nil {
 		// An operation that ran inside the fx app has already reported
-		// its own failure (and suppressed it under --json); errReported
+		// its own failure (`snapshot verify --json` puts it in the
-		// says so. Printing it again here would double the error line.
+		// document instead); errReported says so. Printing it again
+		// here would double the error line.
 		// Every other error — bad arguments, a config that would not
-		// load — reaches Entry unreported, so it is shown here.
+		// load, an interrupt — reaches Entry unreported, so it is shown
+		// here.
 		if !errors.Is(err, errReported) {
 			ReportErrorf("%s", err.Error())
 		}
 
+		if errors.Is(err, errInterrupted) {
+			return exitCodeInterrupted
+		}
+
 		return 1
 	}
 
@@ -49,9 +63,8 @@ func Entry() int {
 
 // emitStartupBanner writes the startup banner to w unless args (the
 // argument vector with the program name already stripped) contains a
-// flag that suppresses it. Split out of Entry so that the decision — the
+// flag that suppresses it. Split out of Entry so that the decision is
-// only thing standing between a --json invocation and a parseable
+// reachable from a test without running the whole CLI.
-// stdout — is reachable from a test without running the whole CLI.
 func emitStartupBanner(args []string, w io.Writer) {
 	if bannerSuppressedInArgs(args) {
 		return
@@ -86,8 +99,7 @@ func ReportErrorf(format string, args ...any) {
 // --json is a subcommand flag rather than a persistent one, but so is
 // --cron (it exists only on `snapshot create`), so this adds no new
 // class of imprecision. The only cost of a false positive is a missing
-// decorative banner; the cost of a false negative is a corrupt document
+// decorative banner.
-// on stdout, so the scan errs deliberately in that direction.
 func bannerSuppressedInArgs(args []string) bool {
 	for _, a := range args {
 		if a == "--" {

internal/cli/entry_banner_test.go

@@ -36,23 +36,14 @@ const (
 	// strips it before scanning, so it has to be present.
 	programName = "vaultik"
 
-	// someSnapshotID is any snapshot identifier: these tests never run
+	// someSnapshotID only fills the positional argument; no test needs
-	// the command, so it only has to occupy the positional argument.
+	// the snapshot to exist.
 	someSnapshotID = "host_2026-01-01T00:00:00Z"
 )
 
-// placeholderJSONDocument stands in for whatever document a --json
-// command writes to stdout. `snapshot list --json` with no snapshots
-// prints exactly this; the other --json commands print an object rather
-// than an array, but this test is not about their shape. It is about
-// what is on stdout *before* them, which is the same for all of them
-// because Entry prints the banner before cobra has parsed anything and
-// therefore before it can know which command is running.
-const placeholderJSONDocument = "[]\n"
-
 // jsonArgumentVectors are the argument vectors of every --json
 // invocation the CLI accepts, with the program name stripped exactly as
-// Entry strips it. Each one must leave stdout untouched by the banner.
+// Entry strips it. Each one must suppress the banner.
 //
 //nolint:gochecknoglobals // read-only test fixture shared by two tests
 var jsonArgumentVectors = map[string][]string{
@@ -74,39 +65,23 @@ var jsonArgumentVectors = map[string][]string{
 	},
 }
 
-// TestJSONInvocationStdoutIsExactlyOneDocument is the CLI-layer
+// TestJSONInvocationSuppressesBanner checks that every --json
-// regression guard for issue #106: `vaultik snapshot list --json | jq`
+// invocation suppresses the startup banner, as the README says --json
-// must work with no other flags.
+// does along with --quiet and --cron. The scan is over the raw argument
-//
+// vector, so each position and spelling of --json is listed.
-// internal/vaultik's TestListSnapshots_JSONStdoutIsOnlyTheDocument
+func TestJSONInvocationSuppressesBanner(t *testing.T) {
-// guards the same contract one layer down, but it calls the library
-// function directly and so cannot see Entry, which is where the
-// contamination was: the startup banner is written to stdout before
-// cobra parses anything, and the suppression scan did not know about
-// --json. The two banner lines and the blank line landed ahead of the
-// document and `jq` refused the result.
-//
-// The document is a constant here because this test is about the
-// argument vectors, one per --json command; the one that runs a real
-// command end to end is TestEntryJSONStdoutIsExactlyOneDocument below.
-func TestJSONInvocationStdoutIsExactlyOneDocument(t *testing.T) {
 	t.Parallel()
 
 	for name, argv := range jsonArgumentVectors {
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 
-			var stdout bytes.Buffer
+			var banner bytes.Buffer
 
-			emitStartupBanner(argv, &stdout)
+			emitStartupBanner(argv, &banner)
 
-			require.Empty(t, stdout.String(),
+			assert.Empty(t, banner.String(),
-				"nothing may reach stdout ahead of a --json document")
+				"--json suppresses the banner")
-
-			_, err := stdout.WriteString(placeholderJSONDocument)
-			require.NoError(t, err)
-
-			requireExactlyOneJSONDocument(t, stdout.String())
 		})
 	}
 }
@@ -127,11 +102,11 @@ func TestBannerStillPrintedWithoutSuppressingFlag(t *testing.T) {
 		t.Run(name, func(t *testing.T) {
 			t.Parallel()
 
-			var stdout bytes.Buffer
+			var banner bytes.Buffer
 
-			emitStartupBanner(argv, &stdout)
+			emitStartupBanner(argv, &banner)
 
-			assert.Contains(t, stdout.String(), "starting up at",
+			assert.Contains(t, banner.String(), "starting up at",
 				"the banner belongs on invocations that did not opt out")
 		})
 	}
@@ -247,9 +222,7 @@ func TestEntryJSONStdoutIsExactlyOneDocument(t *testing.T) {
 // captureProcessStdout redirects the process's own stdout to a pipe for
 // the duration of fn and returns what was written to it. The redirection
 // has to be at the file-descriptor level rather than through an injected
-// writer, because the banner and the JSON encoder reach os.Stdout
+// writer, because the commands Entry runs reach os.Stdout directly.
-// independently and the point of the test is that both land in the same
-// place.
 //
 // Not parallel-safe: os.Stdout is process-global.
 func captureProcessStdout(t *testing.T, fn func()) string {

internal/cli/entry_interrupt_test.go

@@ -0,0 +1,203 @@
+package cli //nolint:testpackage // shares runEntry and the argument constants
+
+import (
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/signal"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/adrg/xdg"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// stalledStoreConfig is hermeticConfig with an s3:// destination store
+// in place of the file:// one. The server behind it accepts any
+// credentials.
+const stalledStoreConfig = `age_recipients:
+  - age1278m9q7dp3chsh2dcy82qk27v047zywyvtxwnj4cvt0z65jw6a7q5dqhfj
+snapshots:
+  test:
+    paths:
+      - %s
+storage_url: s3://bucket?endpoint=%s&ssl=false
+s3:
+  access_key_id: key
+  secret_access_key: secret
+index_path: %s
+hostname: test-host
+`
+
+// interruptRepeat is how often interruptOnFirstRequest sends SIGINT.
+const interruptRepeat = 50 * time.Millisecond
+
+// TestEntryInterruptedRun sends SIGINT to the test process while a
+// command waits on the destination store, and checks that Entry returns
+// 130 and prints one line on stderr saying the run was interrupted. The
+// store is a local HTTP server that holds every request open, so the
+// command is always mid-operation when the signal arrives. The two
+// cases cover --cron and --json, which silence other output.
+//
+// Not parallel: it signals the process and replaces os.Args, os.Stdout,
+// os.Stderr and the xdg globals.
+//
+//nolint:paralleltest // signals the process and replaces process globals
+func TestEntryInterruptedRun(t *testing.T) {
+	for _, testCase := range []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "snapshot create --cron",
+			args: []string{cmdSnapshot, cmdCreate, "--cron"},
+		},
+		{
+			name: "snapshot verify --json",
+			args: []string{cmdSnapshot, cmdVerify, someSnapshotID, flagJSON},
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			endpoint, requestArrived := startStalledStore(t)
+			configPath := writeStalledStoreConfig(t, endpoint)
+			interruptOnFirstRequest(t, requestArrived)
+
+			code, _, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 130, code)
+			assert.Equal(t, 1,
+				strings.Count(stderr, errInterrupted.Error()), stderr)
+		})
+	}
+}
+
+// interruptOnFirstRequest sends SIGINT to the test process every
+// interruptRepeat, from the first request to the destination store until
+// the test ends. One signal is not enough: the command can reach the
+// store before fx has started catching signals. The test catches SIGINT
+// too, so that a signal fx is not catching does not kill the test
+// binary.
+func interruptOnFirstRequest(t *testing.T, requestArrived <-chan struct{}) {
+	t.Helper()
+
+	self, err := os.FindProcess(os.Getpid())
+	require.NoError(t, err)
+
+	caught := make(chan os.Signal, 1)
+	signal.Notify(caught, os.Interrupt)
+
+	testEnded := make(chan struct{})
+	senderDone := make(chan struct{})
+
+	// Stop catching SIGINT only after the sender has returned. The sender
+	// waits for each SIGINT it sends to arrive on caught; one still on
+	// its way after signal.Stop would kill the test binary.
+	t.Cleanup(func() {
+		close(testEnded)
+		<-senderDone
+		signal.Stop(caught)
+	})
+
+	go func() {
+		defer close(senderDone)
+
+		select {
+		case <-requestArrived:
+		case <-testEnded:
+			return
+		}
+
+		ticker := time.NewTicker(interruptRepeat)
+		defer ticker.Stop()
+
+		for {
+			// Empty caught, so that the receive below waits for this
+			// SIGINT rather than an earlier one.
+			select {
+			case <-caught:
+			default:
+			}
+
+			sendErr := self.Signal(os.Interrupt)
+			if sendErr != nil {
+				t.Errorf("sending SIGINT: %v", sendErr)
+
+				return
+			}
+
+			<-caught
+
+			select {
+			case <-testEnded:
+				return
+			case <-ticker.C:
+			}
+		}
+	}()
+}
+
+// startStalledStore starts an HTTP server that never answers: each
+// request is held until the client gives up on it or the test ends.
+// It returns the server's host:port and a channel that receives a value
+// when the first request arrives.
+func startStalledStore(t *testing.T) (string, <-chan struct{}) {
+	t.Helper()
+
+	requestArrived := make(chan struct{}, 1)
+	release := make(chan struct{})
+
+	server := httptest.NewServer(http.HandlerFunc(
+		func(_ http.ResponseWriter, r *http.Request) {
+			select {
+			case requestArrived <- struct{}{}:
+			default:
+			}
+
+			select {
+			case <-r.Context().Done():
+			case <-release:
+			}
+		}))
+
+	// Cleanups run last-registered first, so release lets any held
+	// request return before Close waits for it.
+	t.Cleanup(server.Close)
+	t.Cleanup(func() { close(release) })
+
+	return server.Listener.Addr().String(), requestArrived
+}
+
+// writeStalledStoreConfig writes a config whose destination store is the
+// server at endpoint and whose snapshot source holds one small file, so
+// that `snapshot create` has a blob to upload. Returns the config path.
+func writeStalledStoreConfig(t *testing.T, endpoint string) string {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	sourceDir := filepath.Join(dir, "source")
+
+	require.NoError(t, os.Mkdir(sourceDir, 0o750))
+	require.NoError(t, os.WriteFile(filepath.Join(sourceDir, "file.txt"),
+		[]byte("contents"), 0o600))
+
+	contents := fmt.Sprintf(stalledStoreConfig,
+		sourceDir, endpoint, filepath.Join(dir, "index.sqlite"))
+
+	require.NoError(t,
+		os.WriteFile(configPath, []byte(contents), configFileMode))
+
+	// The PID lock lives under xdg.DataHome, which xdg resolves at
+	// package init; point it at the temp dir so the test neither
+	// touches nor collides with the real one.
+	t.Setenv("XDG_DATA_HOME", filepath.Join(dir, "data"))
+	xdg.Reload()
+	t.Cleanup(xdg.Reload)
+
+	return configPath
+}

internal/cli/entry_prune_json_test.go

@@ -101,7 +101,8 @@ func TestEntryPruneJSONStdoutIsExactlyOneDocument(t *testing.T) {
 // writeHermeticPruneConfig builds a config over a temp directory with an
 // empty destination directory and, if seedStale is set, creates the
 // index database up front with one snapshot record that has no
-// counterpart on the destination store. Returns the config path.
+// counterpart on the destination store, and gives the destination the
+// marker a backup writes. Returns the config path.
 func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 	t.Helper()
 
@@ -131,6 +132,11 @@ func writeHermeticPruneConfig(t *testing.T, seedStale bool) string {
 
 	if seedStale {
 		seedStaleSnapshotRecord(t, indexPath)
+
+		// While the local index holds a snapshot, prune fails on a
+		// destination without the marker a backup writes.
+		require.NoError(t, os.WriteFile(
+			filepath.Join(storeDir, "vaultik-destination.txt"), nil, 0o600))
 	}
 
 	return configPath

internal/cli/entry_snapshot_create_prune_test.go

@@ -0,0 +1,108 @@
+package cli //nolint:testpackage // shares runEntry and the hermetic config helpers
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// Flags of `snapshot create` used to build argument vectors below.
+const (
+	flagPrune         = "--prune"
+	flagKeepNewerThan = "--keep-newer-than"
+)
+
+// TestEntrySnapshotCreateRejectsKeepNewerThanBeforeBackup checks that a
+// --keep-newer-than the post-backup prune could not use stops `snapshot
+// create` before the backup: the command fails and the destination
+// store stays empty. It used to be read only by the prune, after the
+// snapshot was uploaded, and without --prune it was ignored.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateRejectsKeepNewerThanBeforeBackup(t *testing.T) {
+	for _, testCase := range []struct {
+		name string
+		args []string
+	}{
+		{
+			name: "decimal in a day count",
+			args: []string{flagPrune, flagKeepNewerThan, "1.5d"},
+		},
+		{
+			name: "unknown unit",
+			args: []string{flagPrune, flagKeepNewerThan, "4x"},
+		},
+		{
+			name: "without --prune",
+			args: []string{flagKeepNewerThan, "4w"},
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath, storeDir := writeBackupConfig(t)
+
+			code, _, _ := runEntry(t, append(
+				[]string{flagConfig, configPath, cmdSnapshot, cmdCreate},
+				testCase.args...)...)
+
+			assert.Equal(t, 1, code)
+
+			stored, err := os.ReadDir(storeDir)
+			require.NoError(t, err)
+			assert.Empty(t, stored, "nothing may be uploaded")
+		})
+	}
+}
+
+// TestEntrySnapshotCreateReportsPostBackupPruneFailure makes the
+// post-backup prune fail after a good backup, by planting a manifest
+// that cannot be decoded, and checks that the failure is reported as the
+// prune's and not as a failed snapshot creation.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotCreateReportsPostBackupPruneFailure(t *testing.T) {
+	configPath, storeDir := writeBackupConfig(t)
+
+	// A remote key is 64 lowercase hex characters. The prune reads every
+	// manifest on the store and fails on one it cannot decode.
+	plantedDir := filepath.Join(storeDir, "metadata", strings.Repeat("a", 64))
+	require.NoError(t, os.MkdirAll(plantedDir, 0o750))
+	require.NoError(t, os.WriteFile(
+		filepath.Join(plantedDir, "manifest.json.zst"),
+		[]byte("not a valid manifest"), 0o600))
+
+	code, _, stderr := runEntry(t,
+		flagConfig, configPath, cmdSnapshot, cmdCreate, flagPrune)
+
+	assert.Equal(t, 1, code)
+	assert.Contains(t, stderr, "Backup completed, but post-backup prune failed")
+	assert.NotContains(t, stderr, "Snapshot creation failed")
+
+	snapshots, err := os.ReadDir(filepath.Join(storeDir, "metadata"))
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 2,
+		"the new snapshot is on the store beside the planted one")
+}
+
+// writeBackupConfig builds a hermetic config with one small file in its
+// source directory and an empty destination store directory. Returns the
+// config path and the store directory.
+func writeBackupConfig(t *testing.T) (string, string) {
+	t.Helper()
+
+	configPath := writeHermeticPruneConfig(t, false)
+
+	// writeHermeticPruneConfig puts the source and store directories
+	// beside the config file.
+	dir := filepath.Dir(configPath)
+	sourceDir := filepath.Join(dir, "source")
+
+	require.NoError(t, os.Mkdir(sourceDir, 0o750))
+	require.NoError(t, os.WriteFile(
+		filepath.Join(sourceDir, "file.txt"), []byte("hello"), 0o600))
+
+	return configPath, filepath.Join(dir, "store")
+}

internal/cli/entry_status_test.go

@@ -15,8 +15,8 @@ import (
 // run, so a failing command must come back with a non-zero code rather
 // than ending the process here.
 //
-// Stdout is captured only to keep the banner and command output off the
+// Stdout and stderr are captured only to keep the banner and command
-// test log; the assertion is on the returned code.
+// output off the test log; the assertion is on the returned code.
 //
 //nolint:paralleltest // replaces os.Args and rootFlags
 func TestEntryReturnsStatusCode(t *testing.T) {
@@ -50,7 +50,7 @@ func TestEntryReturnsStatusCode(t *testing.T) {
 
 			var code int
 
-			_ = captureProcessStdout(t, func() { code = Entry() })
+			_, _ = captureProcessStdoutAndStderr(t, func() { code = Entry() })
 
 			assert.Equal(t, testCase.want, code)
 		})

internal/cli/entry_stdout_stderr_test.go

@@ -0,0 +1,200 @@
+package cli //nolint:testpackage // shares hermeticConfig and the capture helpers
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"log/slog"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/adrg/xdg"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+)
+
+// TestEntryCompletionStdoutIsTheScript runs `vaultik completion bash`,
+// whose stdout the README tells the user to source. The script has to
+// start on the first line.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryCompletionStdoutIsTheScript(t *testing.T) {
+	code, stdout, _ := runEntry(t, "completion", "bash")
+
+	require.Equal(t, 0, code)
+
+	firstLine, _, _ := strings.Cut(stdout, "\n")
+	assert.True(t, strings.HasPrefix(firstLine, "# bash completion"),
+		"the first line of stdout must be the script's, got %q", firstLine)
+}
+
+// TestEntryConfigGetStdoutIsTheValue runs `vaultik config get`, whose
+// stdout a script reads as the value and nothing else.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryConfigGetStdoutIsTheValue(t *testing.T) {
+	configPath := filepath.Join(t.TempDir(), "config.yml")
+	require.NoError(t, os.WriteFile(configPath,
+		[]byte("hostname: test-host\n"), configFileMode))
+
+	code, stdout, _ := runEntry(t,
+		flagConfig, configPath, "config", "get", "hostname")
+
+	require.Equal(t, 0, code)
+	assert.Equal(t, "test-host\n", stdout)
+}
+
+// TestEntryJSONFailureIsReportedOnStderr runs each --json command that
+// writes no document when it fails, against a destination it cannot
+// use. The error must reach stderr, and stdout must stay empty.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntryJSONFailureIsReportedOnStderr(t *testing.T) {
+	for _, testCase := range []struct {
+		name         string
+		args         []string
+		wantOnStderr string
+	}{
+		{
+			name:         "remote info",
+			args:         []string{cmdRemote, cmdInfo, flagJSON},
+			wantOnStderr: "Failed to get remote info",
+		},
+		{
+			name:         "prune",
+			args:         []string{cmdPrune, flagJSON},
+			wantOnStderr: "Prune failed",
+		},
+		{
+			name:         "snapshot remove",
+			args:         []string{cmdSnapshot, cmdRemove, someSnapshotID, flagJSON},
+			wantOnStderr: "Failed to remove snapshot",
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configPath := writeUnusableDestinationConfig(t)
+
+			code, stdout, stderr := runEntry(t,
+				append([]string{flagConfig, configPath}, testCase.args...)...)
+
+			assert.Equal(t, 1, code)
+			assert.Empty(t, stdout,
+				"a failed --json command has no document to write")
+			assert.Contains(t, stderr, testCase.wantOnStderr,
+				"the failure must be reported on stderr")
+		})
+	}
+}
+
+// TestEntrySnapshotRemoveJSONWarningIsOnStderr runs `snapshot remove
+// --json` on a snapshot in the local index, against a destination
+// directory that does not exist. The command removes the snapshot from
+// the local index and still exits 0. Its stdout must hold the document
+// alone, with the warning about the destination store on stderr: the
+// command to run again once it is reachable, and the snapshot's ID in
+// the record's snapshot_id field.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout, os.Stderr and the xdg globals
+func TestEntrySnapshotRemoveJSONWarningIsOnStderr(t *testing.T) {
+	configPath, indexPath := writeMissingDestinationConfig(t)
+	seedStaleSnapshotRecord(t, indexPath)
+
+	code, stdout, stderr := runEntry(t, flagConfig, configPath,
+		cmdSnapshot, cmdRemove, stalePruneSnapshotID, flagJSON)
+
+	require.Equal(t, 0, code)
+	requireExactlyOneJSONDocument(t, stdout)
+
+	// stderr is a pipe here, so the logger writes one JSON record a line.
+	var warning map[string]any
+
+	for line := range strings.Lines(stderr) {
+		if strings.Contains(line,
+			"Could not remove snapshot metadata from remote storage") {
+			require.NoError(t, json.Unmarshal([]byte(line), &warning))
+		}
+	}
+
+	require.NotNil(t, warning, "the warning must reach stderr")
+	assert.Contains(t, warning[slog.MessageKey],
+		"run 'vaultik snapshot remove' with the snapshot's ID again")
+	assert.Equal(t, stalePruneSnapshotID, warning["snapshot_id"])
+}
+
+// writeMissingDestinationConfig builds a config whose destination
+// directory does not exist. Returns the config path and the path of
+// its local index, which is not created here.
+func writeMissingDestinationConfig(t *testing.T) (string, string) {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	indexPath := filepath.Join(dir, "index.sqlite")
+
+	contents := fmt.Sprintf(hermeticConfig,
+		filepath.Join(dir, "source"),
+		filepath.Join(dir, "missing-store"),
+		indexPath)
+
+	require.NoError(t,
+		os.WriteFile(configPath, []byte(contents), configFileMode))
+
+	// The PID lock lives under xdg.DataHome, which xdg resolves at
+	// package init; point it at the temp dir so the test neither
+	// touches nor collides with the real one.
+	t.Setenv("XDG_DATA_HOME", filepath.Join(dir, "data"))
+	xdg.Reload()
+	t.Cleanup(xdg.Reload)
+
+	return configPath, indexPath
+}
+
+// writeUnusableDestinationConfig builds a config whose destination
+// directory does not exist, which fails `remote info`, and whose local
+// index is bound to another destination, which fails `prune` and
+// `snapshot remove` (a missing destination alone only makes `snapshot
+// remove` warn). Returns the config path.
+func writeUnusableDestinationConfig(t *testing.T) string {
+	t.Helper()
+
+	configPath, indexPath := writeMissingDestinationConfig(t)
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, indexPath)
+	require.NoError(t, err)
+
+	defer func() { require.NoError(t, db.Close()) }()
+
+	require.NoError(t, database.NewRepositories(db).LocalMeta.Set(ctx,
+		database.LocalMetaKeyStorageURL, "file://"+t.TempDir()))
+
+	return configPath
+}
+
+// runEntry runs Entry with args after the program name and returns its
+// exit code and what it wrote to stdout and stderr.
+//
+// Not parallel-safe: it replaces os.Args, os.Stdout and os.Stderr.
+func runEntry(t *testing.T, args ...string) (int, string, string) {
+	t.Helper()
+
+	previousArgs := os.Args
+
+	t.Cleanup(func() {
+		os.Args = previousArgs
+		rootFlags = RootFlags{}
+	})
+
+	os.Args = append([]string{programName}, args...)
+
+	var code int
+
+	stdout, stderr := captureProcessStdoutAndStderr(t,
+		func() { code = Entry() })
+
+	return code, stdout, stderr
+}

internal/cli/entry_unknown_subcommand_test.go

@@ -0,0 +1,43 @@
+package cli //nolint:testpackage // shares runEntry and the command-word constants
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// TestEntryGroupRejectsUnknownSubcommand runs each group command with a
+// mistyped subcommand, as a crontab line might carry one. It must fail
+// with exit 1 and name the word, as `vaultik bogus` does at the top
+// level, instead of printing the group's help and exiting 0 having done
+// nothing. The group name alone still prints its help.
+//
+//nolint:paralleltest // replaces os.Args, os.Stdout and os.Stderr
+func TestEntryGroupRejectsUnknownSubcommand(t *testing.T) {
+	for _, testCase := range []struct {
+		group    string
+		mistyped string
+	}{
+		{group: cmdSnapshot, mistyped: "craete"},
+		{group: "config", mistyped: "inti"},
+		{group: cmdRemote, mistyped: "ifno"},
+		{group: "database", mistyped: "delte"},
+		{group: "completion", mistyped: "nosuchshell"},
+	} {
+		t.Run(testCase.group, func(t *testing.T) {
+			code, stdout, stderr := runEntry(t,
+				testCase.group, testCase.mistyped)
+
+			assert.Equal(t, 1, code)
+			assert.Empty(t, stdout)
+			assert.Contains(t, stderr,
+				`unknown command "`+testCase.mistyped+`"`)
+
+			code, stdout, _ = runEntry(t, testCase.group)
+
+			assert.Equal(t, 0, code)
+			assert.Contains(t, stdout,
+				programName+" "+testCase.group+" [command]")
+		})
+	}
+}

internal/cli/json_help_test.go

@@ -0,0 +1,24 @@
+package cli //nolint:testpackage // exercises the unexported command constructor
+
+import (
+	"testing"
+
+	"github.com/spf13/cobra"
+	"github.com/stretchr/testify/assert"
+)
+
+// TestJSONHelpSaysConfirmationPromptIsSkipped checks that the --json help
+// of `snapshot remove` and `prune` says the flag skips the confirmation
+// prompt. Both delete without asking under --json, since a prompt on
+// stdout would break the JSON document.
+func TestJSONHelpSaysConfirmationPromptIsSkipped(t *testing.T) {
+	t.Parallel()
+
+	for _, cmd := range []*cobra.Command{
+		newSnapshotRemoveCommand(),
+		NewPruneCommand(),
+	} {
+		assert.Contains(t, cmd.Flags().Lookup("json").Usage,
+			"skips the confirmation prompt", cmd.Name())
+	}
+}

internal/cli/prune.go

@@ -22,9 +22,11 @@ scans every snapshot manifest in the destination store, builds the
 set of still-referenced blob hashes, and deletes any blob not in that
 set.
 
-Snapshot create --prune and snapshot remove run the same cleanup
+Snapshot create --prune runs the same cleanup automatically; this
-automatically; this command is the manual entry point for the same
+command is the manual entry point for the same work (e.g. after a
-work (e.g. after a crashed backup or to reclaim storage).`,
+crashed backup or to reclaim storage). Snapshot remove leaves blobs in
+place; run this command afterwards to delete the ones no longer
+referenced.`,
 		Args: cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
 			// Use unified config resolution
@@ -48,10 +50,6 @@ work (e.g. after a crashed backup or to reclaim storage).`,
 			}, func(v *vaultik.Vaultik) error {
 				return v.Prune(opts)
 			}, func(err error) {
-				if opts.JSON {
-					return
-				}
-
 				log.Error("Prune operation failed", "error", err)
 				ReportErrorf("Prune failed: %v", err)
 			})
@@ -59,7 +57,8 @@ work (e.g. after a crashed backup or to reclaim storage).`,
 	}
 
 	cmd.Flags().BoolVar(&opts.Force, "force", false, "Skip confirmation prompt")
-	cmd.Flags().BoolVar(&opts.JSON, "json", false, "Output pruning stats as JSON")
+	cmd.Flags().BoolVar(&opts.JSON, "json", false,
+		"Output pruning stats as JSON; skips the confirmation prompt, as --force does")
 
 	return cmd
 }

internal/cli/remote.go

@@ -18,6 +18,12 @@ func NewRemoteCommand() *cobra.Command {
 		Use:   "remote",
 		Short: "Remote storage management commands",
 		Long:  "Commands for inspecting and managing remote storage",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands
@@ -45,7 +51,7 @@ This is destructive and irreversible. Requires --force.`,
 				return errNukeNeedsForce
 			}
 
-			return runVaultikApp(cmd, mutating, false, false, "Remote nuke failed",
+			return runVaultikApp(cmd, mutating, false, "Remote nuke failed",
 				func(v *vaultik.Vaultik) error {
 					return v.NukeRemote(true)
 				})
@@ -92,10 +98,6 @@ func newRemoteInfoCommand() *cobra.Command {
 			}, func(v *vaultik.Vaultik) error {
 				return v.RemoteInfo(jsonOutput)
 			}, func(err error) {
-				if jsonOutput {
-					return
-				}
-
 				log.Error("Failed to get remote info", "error", err)
 				ReportErrorf("Failed to get remote info: %v", err)
 			})

internal/cli/root.go

@@ -74,6 +74,21 @@ on the source system.`,
 		NewDatabaseCommand(),
 	)
 
+	// cobra adds the completion command only when Execute runs, and with
+	// no Run, so it skipped its own cobra.NoArgs and a mistyped shell name
+	// printed help and exited 0. Adding it here lets it have a Run. Its
+	// scripts go to os.Stdout as it is now: a later SetOut does not
+	// redirect them.
+	cmd.InitDefaultCompletionCmd()
+
+	for _, sub := range cmd.Commands() {
+		if sub.Name() == "completion" {
+			sub.Run = func(completion *cobra.Command, _ []string) {
+				_ = completion.Help()
+			}
+		}
+	}
+
 	return cmd
 }
 

internal/cli/snapshot.go

@@ -41,6 +41,12 @@ func NewSnapshotCommand() *cobra.Command {
 		Use:   "snapshot",
 		Short: "Snapshot management commands",
 		Long:  "Commands for creating, listing, and managing snapshots",
+		// cobra checks Args only on a command that has a Run: without
+		// one, a mistyped subcommand printed help and exited 0.
+		Args: cobra.NoArgs,
+		Run: func(cmd *cobra.Command, _ []string) {
+			_ = cmd.Help()
+		},
 	}
 
 	// Add subcommands
@@ -66,10 +72,16 @@ func newSnapshotCreateCommand() *cobra.Command {
 If snapshot names are provided, only those snapshots are created.
 If no names are provided, all configured snapshots are created.
 
-Config is located at /etc/vaultik/config.yml by default, but can be overridden by
+The config is read from the path given by --config or VAULTIK_CONFIG;
-specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
+otherwise from the platform config directory (~/.config/vaultik/config.yml
+on Linux), then /etc/vaultik/config.yml.`,
 		Args: cobra.ArbitraryArgs,
 		RunE: func(cmd *cobra.Command, args []string) error {
+			err := opts.Validate()
+			if err != nil {
+				return err
+			}
+
 			// Pass snapshot names from args
 			opts.Snapshots = args
 			// --skip-errors is a global flag on the root command.
@@ -96,6 +108,13 @@ specifying a path using --config or by setting VAULTIK_CONFIG to a path.`,
 			}, func(v *vaultik.Vaultik) error {
 				return v.CreateSnapshot(opts)
 			}, func(err error) {
+				if errors.Is(err, vaultik.ErrPostBackupPruneFailed) {
+					log.Error("Post-backup prune failed", "error", err)
+					ReportErrorf("Backup completed, but %v", err)
+
+					return
+				}
+
 				log.Error("Snapshot creation failed", "error", err)
 				ReportErrorf("Snapshot creation failed: %v", err)
 			})
@@ -126,7 +145,7 @@ func newSnapshotListCommand() *cobra.Command {
 		Long:    "Lists all snapshots with their ID, timestamp, and compressed size",
 		Args:    cobra.NoArgs,
 		RunE: func(cmd *cobra.Command, _ []string) error {
-			return runVaultikApp(cmd, readOnly, false, false,
+			return runVaultikApp(cmd, readOnly, false,
 				"Failed to list snapshots",
 				func(v *vaultik.Vaultik) error {
 					return v.ListSnapshots(jsonOutput)
@@ -162,7 +181,7 @@ restrict the operation to specific snapshot names.`,
 				return errPurgeCriteriaBoth
 			}
 
-			return runVaultikApp(cmd, mutating, false, false,
+			return runVaultikApp(cmd, mutating, false,
 				"Failed to purge snapshots",
 				func(v *vaultik.Vaultik) error {
 					return v.PurgeSnapshotsWithOptions(opts)
@@ -254,18 +273,23 @@ manifest (the destination store may hold snapshots this host doesn't
 know about), which is what 'vaultik prune' does. On success the command
 prints the exact 'vaultik prune' invocation to run as a follow-up.
 
+An ID that names no snapshot in the local index or on the destination
+store is an error.
+
 Use --local-only to skip the remote half (e.g. when you want to forget a
-snapshot locally without touching the destination store).
+snapshot locally without touching the destination store); the snapshot
+must then be in the local index.
 
 If the remote is unreachable, the local-database removal still completes
-and a warning is emitted; rerun 'vaultik prune' once the destination store
+and a warning is emitted; run 'vaultik snapshot remove <snapshot-id>' again
-is reachable to finish remote cleanup.
+once the destination store is reachable to remove the snapshot's metadata
+from it ('vaultik prune' does not).
 
 To wipe the entire destination store and start over, use 'vaultik remote
 nuke --force' — it is the single supported entry point for that.`,
 		Args: requireSnapshotIDArg,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			return runVaultikApp(cmd, mutating, opts.JSON, opts.JSON,
+			return runVaultikApp(cmd, mutating, opts.JSON,
 				"Failed to remove snapshot",
 				func(v *vaultik.Vaultik) error {
 					_, err := v.RemoveSnapshot(args[0], opts)
@@ -278,7 +302,8 @@ nuke --force' — it is the single supported entry point for that.`,
 	cmd.Flags().BoolVarP(&opts.Force, "force", "f", false, "Skip confirmation prompt")
 	cmd.Flags().BoolVar(&opts.DryRun, "dry-run", false,
 		"Show what would be removed without removing")
-	cmd.Flags().BoolVar(&opts.JSON, "json", false, "Output result as JSON")
+	cmd.Flags().BoolVar(&opts.JSON, "json", false,
+		"Output result as JSON; skips the confirmation prompt, as --force does")
 	cmd.Flags().BoolVar(&opts.LocalOnly, "local-only", false,
 		"Skip remote cleanup; only touch the local index")
 

internal/config/config.go

@@ -33,11 +33,14 @@ const secretKeyPrefix = "AGE-SECRET-KEY-"
 const (
 	defaultBlobSizeLimit    = Size(10 * 1024 * 1024 * 1024) // 10GB
 	defaultChunkSize        = Size(10 * 1024 * 1024)        // 10MB
-	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MB
+	defaultS3PartSize       = Size(5 * 1024 * 1024)         // 5MiB
 	defaultCompressionLevel = 3
 	minChunkSize            = 1024 * 1024 // 1MB
 	minCompressionLevel     = 1
 	maxCompressionLevel     = 19
+	// S3 accepts a multipart upload part from 5MiB to 5GiB.
+	minS3PartSize = 5 * 1024 * 1024
+	maxS3PartSize = 5 * 1024 * 1024 * 1024
 )
 
 // Sentinel validation errors.
@@ -55,8 +58,12 @@ var (
 		"blob_size_limit must be at least the largest chunk the chunker can " +
 			"emit (chunk_size times the FastCDC size spread)")
 	errBadCompression   = errors.New("compression_level must be between 1 and 19")
+	errBadS3PartSize    = errors.New("s3.part_size must be between 5MiB and 5GiB")
 	errBadStorageScheme = errors.New(
 		"storage_url must start with s3://, file://, or rclone://")
+	errFileURLNotAbsolute = errors.New(
+		"a file:// storage_url needs an absolute path, " +
+			"as in file:///path/to/backup")
 	errStorageNotConfigured = errors.New(
 		"storage not configured; set storage_url or provide s3.endpoint + " +
 			"s3.bucket + credentials")
@@ -244,6 +251,7 @@ func Load(path string) (*Config, error) {
 		ChunkSize:        defaultChunkSize,
 		IndexPath:        filepath.Join(xdg.DataHome, appName, "index.sqlite"),
 		CompressionLevel: defaultCompressionLevel,
+		S3:               S3Config{PartSize: defaultS3PartSize},
 	}
 
 	// Convert smartconfig data to YAML then unmarshal
@@ -294,17 +302,13 @@ func Load(path string) (*Config, error) {
 		cfg.S3.Region = "us-east-1"
 	}
 
-	if cfg.S3.PartSize == 0 {
-		cfg.S3.PartSize = defaultS3PartSize
-	}
-
 	// Check config file permissions (warn if world or group readable)
 	//nolint:gosec // G703: config path is operator-supplied by design
 	info, statErr := os.Stat(path)
 	if statErr == nil {
 		mode := info.Mode().Perm()
 		if mode&0044 != 0 { // group or world readable
-			log.Warn("Config file has insecure permissions (contains S3 credentials)",
+			log.Warn(cfg.readableByOthersWarning(),
 				"path", path,
 				"mode", fmt.Sprintf("%04o", mode),
 				"recommendation", "chmod 600 "+path)
@@ -332,6 +336,7 @@ func Load(path string) (*Config, error) {
 //     (chunk_size times chunker.ChunkSizeSpread), so a single-chunk blob never
 //     exceeds the configured limit
 //   - Compression level must be between 1 and 19
+//   - S3 part size must be between 5MiB and 5GiB, the part sizes S3 accepts
 //
 // Returns an error describing the first validation failure encountered.
 func (c *Config) Validate() error {
@@ -376,6 +381,11 @@ func (c *Config) Validate() error {
 		return errBadCompression
 	}
 
+	if c.S3.PartSize.Int64() < minS3PartSize ||
+		c.S3.PartSize.Int64() > maxS3PartSize {
+		return errBadS3PartSize
+	}
+
 	return nil
 }
 
@@ -411,6 +421,18 @@ func (c *Config) setAgeSecretKey() {
 	}
 }
 
+// readableByOthersWarning is the warning Load logs when others can read
+// the config file. It says "may contain" because the S3 credentials are
+// seen only after smartconfig has replaced any ${...} reference in the
+// file with its value, so a set credential need not be in the file.
+func (c *Config) readableByOthersWarning() string {
+	if c.S3.AccessKeyID != "" || c.S3.SecretAccessKey != "" {
+		return "Config file is readable by others and may contain S3 credentials"
+	}
+
+	return "Config file is readable by others"
+}
+
 // validateStorage validates storage configuration.
 // If StorageURL is set, it takes precedence. S3 URLs require credentials.
 // File URLs don't require any S3 configuration.
@@ -442,11 +464,18 @@ func (c *Config) validateStorage() error {
 
 // validateStorageURL validates URL-based storage configuration. File and
 // rclone URLs need no credentials; S3 URLs require the legacy s3.*
-// credential fields.
+// credential fields. A file URL must name an absolute path: a relative one
+// would resolve against the working directory, so the destination would
+// move with it.
 func (c *Config) validateStorageURL() error {
 	switch {
 	case strings.HasPrefix(c.StorageURL, "file://"):
-		// File storage doesn't need S3 credentials
+		// storage.ParseStorageURL applies the same rule, but config
+		// cannot import storage, which imports config.
+		if !strings.HasPrefix(c.StorageURL, "file:///") {
+			return errFileURLNotAbsolute
+		}
+
 		return nil
 	case strings.HasPrefix(c.StorageURL, "rclone://"):
 		// Rclone storage uses rclone's own config

internal/config/config_test.go

@@ -8,6 +8,7 @@ import (
 	"testing"
 
 	"sneak.berlin/go/vaultik/internal/chunker"
+	"sneak.berlin/go/vaultik/internal/log"
 )
 
 const (
@@ -166,6 +167,7 @@ func TestValidateBlobSizeLimit(t *testing.T) {
 			ChunkSize:        chunkSize,
 			BlobSizeLimit:    blobLimit,
 			CompressionLevel: 3,
+			S3:               S3Config{PartSize: defaultS3PartSize},
 		}
 	}
 
@@ -221,6 +223,149 @@ func TestValidateBlobSizeLimit(t *testing.T) {
 	}
 }
 
+// TestValidateS3PartSize checks that s3.part_size is held to the part sizes
+// S3 accepts, 5MiB to 5GiB, by changing only the part size of the test
+// config. "5MB" in the config file is 5,000,000 bytes, below the minimum.
+func TestValidateS3PartSize(t *testing.T) {
+	t.Parallel()
+
+	base, err := Load(os.Getenv("VAULTIK_CONFIG"))
+	if err != nil {
+		t.Fatalf("Failed to load config: %v", err)
+	}
+
+	tests := []struct {
+		name     string
+		partSize Size
+		wantErr  bool
+	}{
+		{
+			name:     "5MB is rejected",
+			partSize: 5_000_000,
+			wantErr:  true,
+		},
+		{
+			name:     "one byte below 5MiB is rejected",
+			partSize: minS3PartSize - 1,
+			wantErr:  true,
+		},
+		{
+			name:     "5MiB is accepted",
+			partSize: minS3PartSize,
+			wantErr:  false,
+		},
+		{
+			name:     "5GiB is accepted",
+			partSize: maxS3PartSize,
+			wantErr:  false,
+		},
+		{
+			name:     "one byte above 5GiB is rejected",
+			partSize: maxS3PartSize + 1,
+			wantErr:  true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			cfg := *base
+			cfg.S3.PartSize = tt.partSize
+
+			err := cfg.Validate()
+			if tt.wantErr {
+				if !errors.Is(err, errBadS3PartSize) {
+					t.Fatalf("Validate() error = %v, want errBadS3PartSize", err)
+				}
+
+				return
+			}
+
+			if err != nil {
+				t.Fatalf("Validate() unexpected error: %v", err)
+			}
+		})
+	}
+}
+
+// TestLoadS3PartSize checks that a config file without s3.part_size loads
+// with the 5MiB default, and that an explicit 0 fails at load like any other
+// part size S3 refuses.
+func TestLoadS3PartSize(t *testing.T) {
+	t.Parallel()
+
+	const withoutPartSize = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file:///tmp/vaultik-test-storage\n"
+
+	writeConfig := func(t *testing.T, text string) string {
+		t.Helper()
+
+		path := filepath.Join(t.TempDir(), "config.yml")
+
+		err := os.WriteFile(path, []byte(text), 0o600)
+		if err != nil {
+			t.Fatalf("write config: %v", err)
+		}
+
+		return path
+	}
+
+	t.Run("absent loads as 5MiB", func(t *testing.T) {
+		t.Parallel()
+
+		cfg, err := Load(writeConfig(t, withoutPartSize))
+		if err != nil {
+			t.Fatalf("Load() unexpected error: %v", err)
+		}
+
+		if cfg.S3.PartSize != defaultS3PartSize {
+			t.Errorf("s3.part_size = %d, want %d",
+				cfg.S3.PartSize, defaultS3PartSize)
+		}
+	})
+
+	t.Run("0 is rejected", func(t *testing.T) {
+		t.Parallel()
+
+		_, err := Load(writeConfig(t, withoutPartSize+"s3:\n  part_size: 0\n"))
+		if !errors.Is(err, errBadS3PartSize) {
+			t.Fatalf("Load() error = %v, want errBadS3PartSize", err)
+		}
+	})
+}
+
+// TestLoadRejectsRelativeFileURL checks that a file:// storage_url with a
+// relative path fails at load. Accepted, it would resolve against the
+// working directory, so a backup run from another directory would write
+// to a different destination and still succeed.
+func TestLoadRejectsRelativeFileURL(t *testing.T) {
+	t.Parallel()
+
+	const relativeConfig = "snapshots:\n" +
+		"  test:\n" +
+		"    paths: [/tmp/vaultik-test-source]\n" +
+		"storage_url: file://dest\n"
+
+	path := filepath.Join(t.TempDir(), "config.yml")
+
+	err := os.WriteFile(path, []byte(relativeConfig), 0o600)
+	if err != nil {
+		t.Fatalf("write config: %v", err)
+	}
+
+	_, err = Load(path)
+	if !errors.Is(err, errFileURLNotAbsolute) {
+		t.Fatalf("Load() error = %v, want errFileURLNotAbsolute", err)
+	}
+
+	if !strings.Contains(err.Error(), "absolute path") {
+		t.Errorf("error %q does not say to use an absolute path", err.Error())
+	}
+}
+
 // TestValidateAgeRecipients checks that recipients are parsed at config load
 // (a bad entry fails immediately, not mid-backup) and that no invalid entry —
 // least of all a pasted secret key — is echoed in the error. An empty list
@@ -236,6 +381,7 @@ func TestValidateAgeRecipients(t *testing.T) {
 			ChunkSize:        Size(10 * 1024 * 1024),
 			BlobSizeLimit:    Size(10 * 1024 * 1024 * 1024),
 			CompressionLevel: 3,
+			S3:               S3Config{PartSize: defaultS3PartSize},
 		}
 	}
 
@@ -343,3 +489,125 @@ func TestAgeSecretKeySourceName(t *testing.T) {
 		})
 	}
 }
+
+// loadReadableConfig writes configYAML to a file that others can read,
+// loads it, and returns what the logger wrote to stderr meanwhile. The
+// logger writes to the os.Stderr it finds when it is initialized, so
+// os.Stderr is pointed at a file first. Not parallel-safe: os.Stderr and
+// the logger are process-global.
+func loadReadableConfig(t *testing.T, configYAML string) string {
+	t.Helper()
+
+	dir := t.TempDir()
+	configPath := filepath.Join(dir, "config.yml")
+	stderrPath := filepath.Join(dir, "stderr")
+
+	err := os.WriteFile(configPath, []byte(configYAML), 0o600)
+	if err != nil {
+		t.Fatalf("writing config: %v", err)
+	}
+
+	//nolint:gosec // G302: the test needs a config file others can read
+	err = os.Chmod(configPath, 0o644)
+	if err != nil {
+		t.Fatalf("chmod config: %v", err)
+	}
+
+	stderrFile, err := os.Create(stderrPath) //nolint:gosec // G304: test temp path
+	if err != nil {
+		t.Fatalf("creating stderr file: %v", err)
+	}
+
+	previous := os.Stderr
+	os.Stderr = stderrFile
+
+	log.Initialize(log.Config{})
+
+	_, loadErr := Load(configPath)
+
+	os.Stderr = previous
+
+	log.Initialize(log.Config{})
+
+	_ = stderrFile.Close()
+
+	if loadErr != nil {
+		t.Fatalf("Load() error = %v", loadErr)
+	}
+
+	captured, err := os.ReadFile(stderrPath) //nolint:gosec // G304: test temp path
+	if err != nil {
+		t.Fatalf("reading stderr file: %v", err)
+	}
+
+	return string(captured)
+}
+
+// TestLoadWarnsReadableConfigWithoutS3Credentials checks that a config
+// file others can read, holding no S3 credentials, is warned about
+// without a claim that it holds them.
+//
+//nolint:paralleltest // loadReadableConfig replaces os.Stderr
+func TestLoadWarnsReadableConfigWithoutS3Credentials(t *testing.T) {
+	stderr := loadReadableConfig(t, `
+storage_url: file:///var/backups/vaultik
+snapshots:
+  home:
+    paths:
+      - /home
+`)
+
+	if !strings.Contains(stderr, "Config file is readable by others") {
+		t.Errorf("expected a warning that the file is readable by others, got %q",
+			stderr)
+	}
+
+	if strings.Contains(stderr, "S3 credentials") {
+		t.Errorf("warning names S3 credentials the file does not set: %q", stderr)
+	}
+}
+
+// TestLoadWarnsReadableConfigWithS3Credentials checks that a config file
+// others can read and that sets S3 credentials, as values or as ${ENV:...}
+// references, is warned about as one that may contain them.
+//
+//nolint:paralleltest // loadReadableConfig replaces os.Stderr
+func TestLoadWarnsReadableConfigWithS3Credentials(t *testing.T) {
+	t.Setenv("VAULTIK_TEST_ACCESS_KEY_ID", "test-access-key")
+	t.Setenv("VAULTIK_TEST_SECRET_ACCESS_KEY", "test-secret-key")
+
+	configs := map[string]string{
+		"values": `
+storage_url: s3://bucket/prefix?endpoint=s3.example.com
+s3:
+  access_key_id: test-access-key
+  secret_access_key: test-secret-key
+snapshots:
+  home:
+    paths:
+      - /home
+`,
+		"references": `
+storage_url: s3://bucket/prefix?endpoint=s3.example.com
+s3:
+  access_key_id: ${ENV:VAULTIK_TEST_ACCESS_KEY_ID}
+  secret_access_key: ${ENV:VAULTIK_TEST_SECRET_ACCESS_KEY}
+snapshots:
+  home:
+    paths:
+      - /home
+`,
+	}
+
+	for name, configYAML := range configs {
+		t.Run(name, func(t *testing.T) {
+			stderr := loadReadableConfig(t, configYAML)
+
+			if !strings.Contains(stderr,
+				"Config file is readable by others and may contain S3 credentials") {
+				t.Errorf("expected a warning naming the S3 credentials, got %q",
+					stderr)
+			}
+		})
+	}
+}

internal/database/blob_chunks.go

@@ -58,12 +58,7 @@ func (r *BlobChunkRepository) GetByBlobID(
 		return nil, fmt.Errorf("querying blob chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var blobChunks []*BlobChunk
 

internal/database/blobs.go

@@ -83,12 +83,7 @@ func (r *BlobRepository) GetAll(ctx context.Context) (map[string]*Blob, error) {
 		return nil, fmt.Errorf("querying blobs: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	out := make(map[string]*Blob)
 
@@ -159,13 +154,16 @@ func (r *BlobRepository) UpdateFinished(
 	return nil
 }
 
-// UpdateUploaded marks a blob as uploaded
+// UpdateUploaded marks a blob as uploaded and records the size of the
+// stored object as its compressed size. That size can differ from the
+// local copy's when an identical blob was already stored with another
+// compression level or other recipients.
 func (r *BlobRepository) UpdateUploaded(
-	ctx context.Context, tx *sql.Tx, id string,
+	ctx context.Context, tx *sql.Tx, id string, storedSize int64,
 ) error {
 	query := `
-		UPDATE blobs 
+		UPDATE blobs
-		SET uploaded_ts = ?
+		SET uploaded_ts = ?, compressed_size = ?
 		WHERE id = ?
 	`
 
@@ -173,9 +171,9 @@ func (r *BlobRepository) UpdateUploaded(
 
 	var err error
 	if tx != nil {
-		_, err = tx.ExecContext(ctx, query, now, id)
+		_, err = tx.ExecContext(ctx, query, now, storedSize, id)
 	} else {
-		_, err = r.db.ExecWithLog(ctx, query, now, id)
+		_, err = r.db.ExecWithLog(ctx, query, now, storedSize, id)
 	}
 
 	if err != nil {

internal/database/blobs_test.go

@@ -122,8 +122,8 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 		t.Errorf("expected compressed size 500, got %d", updated.CompressedSize)
 	}
 
-	// Test UpdateUploaded
+	// Test UpdateUploaded, with a stored object larger than the local copy
-	err = repo.UpdateUploaded(ctx, nil, blob.ID.String())
+	err = repo.UpdateUploaded(ctx, nil, blob.ID.String(), 600)
 	if err != nil {
 		t.Fatalf("failed to update blob as uploaded: %v", err)
 	}
@@ -141,6 +141,10 @@ func TestBlobRepositoryUpdates(t *testing.T) {
 	if uploaded.UploadedTS.Before(now.Add(-1 * time.Second)) {
 		t.Error("uploaded timestamp should be around test time")
 	}
+
+	if uploaded.CompressedSize != 600 {
+		t.Errorf("expected the stored size 600, got %d", uploaded.CompressedSize)
+	}
 }
 
 func TestBlobRepositoryDuplicate(t *testing.T) {

internal/database/chunk_files.go

@@ -61,12 +61,7 @@ func (r *ChunkFileRepository) GetByChunkHash(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanChunkFiles(rows)
 }
@@ -87,12 +82,7 @@ func (r *ChunkFileRepository) GetByFilePath(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanChunkFiles(rows)
 }
@@ -112,12 +102,7 @@ func (r *ChunkFileRepository) GetByFileID(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanChunkFiles(rows)
 }

internal/database/chunks.go

@@ -107,12 +107,7 @@ func (r *ChunkRepository) GetByHashes(
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var chunks []*Chunk
 
@@ -152,12 +147,7 @@ func (r *ChunkRepository) ListUnpacked(
 		return nil, fmt.Errorf("querying unpacked chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var chunks []*Chunk
 

internal/database/chunks_ext.go

@@ -38,12 +38,7 @@ func (r *ChunkRepository) list(ctx context.Context, query string) ([]*Chunk, err
 		return nil, fmt.Errorf("querying chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var chunks []*Chunk
 

internal/database/errors.go

@@ -1,12 +0,0 @@
-package database
-
-import (
-	"fmt"
-	"os"
-)
-
-// Fatalf prints an error message to stderr and exits with status 1
-func Fatalf(format string, args ...any) {
-	fmt.Fprintf(os.Stderr, "FATAL: "+format+"\n", args...)
-	os.Exit(1)
-}

internal/database/file_chunks.go

@@ -62,12 +62,7 @@ func (r *FileChunkRepository) GetByPath(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanFileChunks(rows)
 }
@@ -88,12 +83,7 @@ func (r *FileChunkRepository) GetByFileID(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanFileChunks(rows)
 }
@@ -117,12 +107,7 @@ func (r *FileChunkRepository) GetByPathTx(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	fileChunks, err := r.scanFileChunks(rows)
 	LogSQL("GetByPathTx", "Complete", path, "count", len(fileChunks))

internal/database/files.go

@@ -33,11 +33,13 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 	}
 
 	query := `
-		INSERT INTO files (id, path, source_path, mtime, size, mode, uid, gid, link_target)
+		INSERT INTO files
-		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+			(id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target)
+		VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
 		ON CONFLICT(path) DO UPDATE SET
 			source_path = excluded.source_path,
 			mtime = excluded.mtime,
+			mtime_nsec = excluded.mtime_nsec,
 			size = excluded.size,
 			mode = excluded.mode,
 			uid = excluded.uid,
@@ -54,16 +56,19 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 	if tx != nil {
 		LogSQL("Execute", query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String())
 		err = tx.QueryRowContext(ctx, query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String()).Scan(&idStr)
 	} else {
 		err = r.db.QueryRowWithLog(ctx, query,
 			file.ID.String(), file.Path.String(), file.SourcePath.String(),
-			file.MTime.Unix(), file.Size, file.Mode, file.UID, file.GID,
+			file.MTime.Unix(), file.MTime.Nanosecond(),
+			file.Size, file.Mode, file.UID, file.GID,
 			file.LinkTarget.String()).Scan(&idStr)
 	}
 
@@ -84,7 +89,7 @@ func (r *FileRepository) Create(ctx context.Context, tx *sql.Tx, file *File) err
 // in the index.
 func (r *FileRepository) GetByPath(ctx context.Context, path string) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE path = ?
 	`
@@ -104,7 +109,7 @@ func (r *FileRepository) GetByPath(ctx context.Context, path string) (*File, err
 // GetByID retrieves a file by its UUID
 func (r *FileRepository) GetByID(ctx context.Context, id types.FileID) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE id = ?
 	`
@@ -127,7 +132,7 @@ func (r *FileRepository) GetByPathTx(
 	ctx context.Context, tx *sql.Tx, path string,
 ) (*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE path = ?
 	`
@@ -158,23 +163,19 @@ func (r *FileRepository) ListModifiedSince(
 	ctx context.Context, since time.Time,
 ) ([]*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
-		WHERE mtime >= ?
+		WHERE (mtime, mtime_nsec) >= (?, ?)
 		ORDER BY path
 	`
 
-	rows, err := r.db.conn.QueryContext(ctx, query, since.Unix())
+	rows, err := r.db.conn.QueryContext(ctx, query,
+		since.Unix(), since.Nanosecond())
 	if err != nil {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var files []*File
 
@@ -239,7 +240,7 @@ func (r *FileRepository) ListUnderPath(
 
 	// LIKE would ignore ASCII case and treat _ and % in path as wildcards.
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		WHERE path = ? OR substr(path, 1, length(?)) = ?
 		ORDER BY path
@@ -250,12 +251,7 @@ func (r *FileRepository) ListUnderPath(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var files []*File
 
@@ -298,12 +294,7 @@ func (r *FileRepository) ListIDsWithChunksNotInUploadedBlobs(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var ids []types.FileID
 
@@ -324,7 +315,7 @@ func (r *FileRepository) ListIDsWithChunksNotInUploadedBlobs(
 // ListAll returns all files in the database
 func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 	query := `
-		SELECT id, path, source_path, mtime, size, mode, uid, gid, link_target
+		SELECT id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target
 		FROM files
 		ORDER BY path
 	`
@@ -334,12 +325,7 @@ func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var files []*File
 
@@ -365,7 +351,7 @@ func (r *FileRepository) CreateBatch(
 	}
 
 	// Each files row binds this many SQL variables.
-	const fileCols = 9
+	const fileCols = 10
 
 	// Batch at 100 rows to be safe with SQLite's variable limit.
 	const batchSize = 100
@@ -376,7 +362,7 @@ func (r *FileRepository) CreateBatch(
 		batch := files[i:end]
 
 		query := `INSERT INTO files
-			(id, path, source_path, mtime, size, mode, uid, gid, link_target)
+			(id, path, source_path, mtime, mtime_nsec, size, mode, uid, gid, link_target)
 			VALUES `
 
 		args := make([]any, 0, len(batch)*fileCols)
@@ -388,11 +374,12 @@ func (r *FileRepository) CreateBatch(
 				querySb325.WriteString(", ")
 			}
 
-			querySb325.WriteString("(?, ?, ?, ?, ?, ?, ?, ?, ?)")
+			querySb325.WriteString("(?, ?, ?, ?, ?, ?, ?, ?, ?, ?)")
 
 			args = append(args,
 				f.ID.String(), f.Path.String(), f.SourcePath.String(),
-				f.MTime.Unix(), f.Size, f.Mode, f.UID, f.GID,
+				f.MTime.Unix(), f.MTime.Nanosecond(),
+				f.Size, f.Mode, f.UID, f.GID,
 				f.LinkTarget.String())
 		}
 
@@ -401,6 +388,7 @@ func (r *FileRepository) CreateBatch(
 		query += ` ON CONFLICT(path) DO UPDATE SET
 			source_path = excluded.source_path,
 			mtime = excluded.mtime,
+			mtime_nsec = excluded.mtime_nsec,
 			size = excluded.size,
 			mode = excluded.mode,
 			uid = excluded.uid,
@@ -460,7 +448,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 	var (
 		file                          File
 		idStr, pathStr, sourcePathStr string
-		mtimeUnix                     int64
+		mtimeUnix, mtimeNsec          int64
 		linkTarget                    sql.NullString
 	)
 
@@ -469,6 +457,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 		&pathStr,
 		&sourcePathStr,
 		&mtimeUnix,
+		&mtimeNsec,
 		&file.Size,
 		&file.Mode,
 		&file.UID,
@@ -487,7 +476,7 @@ func (r *FileRepository) scanFileFrom(row fileRowScanner) (*File, error) {
 	file.Path = types.FilePath(pathStr)
 	file.SourcePath = types.SourcePath(sourcePathStr)
 
-	file.MTime = time.Unix(mtimeUnix, 0).UTC()
+	file.MTime = time.Unix(mtimeUnix, mtimeNsec).UTC()
 	if linkTarget.Valid {
 		file.LinkTarget = types.FilePath(linkTarget.String)
 	}

internal/database/files_test.go

@@ -252,6 +252,115 @@ func TestFileRepositorySymlink(t *testing.T) {
 	}
 }
 
+// An mtime after 2262 or before 1678 does not fit in int64 nanoseconds
+// since the epoch, and must still come back from the database unchanged.
+func TestFileRepositoryMTimeOutsideInt64NanosecondRange(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewFileRepository(db)
+
+	mtimes := []time.Time{
+		time.Date(2300, time.January, 1, 0, 0, 0, 123456789, time.UTC),
+		time.Date(1601, time.January, 1, 0, 0, 0, 987654321, time.UTC),
+	}
+
+	for _, mtime := range mtimes {
+		created := &database.File{
+			Path:  types.FilePath("/created-" + mtime.Format(time.RFC3339Nano)),
+			MTime: mtime,
+		}
+
+		err := repo.Create(ctx, nil, created)
+		if err != nil {
+			t.Fatalf("failed to create file: %v", err)
+		}
+
+		batched := &database.File{
+			ID:    types.NewFileID(),
+			Path:  types.FilePath("/batched-" + mtime.Format(time.RFC3339Nano)),
+			MTime: mtime,
+		}
+
+		err = repo.CreateBatch(ctx, nil, []*database.File{batched})
+		if err != nil {
+			t.Fatalf("failed to batch create file: %v", err)
+		}
+
+		for _, path := range []types.FilePath{created.Path, batched.Path} {
+			retrieved, err := repo.GetByPath(ctx, path.String())
+			if err != nil {
+				t.Fatalf("failed to get file: %v", err)
+			}
+
+			if !retrieved.MTime.Equal(mtime) {
+				t.Errorf("%s: mtime got %v, want %v",
+					path, retrieved.MTime, mtime)
+			}
+		}
+	}
+}
+
+// A file already in the index and rewritten within the same second must get
+// its new nanoseconds stored, through both Create and CreateBatch.
+func TestFileRepositoryUpsertMTimeInSameSecond(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repo := database.NewFileRepository(db)
+
+	indexed := time.Date(2026, time.October, 7, 12, 0, 0, 100000000, time.UTC)
+	rewritten := indexed.Add(800 * time.Millisecond)
+
+	tests := []struct {
+		name   string
+		upsert func(file *database.File) error
+	}{
+		{"Create", func(file *database.File) error {
+			return repo.Create(ctx, nil, file)
+		}},
+		{"CreateBatch", func(file *database.File) error {
+			return repo.CreateBatch(ctx, nil, []*database.File{file})
+		}},
+	}
+
+	for _, tt := range tests {
+		file := &database.File{
+			ID:    types.NewFileID(),
+			Path:  types.FilePath("/" + tt.name),
+			MTime: indexed,
+		}
+
+		err := tt.upsert(file)
+		if err != nil {
+			t.Fatalf("%s: failed to create file: %v", tt.name, err)
+		}
+
+		file.MTime = rewritten
+
+		err = tt.upsert(file)
+		if err != nil {
+			t.Fatalf("%s: failed to update file: %v", tt.name, err)
+		}
+
+		retrieved, err := repo.GetByPath(ctx, file.Path.String())
+		if err != nil {
+			t.Fatalf("%s: failed to get file: %v", tt.name, err)
+		}
+
+		if !retrieved.MTime.Equal(rewritten) {
+			t.Errorf("%s: mtime got %v, want %v",
+				tt.name, retrieved.MTime, rewritten)
+		}
+	}
+}
+
 func TestFileRepositoryTransaction(t *testing.T) {
 	t.Parallel()
 

internal/database/models.go

@@ -14,8 +14,7 @@ type File struct {
 	ID   types.FileID   // UUID primary key
 	Path types.FilePath // Absolute path of the file
 
-	// SourcePath is the source directory this file came from (used for
+	// SourcePath is the source directory this file came from.
-	// restore path stripping).
 	SourcePath types.SourcePath
 	MTime      time.Time
 	Size       int64

internal/database/repositories.go

@@ -3,6 +3,7 @@ package database
 import (
 	"context"
 	"database/sql"
+	"errors"
 	"fmt"
 )
 
@@ -60,22 +61,24 @@ func (r *Repositories) WithTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
+			// The panic is the failure to report, so a rollback error
-			if rollbackErr != nil {
+			// is dropped.
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
+			_ = tx.Rollback()
-			}
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 
@@ -103,22 +106,24 @@ func (r *Repositories) WithReadTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			rollbackErr := tx.Rollback()
+			// The panic is the failure to report, so a rollback error
-			if rollbackErr != nil {
+			// is dropped.
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
+			_ = tx.Rollback()
-			}
 
 			panic(p)
-		} else if err != nil {
-			rollbackErr := tx.Rollback()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
 		}
 	}()
 
 	err = fn(ctx, tx)
 	if err != nil {
+		// Once ctx is cancelled, database/sql rolls the transaction
+		// back itself and Rollback returns sql.ErrTxDone.
+		rollbackErr := tx.Rollback()
+		if rollbackErr != nil && !errors.Is(rollbackErr, sql.ErrTxDone) {
+			return fmt.Errorf("%w; rolling back transaction: %w",
+				err, rollbackErr)
+		}
+
 		return err
 	}
 

internal/database/repositories_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"database/sql"
 	"errors"
+	"strings"
 	"testing"
 	"time"
 
@@ -235,6 +236,125 @@ func TestRepositoriesTransactionRollback(t *testing.T) {
 	}
 }
 
+// When the context is cancelled during the body, database/sql rolls the
+// transaction back itself, so WithTx's own rollback finds it already
+// done. WithTx must return the body's error alone, without that rollback.
+func TestRepositoriesTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		file := &database.File{
+			Path:  "/test/cancelled_file.txt",
+			MTime: time.Now().Truncate(time.Second),
+			Size:  1024,
+			Mode:  0644,
+			UID:   1000,
+			GID:   1000,
+		}
+
+		err := repos.Files.Create(ctx, tx, file)
+		if err != nil {
+			return err
+		}
+
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err = tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+
+	file, err := repos.Files.GetByPath(
+		context.Background(), "/test/cancelled_file.txt",
+	)
+	if err != nil {
+		t.Fatalf("error checking for file: %v", err)
+	}
+
+	if file != nil {
+		t.Error("file should not exist after the cancelled transaction")
+	}
+}
+
+// A body that ends the transaction itself with ROLLBACK makes WithTx's
+// own rollback fail. WithTx must return that failure together with the
+// body's error.
+func TestRepositoriesTransactionRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	err := repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		return errIntentionalRollback
+	})
+	if !errors.Is(err, errIntentionalRollback) {
+		t.Fatalf("expected the body's error, got: %v", err)
+	}
+
+	if !strings.Contains(err.Error(), "rolling back transaction") {
+		t.Errorf("expected the failed rollback in the error, got: %v", err)
+	}
+}
+
+// When the body ends the transaction itself with ROLLBACK and then
+// panics, WithTx's own rollback fails; the panic must still reach the
+// caller.
+func TestRepositoriesTransactionPanicRollbackFails(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	defer func() {
+		if recover() == nil {
+			t.Error("expected the body's panic to reach the caller")
+		}
+	}()
+
+	_ = repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		_, err := tx.ExecContext(ctx, "ROLLBACK")
+		if err != nil {
+			return err
+		}
+
+		panic(errIntentionalRollback)
+	})
+}
+
 func TestRepositoriesReadTransaction(t *testing.T) {
 	t.Parallel()
 
@@ -291,3 +411,41 @@ func TestRepositoriesReadTransaction(t *testing.T) {
 		t.Error("expected to retrieve file in read transaction")
 	}
 }
+
+// When the context is cancelled during a read transaction's body,
+// database/sql rolls the transaction back itself, so WithReadTx's own
+// rollback finds it already done. WithReadTx must return the body's error
+// alone, without that rollback.
+func TestRepositoriesReadTransactionContextCancelled(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	repos := database.NewRepositories(db)
+
+	err := repos.WithReadTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+		cancel()
+
+		// database/sql rolls back from a goroutine of its own; wait
+		// until it has, so the test does not depend on that race.
+		for {
+			_, err := tx.ExecContext(context.WithoutCancel(ctx), "SELECT 1")
+			if errors.Is(err, sql.ErrTxDone) {
+				return ctx.Err()
+			}
+
+			time.Sleep(time.Millisecond)
+		}
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got: %v", err)
+	}
+
+	if errors.Is(err, sql.ErrTxDone) {
+		t.Fatalf("the already-done rollback was reported: %v", err)
+	}
+}

internal/database/repository_edge_cases_test.go

@@ -3,6 +3,7 @@ package database
 
 import (
 	"context"
+	"database/sql"
 	"fmt"
 	"strings"
 	"testing"
@@ -367,7 +368,7 @@ func verifyBlobNullUploadTS(
 }
 
 // createLargeDatasetFiles creates fileCount files and adds every other
-// one to the snapshot.
+// one to the snapshot, in one transaction as a backup writes them.
 func createLargeDatasetFiles(
 	t *testing.T,
 	repos *Repositories,
@@ -376,31 +377,38 @@ func createLargeDatasetFiles(
 ) {
 	t.Helper()
 
-	ctx := context.Background()
 	start := time.Now()
 
-	for i := range fileCount {
+	err := repos.WithTx(context.Background(),
-		file := &File{
+		func(ctx context.Context, tx *sql.Tx) error {
-			Path:  types.FilePath(fmt.Sprintf("/large/file%05d.txt", i)),
+			for i := range fileCount {
-			MTime: time.Now(),
+				file := &File{
-			Size:  int64(i * 1024),
+					Path:  types.FilePath(fmt.Sprintf("/large/file%05d.txt", i)),
-			Mode:  0644,
+					MTime: time.Now(),
-			UID:   uint32(1000 + (i % 10)),
+					Size:  int64(i * 1024),
-			GID:   uint32(1000 + (i % 10)),
+					Mode:  0644,
-		}
+					UID:   uint32(1000 + (i % 10)),
+					GID:   uint32(1000 + (i % 10)),
+				}
 
-		err := repos.Files.Create(ctx, nil, file)
+				err := repos.Files.Create(ctx, tx, file)
-		if err != nil {
+				if err != nil {
-			t.Fatalf("failed to create file %d: %v", i, err)
+					return fmt.Errorf("creating file %d: %w", i, err)
-		}
+				}
 
-		// Add half to snapshot
+				// Add half to snapshot
-		if i%2 == 0 {
+				if i%2 == 0 {
-			err = repos.Snapshots.AddFileByID(ctx, nil, snapshotID, file.ID)
+					err = repos.Snapshots.AddFileByID(ctx, tx, snapshotID, file.ID)
-			if err != nil {
+					if err != nil {
-				t.Fatal(err)
+						return err
+					}
+				}
 			}
-		}
+
+			return nil
+		})
+	if err != nil {
+		t.Fatal(err)
 	}
 
 	t.Logf("Created %d files in %v", fileCount, time.Since(start))
@@ -606,8 +614,7 @@ func TestTimezoneHandling(t *testing.T) {
 		t.Skip("timezone not available")
 	}
 
-	// Use Truncate to remove sub-second precision since we store as Unix timestamps
+	nyTime := time.Now().In(loc)
-	nyTime := time.Now().In(loc).Truncate(time.Second)
 	file := &File{
 		Path:  "/timezone-test.txt",
 		MTime: nyTime,

internal/database/schema/001.sql

@@ -5,8 +5,9 @@
 CREATE TABLE IF NOT EXISTS files (
     id TEXT PRIMARY KEY,  -- UUID
     path TEXT NOT NULL UNIQUE,
-    source_path TEXT NOT NULL DEFAULT '',  -- The source directory this file came from (for restore path stripping)
+    source_path TEXT NOT NULL DEFAULT '',  -- The source directory this file came from
-    mtime INTEGER NOT NULL,
+    mtime INTEGER NOT NULL,  -- whole seconds since the Unix epoch
+    mtime_nsec INTEGER NOT NULL,  -- nanoseconds within that second, 0 to 999999999
     size INTEGER NOT NULL,
     mode INTEGER NOT NULL,
     uid INTEGER NOT NULL,
@@ -85,6 +86,7 @@ CREATE TABLE IF NOT EXISTS snapshots (
     completed_at INTEGER,
     file_count INTEGER NOT NULL DEFAULT 0,
     chunk_count INTEGER NOT NULL DEFAULT 0,
+    new_chunk_size INTEGER NOT NULL DEFAULT 0,
     blob_count INTEGER NOT NULL DEFAULT 0,
     total_size INTEGER NOT NULL DEFAULT 0,
     blob_size INTEGER NOT NULL DEFAULT 0,

internal/database/snapshots.go

@@ -129,6 +129,7 @@ func (r *SnapshotRepository) UpdateExtendedStats(
 	compressionLevel int,
 	uploadBytes int64,
 	uploadDurationMs int64,
+	newChunkSize int64,
 ) error {
 	compressionRatio, err := r.extendedCompressionRatio(
 		ctx, tx, snapshotID, blobUncompressedSize,
@@ -143,18 +144,19 @@ func (r *SnapshotRepository) UpdateExtendedStats(
 		    compression_ratio = ?,
 		    compression_level = ?,
 		    upload_bytes = ?,
-		    upload_duration_ms = ?
+		    upload_duration_ms = ?,
+		    new_chunk_size = ?
 		WHERE id = ?
 	`
 
 	if tx != nil {
 		_, err = tx.ExecContext(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadBytes, uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	} else {
 		_, err = r.db.ExecWithLog(ctx, query,
 			blobUncompressedSize, compressionRatio, compressionLevel,
-			uploadBytes, uploadDurationMs, snapshotID)
+			uploadBytes, uploadDurationMs, newChunkSize, snapshotID)
 	}
 
 	if err != nil {
@@ -239,12 +241,7 @@ func (r *SnapshotRepository) GetOnlySnapshot(ctx context.Context) (*Snapshot, er
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	snapshots, err := r.scanSnapshotRows(rows)
 	if err != nil {
@@ -279,12 +276,7 @@ func (r *SnapshotRepository) ListRecent(
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -499,12 +491,7 @@ func (r *SnapshotRepository) GetBlobHashes(
 		return nil, fmt.Errorf("querying blob hashes: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	var blobs []string
 
@@ -544,6 +531,30 @@ func (r *SnapshotRepository) GetSnapshotTotalCompressedSize(
 	return totalSize, nil
 }
 
+// GetSnapshotBlobSizes returns the total compressed and uncompressed sizes
+// of all blobs referenced by a snapshot.
+func (r *SnapshotRepository) GetSnapshotBlobSizes(
+	ctx context.Context, snapshotID string,
+) (int64, int64, error) {
+	query := `
+		SELECT COALESCE(SUM(b.compressed_size), 0),
+			COALESCE(SUM(b.uncompressed_size), 0)
+		FROM snapshot_blobs sb
+		JOIN blobs b ON sb.blob_hash = b.blob_hash
+		WHERE sb.snapshot_id = ?
+	`
+
+	var compressed, uncompressed int64
+
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(
+		&compressed, &uncompressed)
+	if err != nil {
+		return 0, 0, fmt.Errorf("querying snapshot blob sizes: %w", err)
+	}
+
+	return compressed, uncompressed, nil
+}
+
 // GetSnapshotUncompressedChunkSize returns the sum of plaintext sizes of all unique
 // chunks referenced by a snapshot (via snapshot_files → file_chunks → chunks).
 func (r *SnapshotRepository) GetSnapshotUncompressedChunkSize(
@@ -570,45 +581,24 @@ func (r *SnapshotRepository) GetSnapshotUncompressedChunkSize(
 	return totalSize, nil
 }
 
-// GetSnapshotNewChunkSize returns the sum of plaintext sizes of chunks that are
+// GetSnapshotNewChunkSize returns the total plaintext size of the chunks the
-// referenced by this snapshot but not by any earlier completed snapshot known to
+// snapshot's backup stored that were not stored before, as the backup
-// the local database. The result is the marginal uncompressed data this snapshot
+// recorded it. It is not derived from the snapshot's files: a file's row is
-// added to the dedup pool — i.e., the delta from prior snapshots.
+// shared by every snapshot that lists it and is rewritten when the file
+// changes, so it shows only the file's latest chunks.
 func (r *SnapshotRepository) GetSnapshotNewChunkSize(
 	ctx context.Context, snapshotID string,
 ) (int64, error) {
-	query := `
+	query := `SELECT new_chunk_size FROM snapshots WHERE id = ?`
-		WITH this_snap_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshot_files sf
-			JOIN file_chunks fc ON sf.file_id = fc.file_id
-			WHERE sf.snapshot_id = ?
-		),
-		prior_chunks AS (
-			SELECT DISTINCT fc.chunk_hash
-			FROM snapshots s
-			JOIN snapshot_files sf ON sf.snapshot_id = s.id
-			JOIN file_chunks fc ON fc.file_id = sf.file_id
-			WHERE s.completed_at IS NOT NULL
-			  AND s.id != ?
-			  AND s.started_at < (SELECT started_at FROM snapshots WHERE id = ?)
-		)
-		SELECT COALESCE(SUM(c.size), 0)
-		FROM chunks c
-		JOIN this_snap_chunks t ON c.chunk_hash = t.chunk_hash
-		WHERE c.chunk_hash NOT IN (SELECT chunk_hash FROM prior_chunks)
-	`
 
-	var totalSize int64
+	var newChunkSize int64
 
-	err := r.db.conn.QueryRowContext(
+	err := r.db.conn.QueryRowContext(ctx, query, snapshotID).Scan(&newChunkSize)
-		ctx, query, snapshotID, snapshotID, snapshotID,
-	).Scan(&totalSize)
 	if err != nil {
 		return 0, fmt.Errorf("querying new chunk size: %w", err)
 	}
 
-	return totalSize, nil
+	return newChunkSize, nil
 }
 
 // GetIncompleteSnapshots returns all snapshots that haven't been completed
@@ -629,12 +619,7 @@ func (r *SnapshotRepository) GetIncompleteSnapshots(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -657,12 +642,7 @@ func (r *SnapshotRepository) GetIncompleteByHostname(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
 
 	// Same column set as every other multi-row snapshot query, so the
 	// shared scanner applies — including its timestamp normalization.

internal/database/snapshots_test.go

@@ -145,6 +145,65 @@ func TestSnapshotRepositoryUpdateCounts(t *testing.T) {
 	}
 }
 
+// GetSnapshotBlobSizes totals the blobs the snapshot references, and only
+// those.
+func TestSnapshotRepositoryGetSnapshotBlobSizes(t *testing.T) {
+	t.Parallel()
+
+	db, cleanup := setupTestDB(t)
+	defer cleanup()
+
+	ctx := context.Background()
+	repos := database.NewRepositories(db)
+
+	snapshot := &database.Snapshot{
+		ID:             "2024-01-03T12:00:00Z",
+		Hostname:       testHostname,
+		VaultikVersion: testVersion,
+		StartedAt:      time.Now().Truncate(time.Second),
+	}
+
+	err := repos.Snapshots.Create(ctx, nil, snapshot)
+	if err != nil {
+		t.Fatalf("failed to create snapshot: %v", err)
+	}
+
+	blobs := []*database.Blob{
+		{Hash: "referenced-1", CompressedSize: 10, UncompressedSize: 100},
+		{Hash: "referenced-2", CompressedSize: 20, UncompressedSize: 200},
+		{Hash: "unreferenced", CompressedSize: 40, UncompressedSize: 400},
+	}
+
+	for _, blob := range blobs {
+		blob.ID = types.NewBlobID()
+		blob.CreatedTS = time.Now().Truncate(time.Second)
+
+		err = repos.Blobs.Create(ctx, nil, blob)
+		if err != nil {
+			t.Fatalf("failed to create blob %s: %v", blob.Hash, err)
+		}
+	}
+
+	for _, blob := range blobs[:2] {
+		err = repos.Snapshots.AddBlob(ctx, nil, snapshot.ID.String(),
+			blob.ID, blob.Hash)
+		if err != nil {
+			t.Fatalf("failed to add blob %s to snapshot: %v", blob.Hash, err)
+		}
+	}
+
+	compressed, uncompressed, err := repos.Snapshots.GetSnapshotBlobSizes(
+		ctx, snapshot.ID.String())
+	if err != nil {
+		t.Fatalf("failed to get snapshot blob sizes: %v", err)
+	}
+
+	if compressed != 30 || uncompressed != 300 {
+		t.Errorf("blob sizes: got %d and %d, want 30 and 300",
+			compressed, uncompressed)
+	}
+}
+
 func TestSnapshotRepositoryListRecent(t *testing.T) {
 	t.Parallel()
 

internal/database/uploads.go

@@ -158,19 +158,3 @@ type UploadStats struct {
 	MinDurationMs int64
 	MaxDurationMs int64
 }
-
-// GetCountBySnapshot returns the count of uploads for a specific snapshot
-func (r *UploadRepository) GetCountBySnapshot(
-	ctx context.Context, snapshotID string,
-) (int64, error) {
-	query := `SELECT COUNT(*) FROM uploads WHERE snapshot_id = ?`
-
-	var count int64
-
-	err := r.conn.QueryRowContext(ctx, query, snapshotID).Scan(&count)
-	if err != nil {
-		return 0, err
-	}
-
-	return count, nil
-}

internal/log/tty_handler.go

@@ -10,6 +10,8 @@ import (
 	"sync"
 	"time"
 	"unicode"
+
+	"github.com/dustin/go-humanize"
 )
 
 // groupSeparator joins an open group path to an attribute key. This
@@ -304,18 +306,13 @@ func formatDuration(d time.Duration) string {
 	}
 }
 
-// formatBytes formats bytes in a human-readable way
+// formatBytes formats a byte count with humanize.Bytes, as every command
+// prints one. humanize.Bytes takes no negative count, so one is printed
+// as it is.
 func formatBytes(b int64) string {
-	const unit = 1024
+	if b < 0 {
-	if b < unit {
 		return fmt.Sprintf("%d B", b)
 	}
 
-	div, exp := int64(unit), 0
+	return humanize.Bytes(uint64(b))
-	for n := b / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(b)/float64(div), "KMGTPE"[exp])
 }

internal/log/tty_handler_test.go

@@ -200,7 +200,8 @@ func TestTTYHandlerWithGroupQualifiesKeys(t *testing.T) {
 func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 	t.Parallel()
 
-	const oneAndAHalfKiB = 1536
+	// Printed as "1.0 MB", in powers of 1000 like every other byte count.
+	const oneMegabyte = 1_000_000
 
 	for name, testCase := range map[string]struct {
 		derive func(*slog.Logger) *slog.Logger
@@ -223,13 +224,13 @@ func TestTTYHandlerByteFormattingSurvivesGrouping(t *testing.T) {
 			var buf bytes.Buffer
 
 			logger := slog.New(log.NewTTYHandler(&buf, debugHandlerOptions()))
-			testCase.derive(logger).Info("uploaded", "bytes", oneAndAHalfKiB)
+			testCase.derive(logger).Info("uploaded", "bytes", oneMegabyte)
 
 			line := ansiEscape.ReplaceAllString(buf.String(), "")
 
-			assert.Contains(t, line, testCase.key+"=1.5 KB",
+			assert.Contains(t, line, testCase.key+"=1.0 MB",
 				"a byte count must be human-readable however it is qualified")
-			assert.NotContains(t, line, strconv.Itoa(oneAndAHalfKiB),
+			assert.NotContains(t, line, strconv.Itoa(oneMegabyte),
 				"the raw number must not survive the formatting")
 		})
 	}

internal/pidlock/pidlock.go

@@ -10,15 +10,18 @@ import (
 	"path/filepath"
 	"strconv"
 	"strings"
-	"syscall"
+
+	"golang.org/x/sys/unix"
 )
 
 // ErrAlreadyRunning indicates another vaultik instance is running.
 var ErrAlreadyRunning = errors.New("another vaultik instance is already running")
 
-// Lock represents an acquired PID lock.
+// Lock represents an acquired PID lock: an flock(2) on the PID file,
+// held while the file stays open. The kernel drops it when the process
+// exits, however it exits, so a crashed run never leaves the lock held.
 type Lock struct {
-	path string
+	file *os.File
 }
 
 const (
@@ -29,10 +32,9 @@ const (
 )
 
 // Acquire attempts to acquire a PID lock in the specified directory.
-// If the lock file exists and the process is still running, it returns
+// If another process holds the lock, it returns ErrAlreadyRunning with
-// ErrAlreadyRunning with details about the existing process.
+// that process's PID. On success, it writes the current PID to the lock
-// On success, it writes the current PID to the lock file and returns
+// file and returns a Lock that must be released with Release().
-// a Lock that must be released with Release().
 func Acquire(lockDir string) (*Lock, error) {
 	// Ensure lock directory exists
 	err := os.MkdirAll(lockDir, lockDirPerm)
@@ -42,56 +44,82 @@ func Acquire(lockDir string) (*Lock, error) {
 
 	lockPath := filepath.Join(lockDir, "vaultik.pid")
 
-	// Check for existing lock
+	// No O_TRUNC: the file may hold the PID of the process that has the
-	existingPID, err := readPIDFile(lockPath)
+	// lock, which the error below reports.
-	if err == nil {
+	file, err := os.OpenFile( //nolint:gosec // G304: path is our own lock file
-		// Lock file exists, check if process is running
+		lockPath, os.O_RDWR|os.O_CREATE, pidFilePerm)
-		if isProcessRunning(existingPID) {
-			return nil, fmt.Errorf("%w (PID %d)", ErrAlreadyRunning, existingPID)
-		}
-		// Process is not running, stale lock file - we can take over
-	}
-
-	// Write our PID
-	pid := os.Getpid()
-
-	err = os.WriteFile(lockPath, []byte(strconv.Itoa(pid)), pidFilePerm)
 	if err != nil {
-		return nil, fmt.Errorf("writing PID file: %w", err)
+		return nil, fmt.Errorf("opening PID file: %w", err)
 	}
 
-	return &Lock{path: lockPath}, nil
+	err = unix.Flock(int(file.Fd()), unix.LOCK_EX|unix.LOCK_NB)
+	if err != nil {
+		_ = file.Close()
+
+		if errors.Is(err, unix.EWOULDBLOCK) {
+			return nil, alreadyRunningError(lockPath)
+		}
+
+		return nil, fmt.Errorf("locking PID file: %w", err)
+	}
+
+	err = writePID(file)
+	if err != nil {
+		_ = file.Close()
+
+		return nil, err
+	}
+
+	return &Lock{file: file}, nil
 }
 
-// Release removes the PID lock file.
+// Release empties the PID file and closes it, which drops the lock.
 // It is safe to call Release multiple times.
 func (l *Lock) Release() error {
-	if l == nil || l.path == "" {
+	if l == nil || l.file == nil {
 		return nil
 	}
 
-	// Verify we still own the lock (our PID is in the file)
+	file := l.file
-	existingPID, err := readPIDFile(l.path)
+	l.file = nil
+
+	// Do not remove the file here. A process that opened it a moment
+	// earlier could then lock the removed file while another creates and
+	// locks a new one, and both would run.
+	truncateErr := file.Truncate(0)
+	closeErr := file.Close()
+
+	return errors.Join(truncateErr, closeErr)
+}
+
+// writePID replaces the contents of the locked PID file with the current
+// PID.
+func writePID(file *os.File) error {
+	err := file.Truncate(0)
 	if err != nil {
-		// File already gone or unreadable - that's fine
+		return fmt.Errorf("truncating PID file: %w", err)
-		return nil //nolint:nilerr // unreadable lock file means nothing to release
 	}
 
-	if existingPID != os.Getpid() {
+	_, err = file.WriteAt([]byte(strconv.Itoa(os.Getpid())), 0)
-		// Someone else wrote to our lock file - don't remove it
+	if err != nil {
-		return nil
+		return fmt.Errorf("writing PID file: %w", err)
 	}
 
-	err = os.Remove(l.path)
-	if err != nil && !os.IsNotExist(err) {
-		return fmt.Errorf("removing PID file: %w", err)
-	}
-
-	l.path = "" // Prevent double-release
-
 	return nil
 }
 
+// alreadyRunningError reports that another process holds the lock,
+// naming its PID when the file holds one. The holder writes its PID just
+// after it locks, so the file can briefly be empty.
+func alreadyRunningError(lockPath string) error {
+	pid, err := readPIDFile(lockPath)
+	if err != nil {
+		return ErrAlreadyRunning
+	}
+
+	return fmt.Errorf("%w (PID %d)", ErrAlreadyRunning, pid)
+}
+
 // readPIDFile reads and parses the PID from a lock file.
 func readPIDFile(path string) (int, error) {
 	data, err := os.ReadFile(path) //nolint:gosec // G304: path is our own lock file
@@ -106,16 +134,3 @@ func readPIDFile(path string) (int, error) {
 
 	return pid, nil
 }
-
-// isProcessRunning checks if a process with the given PID is running.
-func isProcessRunning(pid int) bool {
-	process, err := os.FindProcess(pid)
-	if err != nil {
-		return false
-	}
-
-	// On Unix, FindProcess always succeeds. We need to send signal 0 to check.
-	err = process.Signal(syscall.Signal(0))
-
-	return err == nil
-}

internal/pidlock/pidlock_test.go

@@ -4,6 +4,7 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
+	"sync"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -33,9 +34,10 @@ func TestAcquireAndRelease(t *testing.T) {
 	err = lock.Release()
 	require.NoError(t, err)
 
-	// Verify PID file is gone
+	// Verify PID file is empty
-	_, err = os.Stat(pidPath)
+	data, err = os.ReadFile(pidPath) //nolint:gosec // G304: test's own temp file
-	assert.True(t, os.IsNotExist(err))
+	require.NoError(t, err)
+	assert.Empty(t, data)
 }
 
 func TestAcquireBlocksSecondInstance(t *testing.T) {
@@ -55,6 +57,64 @@ func TestAcquireBlocksSecondInstance(t *testing.T) {
 	lock2, err := pidlock.Acquire(tmpDir)
 	require.ErrorIs(t, err, pidlock.ErrAlreadyRunning)
 	assert.Nil(t, lock2)
+
+	// Once the first lock is released, the next Acquire succeeds
+	require.NoError(t, lock1.Release())
+
+	lock3, err := pidlock.Acquire(tmpDir)
+	require.NoError(t, err)
+	require.NoError(t, lock3.Release())
+}
+
+// TestConcurrentAcquireAdmitsOne starts many Acquire calls at the same
+// moment, as two cron entries firing together would, and checks that
+// exactly one of them gets the lock.
+func TestConcurrentAcquireAdmitsOne(t *testing.T) {
+	t.Parallel()
+
+	const callers = 50
+
+	tmpDir := t.TempDir()
+	start := make(chan struct{})
+
+	var (
+		mu       sync.Mutex
+		acquired []*pidlock.Lock
+		failures []error
+		wg       sync.WaitGroup
+	)
+
+	for range callers {
+		wg.Go(func() {
+			<-start
+
+			lock, err := pidlock.Acquire(tmpDir)
+
+			mu.Lock()
+			defer mu.Unlock()
+
+			if err != nil {
+				failures = append(failures, err)
+
+				return
+			}
+
+			acquired = append(acquired, lock)
+		})
+	}
+
+	close(start)
+	wg.Wait()
+
+	for _, lock := range acquired {
+		require.NoError(t, lock.Release())
+	}
+
+	assert.Len(t, acquired, 1, "exactly one caller should hold the lock")
+
+	for _, err := range failures {
+		require.ErrorIs(t, err, pidlock.ErrAlreadyRunning)
+	}
 }
 
 func TestAcquireWithStaleLock(t *testing.T) {

internal/s3/client.go

@@ -18,6 +18,13 @@ import (
 	"github.com/aws/smithy-go/logging"
 )
 
+// ErrNoContinuationToken is returned by ListObjects and ListObjectsStream
+// when the server marks a page of a listing truncated but sends no
+// continuation token to fetch the rest with, as a server answering V2
+// listings V1-style does.
+var ErrNoContinuationToken = errors.New(
+	"server marked the listing truncated but sent no continuation token")
+
 // Client wraps the AWS S3 client for vaultik operations.
 // It provides a simplified interface for S3 operations with automatic
 // prefix handling and connection management. All operations are performed
@@ -27,10 +34,12 @@ type Client struct {
 	bucket   string
 	prefix   string
 	endpoint string
+	partSize int64
 }
 
 // Config contains S3 client configuration.
-// All fields are required except Prefix, which defaults to an empty string.
+// All fields are required except Prefix, which defaults to an empty string,
+// and PartSize, where zero means the SDK default of 5 MiB.
 // A non-empty Prefix is joined to every key with one "/", whether or not
 // it ends with one.
 // The Endpoint field should include the protocol (http:// or https://).
@@ -41,6 +50,9 @@ type Config struct {
 	AccessKeyID     string
 	SecretAccessKey string
 	Region          string
+	// PartSize is the size in bytes of each part of a multipart upload.
+	// An upload too large for S3's limit of 10,000 parts gets larger parts.
+	PartSize int64
 }
 
 // nopLogger is a logger that discards all output.
@@ -90,6 +102,7 @@ func NewClient(ctx context.Context, cfg Config) (*Client, error) {
 		bucket:   cfg.Bucket,
 		prefix:   prefix,
 		endpoint: cfg.Endpoint,
+		partSize: cfg.PartSize,
 	}, nil
 }
 
@@ -123,12 +136,9 @@ func (c *Client) PutObjectWithProgress(
 ) error {
 	fullKey := c.prefix + key
 
-	// uploadPartSize is 10MB for better progress granularity.
-	const uploadPartSize = 10 * 1024 * 1024
-
 	// Create an uploader with the S3 client
 	uploader := manager.NewUploader(c.s3Client, func(u *manager.Uploader) {
-		u.PartSize = uploadPartSize
+		u.PartSize = uploadPartSize(c.partSize, size)
 	})
 
 	// Create a progress reader that tracks upload progress
@@ -149,6 +159,21 @@ func (c *Client) PutObjectWithProgress(
 	return err
 }
 
+// uploadPartSize returns the part size for an upload of size bytes: the
+// configured part size (the SDK default when zero), raised where needed so
+// the upload fits in S3's limit of 10,000 parts. The uploader cannot raise
+// it itself, because it cannot seek the progress reader to learn its size.
+func uploadPartSize(configured, size int64) int64 {
+	if configured == 0 {
+		configured = manager.DefaultUploadPartSize
+	}
+
+	maxParts := int64(manager.MaxUploadParts)
+	smallestThatFits := (size + maxParts - 1) / maxParts // rounded up
+
+	return max(configured, smallestThatFits)
+}
+
 // GetObject downloads an object from S3 with the specified key.
 // The key is automatically prefixed with the configured prefix.
 // Returns a ReadCloser containing the object data. The caller must
@@ -196,7 +221,7 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	})
 
 	for paginator.HasMorePages() {
-		page, err := paginator.NextPage(ctx)
+		page, err := nextListPage(ctx, paginator)
 		if err != nil {
 			return nil, err
 		}
@@ -217,6 +242,26 @@ func (c *Client) ListObjects(ctx context.Context, prefix string) ([]string, erro
 	return keys, nil
 }
 
+// nextListPage fetches the next page of a listing. The SDK paginator takes
+// a page marked truncated without a continuation token for the last page
+// and stops without an error, so such a page is ErrNoContinuationToken
+// here instead.
+func nextListPage(
+	ctx context.Context, paginator *s3.ListObjectsV2Paginator,
+) (*s3.ListObjectsV2Output, error) {
+	page, err := paginator.NextPage(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	truncated := aws.ToBool(page.IsTruncated)
+	if truncated && aws.ToString(page.NextContinuationToken) == "" {
+		return nil, ErrNoContinuationToken
+	}
+
+	return page, nil
+}
+
 // HeadObject checks if an object exists in S3.
 // Returns true if the object exists, false otherwise.
 // The key is automatically prefixed with the configured prefix.
@@ -282,7 +327,7 @@ func (c *Client) ListObjectsStream(
 		})
 
 		for paginator.HasMorePages() {
-			page, err := paginator.NextPage(ctx)
+			page, err := nextListPage(ctx, paginator)
 			if err != nil {
 				ch <- ObjectInfo{Err: err}
 

internal/s3/client_internal_test.go

@@ -0,0 +1,55 @@
+package s3
+
+import "testing"
+
+// TestUploadPartSize checks that an upload too large for 10,000 parts of the
+// configured size gets parts just large enough to fit in 10,000.
+func TestUploadPartSize(t *testing.T) {
+	t.Parallel()
+
+	const mib = 1024 * 1024
+
+	tests := []struct {
+		name       string
+		configured int64
+		size       int64
+		want       int64
+	}{
+		{
+			name:       "an upload that fits keeps the configured size",
+			configured: 5 * mib,
+			size:       10 * 1024 * mib,
+			want:       5 * mib,
+		},
+		{
+			name:       "exactly 10,000 parts keeps the configured size",
+			configured: 6 * mib,
+			size:       10_000 * 6 * mib,
+			want:       6 * mib,
+		},
+		{
+			name:       "one byte more than 10,000 parts adds a byte to each",
+			configured: 6 * mib,
+			size:       10_000*6*mib + 1,
+			want:       6*mib + 1,
+		},
+		{
+			name:       "zero means the SDK default of 5MiB",
+			configured: 0,
+			size:       1,
+			want:       5 * mib,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			got := uploadPartSize(tt.configured, tt.size)
+			if got != tt.want {
+				t.Errorf("uploadPartSize(%d, %d) = %d, want %d",
+					tt.configured, tt.size, got, tt.want)
+			}
+		})
+	}
+}

internal/s3/module.go

@@ -28,6 +28,7 @@ func provideClient(lc fx.Lifecycle, cfg *config.Config) (*Client, error) {
 		AccessKeyID:     cfg.S3.AccessKeyID,
 		SecretAccessKey: cfg.S3.SecretAccessKey,
 		Region:          cfg.S3.Region,
+		PartSize:        cfg.S3.PartSize.Int64(),
 	})
 	if err != nil {
 		return nil, err

internal/snapshot/blob_temp_file_test.go

@@ -0,0 +1,62 @@
+//nolint:testpackage // exercises the unexported cleanupBlobTempFile
+package snapshot
+
+import (
+	"errors"
+	"os"
+	"testing"
+
+	"github.com/spf13/afero"
+	"sneak.berlin/go/vaultik/internal/blob"
+)
+
+// A blob temp file that cannot be closed or removed must make the cleanup
+// return an error, not end the process.
+func TestCleanupBlobTempFileReturnsErrors(t *testing.T) {
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	s := &Scanner{fs: fs}
+
+	tests := []struct {
+		name string
+		// prepare runs before the cleanup and makes its close or remove
+		// fail.
+		prepare func(tempFile afero.File) error
+		want    error
+	}{
+		{
+			name:    "close fails",
+			prepare: func(tempFile afero.File) error { return tempFile.Close() },
+			want:    os.ErrClosed,
+		},
+		{
+			name: "remove fails",
+			prepare: func(tempFile afero.File) error {
+				return fs.Remove(tempFile.Name())
+			},
+			want: os.ErrNotExist,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			tempFile, err := afero.TempFile(fs, t.TempDir(), "blob-*.tmp")
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = tt.prepare(tempFile)
+			if err != nil {
+				t.Fatal(err)
+			}
+
+			err = s.cleanupBlobTempFile(&blob.WithReader{TempFile: tempFile})
+			if !errors.Is(err, tt.want) {
+				t.Fatalf("expected an error wrapping %v, got: %v", tt.want, err)
+			}
+		})
+	}
+}

internal/snapshot/exclude_test.go

@@ -531,3 +531,50 @@ func TestExcludePatterns_UnanchoredPatternFile(t *testing.T) {
 	require.Equal(t, 1, result.FilesScanned,
 		"Unanchored pattern for file should match anywhere")
 }
+
+func TestExcludePatterns_DoubleStarDirectoryAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/projectname"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/projectname should exclude BOTH:
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 2
+	require.Equal(t, 2, result.FilesScanned,
+		"**/ directory pattern should match at the top level and nested")
+}
+
+func TestExcludePatterns_DoubleStarFileAtEveryDepth(t *testing.T) {
+	t.Parallel()
+
+	// As in gitignore, a leading **/ also matches directly under the source dir
+	fs := setupAnchoredTestFS(t)
+
+	scanner, repos, cleanup := createTestScanner(t, fs, []string{"**/*.txt"})
+	defer cleanup()
+
+	ctx := context.Background()
+	createSnapshotRecord(ctx, t, repos, "test-snapshot")
+
+	result, err := scanner.Scan(ctx, "/backup", "test-snapshot")
+	require.NoError(t, err)
+
+	// **/*.txt should exclude ALL .txt files:
+	// - /backup/file.txt
+	// - /backup/projectname/file.txt
+	// - /backup/otherproject/projectname/file.txt
+	// Total files: 4, excluded: 3
+	require.Equal(t, 1, result.FilesScanned,
+		"**/ file pattern should match at the top level and nested")
+}

internal/snapshot/progress.go

@@ -56,23 +56,27 @@ const (
 
 // ProgressStats holds atomic counters for progress tracking
 type ProgressStats struct {
-	FilesScanned     atomic.Int64 // Total files seen during scan (includes skipped)
+	FilesScanned   atomic.Int64 // Total files seen during scan (includes skipped)
-	FilesProcessed   atomic.Int64 // Files actually processed in phase 2
+	FilesProcessed atomic.Int64 // Files actually processed in phase 2
-	FilesSkipped     atomic.Int64 // Files skipped due to no changes
+	FilesSkipped   atomic.Int64 // Files skipped due to no changes
-	BytesScanned     atomic.Int64 // Bytes from new/changed files only
+	BytesScanned   atomic.Int64 // Bytes from new/changed files only
-	BytesSkipped     atomic.Int64 // Bytes from unchanged files
+	BytesSkipped   atomic.Int64 // Bytes from unchanged files
-	BytesProcessed   atomic.Int64 // Actual bytes processed (for ETA calculation)
+	BytesProcessed atomic.Int64 // Actual bytes processed (for ETA calculation)
-	ChunksCreated    atomic.Int64
+	ChunksCreated  atomic.Int64
-	BlobsCreated     atomic.Int64
+	BlobsCreated   atomic.Int64
-	BlobsUploaded    atomic.Int64
+	BlobsUploaded  atomic.Int64
-	BytesUploaded    atomic.Int64
+	BytesUploaded  atomic.Int64
-	CurrentFile      atomic.Value // stores string
+	CurrentFile    atomic.Value // stores string
-	TotalSize        atomic.Int64 // Total size to process (set after scan phase)
+	TotalSize      atomic.Int64 // Size to process in the paths scanned so far
-	TotalFiles       atomic.Int64 // Total files to process in phase 2
+	TotalFiles     atomic.Int64 // Files to process in the paths scanned so far
-	ProcessStartTime atomic.Value // stores time.Time when processing starts
+	StartTime      time.Time
-	StartTime        time.Time
+	mu             sync.RWMutex
-	mu               sync.RWMutex
+	lastDetailTime time.Time
-	lastDetailTime   time.Time
+
+	// Guarded by mu: when the last scan phase ended, and
+	// BytesProcessed at that moment.
+	processStartTime  time.Time
+	processStartBytes int64
 
 	// Upload tracking
 	CurrentUpload    atomic.Value // stores *UploadInfo
@@ -148,10 +152,36 @@ func (pr *ProgressReporter) GetStats() *ProgressStats {
 	return pr.stats
 }
 
-// SetTotalSize sets the total size to process (after scan phase)
+// AddTotalSize adds the size one path of the snapshot has to process to
-func (pr *ProgressReporter) SetTotalSize(size int64) {
+// the total, once that path's scan phase is done, and starts measuring
-	pr.stats.TotalSize.Store(size)
+// the processing rate again. The processed counts run across every path,
-	pr.stats.ProcessStartTime.Store(time.Now().UTC())
+// so the total does too. Restarting the rate keeps a path's scan phase,
+// which processes nothing, out of the rate while that path is processed.
+// During a later path's scan phase the rate is still the previous
+// path's, and falls as that scan goes on.
+func (pr *ProgressReporter) AddTotalSize(size int64) {
+	pr.stats.TotalSize.Add(size)
+
+	pr.stats.mu.Lock()
+	defer pr.stats.mu.Unlock()
+
+	pr.stats.processStartTime = time.Now().UTC()
+	pr.stats.processStartBytes = pr.stats.BytesProcessed.Load()
+}
+
+// processRate returns the bytes processed per second since the last scan
+// phase ended, or 0 before the first path's scan phase is done.
+func (s *ProgressStats) processRate() float64 {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	if s.processStartTime.IsZero() {
+		return 0
+	}
+
+	processed := s.BytesProcessed.Load() - s.processStartBytes
+
+	return float64(processed) / time.Since(s.processStartTime).Seconds()
 }
 
 // Helper functions
@@ -357,19 +387,12 @@ func (pr *ProgressReporter) printSummaryStatus() {
 	// Calculate ETA if we have total size and are processing
 	etaStr := ""
 
-	if totalSize > 0 && bytesProcessed > 0 {
+	processRate := pr.stats.processRate()
-		processStart, ok := pr.stats.ProcessStartTime.Load().(time.Time)
+	if totalSize > 0 && processRate > 0 {
-		if ok && !processStart.IsZero() {
+		remainingBytes := totalSize - bytesProcessed
-			processElapsed := time.Since(processStart)
+		remainingSeconds := float64(remainingBytes) / processRate
-
+		eta := time.Duration(remainingSeconds * float64(time.Second))
-			rate := float64(bytesProcessed) / processElapsed.Seconds()
+		etaStr = " | ETA: " + formatDuration(eta)
-			if rate > 0 {
-				remainingBytes := totalSize - bytesProcessed
-				remainingSeconds := float64(remainingBytes) / rate
-				eta := time.Duration(remainingSeconds * float64(time.Second))
-				etaStr = " | ETA: " + formatDuration(eta)
-			}
-		}
 	}
 
 	rate := float64(bytesScanned+bytesSkipped) / elapsed.Seconds()
@@ -421,25 +444,18 @@ func (pr *ProgressReporter) printDetailedStatus() {
 	log.Info("Elapsed time", "duration", formatDuration(elapsed))
 
 	// Calculate and show ETA if we have data
-	if totalSize > 0 && bytesProcessed > 0 {
+	processRate := pr.stats.processRate()
-		processStart, ok := pr.stats.ProcessStartTime.Load().(time.Time)
+	if totalSize > 0 && processRate > 0 {
-		if ok && !processStart.IsZero() {
+		remainingBytes := totalSize - bytesProcessed
-			processElapsed := time.Since(processStart)
+		remainingSeconds := float64(remainingBytes) / processRate
-
+		eta := time.Duration(remainingSeconds * float64(time.Second))
-			processRate := float64(bytesProcessed) / processElapsed.Seconds()
+		percentComplete := float64(bytesProcessed) / float64(totalSize) * percentScale
-			if processRate > 0 {
+		log.Info("Overall progress",
-				remainingBytes := totalSize - bytesProcessed
+			"percent", fmt.Sprintf("%.1f%%", percentComplete),
-				remainingSeconds := float64(remainingBytes) / processRate
+			"processed", humanize.Bytes(safeUint64(bytesProcessed)),
-				eta := time.Duration(remainingSeconds * float64(time.Second))
+			"total", humanize.Bytes(safeUint64(totalSize)),
-				percentComplete := float64(bytesProcessed) / float64(totalSize) * percentScale
+			"rate", humanize.Bytes(uint64(processRate))+"/s",
-				log.Info("Overall progress",
+			"eta", formatDuration(eta))
-					"percent", fmt.Sprintf("%.1f%%", percentComplete),
-					"processed", humanize.Bytes(safeUint64(bytesProcessed)),
-					"total", humanize.Bytes(safeUint64(totalSize)),
-					"rate", humanize.Bytes(uint64(processRate))+"/s",
-					"eta", formatDuration(eta))
-			}
-		}
 	}
 
 	log.Info("Files processed",

internal/snapshot/progress_rate_test.go

@@ -0,0 +1,54 @@
+//nolint:testpackage // exercises the unexported processRate helper
+package snapshot
+
+import (
+	"testing"
+	"time"
+)
+
+// TestProcessRateNotLoweredByLaterScanPhase feeds the reporter two paths
+// that each take 10 seconds to process, the second after a 30-second scan
+// phase. That scan phase processes nothing, so the rate while the second
+// path is processed must be that path's own.
+func TestProcessRateNotLoweredByLaterScanPhase(t *testing.T) {
+	t.Parallel()
+
+	const (
+		pathSize       = 1000
+		processingTime = 10 * time.Second
+		scanTime       = 30 * time.Second
+	)
+
+	// Never started, but Stop releases its tickers and signal handler.
+	progress := NewProgressReporter()
+	defer progress.Stop()
+
+	stats := progress.GetStats()
+
+	// Moves the processing start time back instead of sleeping.
+	elapse := func(d time.Duration) {
+		stats.mu.Lock()
+		defer stats.mu.Unlock()
+
+		stats.processStartTime = stats.processStartTime.Add(-d)
+	}
+
+	progress.AddTotalSize(pathSize)
+	stats.BytesProcessed.Add(pathSize)
+	elapse(processingTime)
+
+	elapse(scanTime)
+	progress.AddTotalSize(pathSize)
+	stats.BytesProcessed.Add(pathSize)
+	elapse(processingTime)
+
+	want := pathSize / processingTime.Seconds()
+	got := stats.processRate()
+
+	// The test's own run time adds to the elapsed time, so got is a hair
+	// under want.
+	if got > want || got < want*0.99 {
+		t.Errorf("rate while the second path is processed is %.1f bytes/s, "+
+			"want %.1f", got, want)
+	}
+}

internal/snapshot/progress_test.go

@@ -0,0 +1,94 @@
+package snapshot_test
+
+import (
+	"context"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/spf13/afero"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// TestProgressPercentWithSecondPathSmaller backs up two paths with one
+// scanner, as a snapshot with two paths does, the second path smaller
+// than the first. The progress line divides the bytes processed by the
+// total size, so both must count both paths to stay within 100%.
+func TestProgressPercentWithSecondPathSmaller(t *testing.T) {
+	t.Parallel()
+
+	fs := afero.NewMemMapFs()
+	files := map[string]string{
+		"/large/one.txt":   strings.Repeat("1", 4000),
+		"/large/two.txt":   strings.Repeat("2", 4000),
+		"/small/three.txt": strings.Repeat("3", 1000),
+	}
+
+	for path, content := range files {
+		err := fs.MkdirAll(filepath.Dir(path), 0755)
+		if err != nil {
+			t.Fatalf("mkdir: %v", err)
+		}
+
+		err = afero.WriteFile(fs, path, []byte(content), 0644)
+		if err != nil {
+			t.Fatalf("write %s: %v", path, err)
+		}
+	}
+
+	db, err := database.NewTestDB()
+	if err != nil {
+		t.Fatalf("create test db: %v", err)
+	}
+
+	t.Cleanup(func() {
+		cerr := db.Close()
+		if cerr != nil {
+			t.Errorf("close db: %v", cerr)
+		}
+	})
+
+	repos := database.NewRepositories(db)
+
+	scanner := snapshot.NewScanner(snapshot.ScannerConfig{
+		FS:               fs,
+		ChunkSize:        int64(1024 * 16),
+		Repositories:     repos,
+		MaxBlobSize:      int64(1024 * 1024),
+		CompressionLevel: 3,
+		AgeRecipients:    []string{testAgePublicKey},
+		EnableProgress:   true,
+	})
+
+	// Never started, but Stop releases its tickers and signal handler.
+	progress := scanner.GetProgress()
+	defer progress.Stop()
+
+	ctx := context.Background()
+	snapshotID := "test-snapshot-progress"
+	createTestSnapshotRecord(ctx, t, repos, snapshotID)
+
+	for _, path := range []string{"/large", "/small"} {
+		_, err := scanner.Scan(ctx, path, snapshotID)
+		if err != nil {
+			t.Fatalf("scanning %s: %v", path, err)
+		}
+	}
+
+	stats := progress.GetStats()
+
+	// Directories count toward the total size but produce no chunks, so
+	// the percentage ends just under 100%.
+	percent := float64(stats.BytesProcessed.Load()) /
+		float64(stats.TotalSize.Load()) * 100
+	if percent > 100 {
+		t.Errorf("progress after both paths is %.1f%%, want at most 100%%",
+			percent)
+	}
+
+	if stats.FilesProcessed.Load() != stats.TotalFiles.Load() {
+		t.Errorf("progress after both paths is %d of %d files, want all",
+			stats.FilesProcessed.Load(), stats.TotalFiles.Load())
+	}
+}

internal/snapshot/scanner.go

@@ -58,8 +58,8 @@ type Scanner struct {
 	compressionLevel int
 	ageRecipient     string
 	snapshotID       string // Current snapshot being processed
-	// currentSourcePath is the source directory being scanned (used for
+	// currentSourcePath is the source directory being scanned, stored with
-	// restore path stripping).
+	// each file record.
 	currentSourcePath string
 	exclude           []string          // Glob patterns for files/directories to exclude
 	compiledExclude   []compiledPattern // Compiled glob patterns
@@ -133,15 +133,21 @@ type ScannerConfig struct {
 
 // ScanResult contains the results of a scan operation. Files and bytes
 // are counted per file: BytesScanned is the size of the new and changed
-// files, BytesSkipped that of the unchanged ones.
+// files, BytesSkipped that of the unchanged ones. FilesFailed counts the
+// new and changed files that phase 2 could not store; FilesScanned
+// includes them and BytesScanned does not. ChunksCreated counts the
+// chunks that were not already stored, and NewChunkSize is their total
+// size.
 type ScanResult struct {
 	FilesScanned   int
 	FilesSkipped   int
+	FilesFailed    int
 	FilesDeleted   int
 	BytesScanned   int64
 	BytesSkipped   int64
 	BytesDeleted   int64
 	ChunksCreated  int
+	NewChunkSize   int64
 	BlobsCreated   int
 	BlobsUploaded  int
 	BytesUploaded  int64
@@ -211,7 +217,6 @@ func (s *Scanner) Scan(
 	ctx context.Context, path string, snapshotID string,
 ) (*ScanResult, error) {
 	s.snapshotID = snapshotID
-	// Store source path for file records (used during restore)
 	s.currentSourcePath = path
 	result := &ScanResult{
 		StartTime: time.Now().UTC(),
@@ -227,12 +232,6 @@ func (s *Scanner) Scan(
 		log.Debug("No storage configured, blobs will not be uploaded")
 	}
 
-	// Start progress reporting if enabled
-	if s.progress != nil {
-		s.progress.Start()
-		defer s.progress.Stop()
-	}
-
 	// Phase 0: Repair any state left by an interrupted previous run, then
 	// load known files and chunks from the database into memory for fast
 	// lookup.
@@ -302,6 +301,13 @@ func (s *Scanner) Scan(
 	return result, nil
 }
 
+// GetProgress returns the progress reporter for this scanner, or nil when
+// progress is off. Scan neither starts nor stops it: the caller does,
+// once for all the paths it scans, because a second Stop panics.
+func (s *Scanner) GetProgress() *ProgressReporter {
+	return s.progress
+}
+
 // loadDatabaseState loads known files and chunks from the database into
 // memory for fast lookup. This avoids per-file and per-chunk database
 // queries during the scan and process phases.
@@ -407,8 +413,8 @@ func (s *Scanner) summarizeScanPhase(
 	}
 
 	if s.progress != nil {
-		s.progress.SetTotalSize(totalSizeToProcess)
+		s.progress.AddTotalSize(totalSizeToProcess)
-		s.progress.GetStats().TotalFiles.Store(int64(len(filesToProcess)))
+		s.progress.GetStats().TotalFiles.Add(int64(len(filesToProcess)))
 	}
 
 	log.Info("Phase 1 complete",
@@ -889,9 +895,10 @@ func (s *Scanner) scanPhase(
 		}
 
 		// Handle symlinks and directories
-		if handled := s.recordSpecialEntry(
+		handled, err := s.recordSpecialEntry(
-			filePath, info, existingFiles, collector, result); handled {
+			filePath, info, existingFiles, collector, result)
-			return nil
+		if handled {
+			return err
 		}
 
 		// Skip other non-regular files (devices, sockets, etc.)
@@ -933,22 +940,25 @@ func (s *Scanner) scanPhase(
 }
 
 // recordSpecialEntry records symlinks and directories (which have no
-// data to chunk) and reports whether it handled the entry.
+// data to chunk) and reports whether it handled the entry. For a symlink
+// whose target cannot be read it returns handleWalkError's result.
 func (s *Scanner) recordSpecialEntry(
 	filePath string, info os.FileInfo,
 	existingFiles map[string]struct{},
 	collector *scanCollector, result *ScanResult,
-) bool {
+) (bool, error) {
 	// Handle symlinks
 	if info.Mode()&os.ModeSymlink != 0 {
-		file := s.buildSymlinkEntry(filePath, info)
+		file, err := s.buildSymlinkEntry(filePath, info)
-		if file != nil {
+		if err != nil {
-			existingFiles[filePath] = struct{}{}
+			return true, s.handleWalkError(filePath, err)
-			collector.addToProcess(filePath, info, file)
-			s.updateScanEntryStats(result, true, info)
 		}
 
-		return true
+		existingFiles[filePath] = struct{}{}
+		collector.addToProcess(filePath, info, file)
+		s.updateScanEntryStats(result, true, info)
+
+		return true, nil
 	}
 
 	// Handle directories (record for permission/ownership preservation
@@ -958,10 +968,10 @@ func (s *Scanner) recordSpecialEntry(
 		existingFiles[filePath] = struct{}{}
 		collector.addToProcess(filePath, info, file)
 
-		return true
+		return true, nil
 	}
 
-	return false
+	return false, nil
 }
 
 // handleWalkError deals with a filesystem error surfaced by the walk:
@@ -1111,13 +1121,12 @@ func (s *Scanner) printScanProgressLine(
 }
 
 // buildSymlinkEntry creates a File record for a symlink.
-// Returns nil if the link target cannot be read.
+func (s *Scanner) buildSymlinkEntry(
-func (s *Scanner) buildSymlinkEntry(path string, info os.FileInfo) *database.File {
+	path string, info os.FileInfo,
+) (*database.File, error) {
 	target, err := os.Readlink(path)
 	if err != nil {
-		log.Debug("Cannot read symlink target", "path", path, "error", err)
+		return nil, err
-
-		return nil
 	}
 
 	var uid, gid uint32
@@ -1136,7 +1145,7 @@ func (s *Scanner) buildSymlinkEntry(path string, info os.FileInfo) *database.Fil
 		UID:        uid,
 		GID:        gid,
 		LinkTarget: types.FilePath(target),
-	}
+	}, nil
 }
 
 // buildDirectoryEntry creates a File record for a directory.
@@ -1197,9 +1206,8 @@ func (s *Scanner) checkFileInMemory(
 	}
 
 	file := &database.File{
-		ID:   fileID,
+		ID:         fileID,
-		Path: types.FilePath(path),
+		Path:       types.FilePath(path),
-		// Store source directory for restore path stripping
 		SourcePath: types.SourcePath(s.currentSourcePath),
 		MTime:      info.ModTime(),
 		Size:       info.Size(),
@@ -1220,7 +1228,7 @@ func (s *Scanner) checkFileInMemory(
 
 	// Check if file has changed
 	if existingFile.Size != file.Size ||
-		existingFile.MTime.Unix() != file.MTime.Unix() ||
+		!existingFile.MTime.Equal(file.MTime) ||
 		existingFile.Mode != file.Mode ||
 		existingFile.UID != file.UID ||
 		existingFile.GID != file.GID {
@@ -1290,6 +1298,11 @@ func (s *Scanner) batchAddFilesToSnapshot(
 func (s *Scanner) processPhase(
 	ctx context.Context, filesToProcess []*FileToProcess, result *ScanResult,
 ) error {
+	// On an error or cancellation the blob being packed is never
+	// finalized, so its temp file must be removed here. After a
+	// successful flush there is no such blob and this does nothing.
+	defer s.packer.DiscardCurrentBlob()
+
 	// Calculate total bytes to process
 	var totalBytes int64
 	for _, f := range filesToProcess {
@@ -1307,6 +1320,13 @@ func (s *Scanner) processPhase(
 
 	// Process each file
 	for _, fileToProcess := range filesToProcess {
+		// Check context cancellation
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+
 		// Update progress
 		if s.progress != nil {
 			s.progress.GetStats().CurrentFile.Store(fileToProcess.Path)
@@ -1343,13 +1363,32 @@ func (s *Scanner) processPhase(
 	return s.finalizeProcessPhase(ctx, result)
 }
 
-// processFileWithErrorHandling wraps processFileStreaming with error recovery for
+// processFileWithErrorHandling records a directory or symlink, or wraps
-// deleted files and skip-errors mode. Returns (skipped, error).
+// processFileStreaming for a regular file with error recovery for deleted
+// files and skip-errors mode. Returns (skipped, error).
 func (s *Scanner) processFileWithErrorHandling(
 	ctx context.Context, fileToProcess *FileToProcess, result *ScanResult,
 ) (bool, error) {
+	// A directory or symlink has no data to open or read; it is only
+	// recorded in the local index. An error recording it stops the run
+	// even under --skip-errors, or the snapshot would complete without it.
+	mode := os.FileMode(fileToProcess.File.Mode)
+	if mode&os.ModeSymlink != 0 || mode.IsDir() {
+		err := s.recordNonRegularFile(ctx, fileToProcess)
+		if err != nil {
+			return false, fmt.Errorf("processing file %s: %w", fileToProcess.Path, err)
+		}
+
+		return false, nil
+	}
+
 	err := s.processFileStreaming(ctx, fileToProcess, result)
 	if err != nil {
+		// A cancelled run stops here even under --skip-errors, rather than
+		// counting this file as failed and going on to the next one.
+		if ctx.Err() != nil {
+			return false, fmt.Errorf("processing file %s: %w", fileToProcess.Path, err)
+		}
 		// A packer/database/encryption/upload failure means the chunk's data
 		// may not have been stored. Skipping the file would let the snapshot
 		// record a file whose chunk is in no blob and cannot be restored, so
@@ -1363,7 +1402,7 @@ func (s *Scanner) processFileWithErrorHandling(
 			log.Warn("File was deleted during backup, skipping",
 				"path", fileToProcess.Path)
 
-			result.FilesSkipped++
+			countFailedFile(fileToProcess, result)
 
 			return true, nil
 		}
@@ -1374,7 +1413,7 @@ func (s *Scanner) processFileWithErrorHandling(
 			s.ui.Errorf("Failed to process %s: %v. Skipping (--skip-errors).",
 				s.ui.Path(fileToProcess.Path), err)
 
-			result.FilesSkipped++
+			countFailedFile(fileToProcess, result)
 
 			return true, nil
 		}
@@ -1385,6 +1424,13 @@ func (s *Scanner) processFileWithErrorHandling(
 	return false, nil
 }
 
+// countFailedFile counts a file that phase 2 could not store as failed
+// and takes its size back out of BytesScanned, where phase 1 put it.
+func countFailedFile(fileToProcess *FileToProcess, result *ScanResult) {
+	result.FilesFailed++
+	result.BytesScanned -= fileToProcess.FileInfo.Size()
+}
+
 // printProcessingProgress prints a periodic progress line during the process phase,
 // showing files processed, bytes transferred, throughput, and ETA
 func (s *Scanner) printProcessingProgress(
@@ -1471,7 +1517,8 @@ func (s *Scanner) finalizeProcessPhase(ctx context.Context, result *ScanResult)
 			// snapshot references has uploaded_ts set, so deduplication and
 			// interrupted-run repair treat these blobs as trustworthy.
 			err = s.repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
-				err := s.repos.Blobs.UpdateUploaded(ctx, tx, b.ID)
+				err := s.repos.Blobs.UpdateUploaded(
+					ctx, tx, b.ID, b.Compressed)
 				if err != nil {
 					return fmt.Errorf("marking blob uploaded: %w", err)
 				}
@@ -1507,23 +1554,28 @@ func (s *Scanner) handleBlobReady(
 	blobPath := fmt.Sprintf("blobs/%s/%s/%s",
 		finishedBlob.Hash[:2], finishedBlob.Hash[2:4], finishedBlob.Hash)
 
-	blobExists, err := s.uploadBlobIfNeeded(
+	storedSize, blobExists, err := s.uploadBlobIfNeeded(
 		ctx, blobPath, blobWithReader, startTime, result)
 	if err != nil {
-		s.cleanupBlobTempFile(blobWithReader)
+		cleanupErr := s.cleanupBlobTempFile(blobWithReader)
 
-		return fmt.Errorf("uploading blob %s: %w", finishedBlob.Hash, err)
+		return errors.Join(
+			fmt.Errorf("uploading blob %s: %w", finishedBlob.Hash, err), cleanupErr)
 	}
 
-	err = s.recordBlobMetadata(ctx, finishedBlob, blobExists, startTime)
+	err = s.recordBlobMetadata(
+		ctx, finishedBlob, storedSize, blobExists, startTime)
 	if err != nil {
-		s.cleanupBlobTempFile(blobWithReader)
+		cleanupErr := s.cleanupBlobTempFile(blobWithReader)
 
+		return errors.Join(err, cleanupErr)
+	}
+
+	err = s.cleanupBlobTempFile(blobWithReader)
+	if err != nil {
 		return err
 	}
 
-	s.cleanupBlobTempFile(blobWithReader)
-
 	// Chunks from this blob are now committed to DB - remove from pending set
 	s.removePendingChunkHashes(blobWithReader.InsertedChunkHashes)
 
@@ -1537,20 +1589,22 @@ func (s *Scanner) handleBlobReady(
 }
 
 // uploadBlobIfNeeded uploads the blob to storage if it doesn't already
-// exist, returns whether it existed
+// exist. It returns the size of the object in storage, which for an
+// existing object can differ from the local copy's, and whether it
+// already existed.
 func (s *Scanner) uploadBlobIfNeeded(
 	ctx context.Context,
 	blobPath string,
 	blobWithReader *blob.WithReader,
 	startTime time.Time,
 	result *ScanResult,
-) (bool, error) {
+) (int64, bool, error) {
 	finishedBlob := blobWithReader.FinishedBlob
 
 	// Check if blob already exists (deduplication after restart)
 	destination := s.storage.Info().Location
 
-	_, err := s.storage.Stat(ctx, blobPath)
+	stored, err := s.storage.Stat(ctx, blobPath)
 	if err == nil {
 		log.Info("Blob already exists in storage, skipping upload",
 			"hash", finishedBlob.Hash,
@@ -1559,7 +1613,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 			s.ui.Hex(finishedBlob.Hash), s.ui.Size(finishedBlob.Compressed),
 			s.ui.Path(destination))
 
-		return true, nil
+		return stored.Size, true, nil
 	}
 
 	s.ui.Beginf("Uploading blob %s (%s) to %s.",
@@ -1573,7 +1627,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 	if err != nil {
 		log.Error("Failed to upload blob", "hash", finishedBlob.Hash, "error", err)
 
-		return false, fmt.Errorf("uploading blob to storage: %w", err)
+		return 0, false, fmt.Errorf("uploading blob to storage: %w", err)
 	}
 
 	uploadDuration := time.Since(startTime)
@@ -1604,7 +1658,7 @@ func (s *Scanner) uploadBlobIfNeeded(
 		stats.BytesUploaded.Add(finishedBlob.Compressed)
 	}
 
-	return false, nil
+	return finishedBlob.Compressed, false, nil
 }
 
 // makeUploadProgressCallback creates a progress callback for blob uploads.
@@ -1674,6 +1728,7 @@ func (s *Scanner) makeUploadProgressCallback(
 func (s *Scanner) recordBlobMetadata(
 	ctx context.Context,
 	finishedBlob *blob.FinishedBlob,
+	storedSize int64,
 	blobExists bool,
 	startTime time.Time,
 ) error {
@@ -1685,7 +1740,8 @@ func (s *Scanner) recordBlobMetadata(
 	uploadDuration := time.Since(startTime)
 
 	return s.repos.WithTx(ctx, func(txCtx context.Context, tx *sql.Tx) error {
-		err := s.repos.Blobs.UpdateUploaded(txCtx, tx, finishedBlob.ID)
+		err := s.repos.Blobs.UpdateUploaded(
+			txCtx, tx, finishedBlob.ID, storedSize)
 		if err != nil {
 			return fmt.Errorf("updating blob upload timestamp: %w", err)
 		}
@@ -1716,20 +1772,22 @@ func (s *Scanner) recordBlobMetadata(
 }
 
 // cleanupBlobTempFile closes and removes the blob's temporary file
-func (s *Scanner) cleanupBlobTempFile(blobWithReader *blob.WithReader) {
+func (s *Scanner) cleanupBlobTempFile(blobWithReader *blob.WithReader) error {
 	if blobWithReader.TempFile != nil {
 		tempName := blobWithReader.TempFile.Name()
 
 		err := blobWithReader.TempFile.Close()
 		if err != nil {
-			log.Fatal("Failed to close temp file", "file", tempName, "error", err)
+			return fmt.Errorf("closing blob temp file: %w", err)
 		}
 
 		err = s.fs.Remove(tempName)
 		if err != nil {
-			log.Fatal("Failed to remove temp file", "file", tempName, "error", err)
+			return fmt.Errorf("removing blob temp file: %w", err)
 		}
 	}
+
+	return nil
 }
 
 // streamingChunkInfo tracks chunk metadata collected during streaming
@@ -1753,16 +1811,11 @@ func (e *packerError) Error() string { return e.err.Error() }
 
 func (e *packerError) Unwrap() error { return e.err }
 
-// processFileStreaming processes a file by streaming chunks directly to the packer
+// processFileStreaming processes a regular file by streaming chunks directly
+// to the packer
 func (s *Scanner) processFileStreaming(
 	ctx context.Context, fileToProcess *FileToProcess, result *ScanResult,
 ) error {
-	// Symlinks and directories have no data to chunk — just record them in the DB.
-	mode := os.FileMode(fileToProcess.File.Mode)
-	if mode&os.ModeSymlink != 0 || mode.IsDir() {
-		return s.recordNonRegularFile(ctx, fileToProcess)
-	}
-
 	file, err := s.fs.Open(fileToProcess.Path)
 	if err != nil {
 		return fmt.Errorf("opening file: %w", wrapPermissionError(fileToProcess.Path, err))
@@ -1828,10 +1881,12 @@ func (s *Scanner) processFileStreaming(
 	return nil
 }
 
-// updateChunkStats counts a chunk that was not already stored. File and
+// updateChunkStats counts a chunk that was not already stored, and its
-// byte counts are not touched here: the scan phase counts each file once.
+// size. The scan result's file counts, BytesScanned and BytesSkipped are
+// not touched here: the scan phase counts each file once.
 func (s *Scanner) updateChunkStats(chunkSize int64, result *ScanResult) {
 	result.ChunksCreated++
+	result.NewChunkSize += chunkSize
 
 	if s.progress != nil {
 		s.progress.GetStats().ChunksCreated.Add(1)
@@ -1979,6 +2034,14 @@ func compileExcludePatterns(patterns []string) []compiledPattern {
 		// Remove trailing slash if present (directory indicator)
 		pattern = strings.TrimSuffix(pattern, "/")
 
+		// A leading "**/" demands a "/" before the name and so misses an
+		// entry directly under the source path. An unanchored pattern
+		// already matches at every depth, so drop the prefix: as in
+		// gitignore, "**/x" then also matches a top-level "x".
+		if !anchored {
+			pattern = strings.TrimPrefix(pattern, "**/")
+		}
+
 		// Compile the glob pattern
 		// For patterns without path separators, we need to match them as components
 		// e.g., ".git" should match ".git" anywhere in the path

internal/snapshot/skip_errors_test.go

@@ -3,6 +3,7 @@ package snapshot_test
 import (
 	"context"
 	"errors"
+	"io"
 	"os"
 	"path/filepath"
 	"strings"
@@ -13,6 +14,7 @@ import (
 	"github.com/spf13/afero"
 	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/ui"
 )
 
 // errSimTempFail is the one-time temp-file creation failure blobTempFailFs
@@ -80,6 +82,56 @@ func (f *readFailFs) Open(name string) (afero.File, error) {
 	return file, nil
 }
 
+// cancelOnOpenFs cancels the run when the scanner opens the target file to
+// back it up, as Ctrl-C partway through processing would, and records each
+// file opened after that.
+type cancelOnOpenFs struct {
+	afero.Fs
+
+	target            string
+	cancel            context.CancelFunc
+	cancelled         bool
+	openedAfterCancel []string
+}
+
+//nolint:ireturn // afero.Fs.Open is defined to return the interface.
+func (f *cancelOnOpenFs) Open(name string) (afero.File, error) {
+	if f.cancelled {
+		f.openedAfterCancel = append(f.openedAfterCancel, name)
+	}
+
+	if name == f.target {
+		f.cancel()
+		f.cancelled = true
+	}
+
+	return f.Fs.Open(name)
+}
+
+// linkRemovedAfterLstatFs is the real filesystem, except that the symlink at
+// target is removed right after the walk lstats it, as happens when a link is
+// deleted during a backup. The scanner's readlink of it then fails.
+type linkRemovedAfterLstatFs struct {
+	afero.OsFs
+
+	t      *testing.T
+	target string
+}
+
+func (f *linkRemovedAfterLstatFs) LstatIfPossible(
+	name string,
+) (os.FileInfo, bool, error) {
+	info, lstatCalled, err := f.OsFs.LstatIfPossible(name)
+	if err == nil && name == f.target {
+		rmErr := os.Remove(name)
+		if rmErr != nil {
+			f.t.Errorf("removing %s: %v", name, rmErr)
+		}
+	}
+
+	return info, lstatCalled, err
+}
+
 // writeSkipErrorTestFile writes one file into fs with a fixed mtime.
 func writeSkipErrorTestFile(t *testing.T, fs afero.Fs, path, content string) {
 	t.Helper()
@@ -102,14 +154,16 @@ func writeSkipErrorTestFile(t *testing.T, fs afero.Fs, path, content string) {
 	}
 }
 
-// runSkipErrorScan scans /source on fs with the given skip-errors setting and
+// runSkipErrorScan scans source on fs under ctx with the given skip-errors
+// setting, printing user-facing messages to uiw (nil discards them), and
 // returns the repositories (for inspection) and the scan error.
 func runSkipErrorScan(
-	t *testing.T, fs afero.Fs, skipErrors bool,
+	ctx context.Context, t *testing.T, fs afero.Fs, source string,
+	skipErrors bool, uiw *ui.Writer,
 ) (*database.Repositories, error) {
 	t.Helper()
 
-	db, err := database.NewTestDB()
+	db, err := database.New(ctx, ":memory:")
 	if err != nil {
 		t.Fatalf("create test db: %v", err)
 	}
@@ -130,14 +184,14 @@ func runSkipErrorScan(
 		MaxBlobSize:      int64(1024 * 1024),
 		CompressionLevel: 3,
 		AgeRecipients:    []string{testAgePublicKey},
+		UI:               uiw,
 		SkipErrors:       skipErrors,
 	})
 
-	ctx := context.Background()
 	snapshotID := "test-snapshot-skip-errors"
 	createTestSnapshotRecord(ctx, t, repos, snapshotID)
 
-	_, err = scanner.Scan(ctx, "/source", snapshotID)
+	_, err = scanner.Scan(ctx, source, snapshotID)
 
 	return repos, err
 }
@@ -157,7 +211,7 @@ func TestScannerPackingFailureAbortsUnderSkipErrors(t *testing.T) {
 	writeSkipErrorTestFile(t, fs, "/source/file1.txt", "first file content")
 	writeSkipErrorTestFile(t, fs, "/source/file2.txt", "second file content")
 
-	repos, err := runSkipErrorScan(t, fs, true)
+	repos, err := runSkipErrorScan(context.Background(), t, fs, "/source", true, nil)
 	if err == nil {
 		t.Fatal("expected scan to abort on the packer error, got nil")
 	}
@@ -184,7 +238,7 @@ func TestScannerReadErrorAbortsWithoutSkipErrors(t *testing.T) {
 	fs := &readFailFs{Fs: afero.NewMemMapFs(), target: target}
 	writeSkipErrorTestFile(t, fs, target, "content that cannot be read")
 
-	_, err := runSkipErrorScan(t, fs, false)
+	_, err := runSkipErrorScan(context.Background(), t, fs, "/source", false, nil)
 	if err == nil {
 		t.Fatal("expected scan to fail on the read error, got nil")
 	}
@@ -200,7 +254,7 @@ func TestScannerReadErrorSkippedWithSkipErrors(t *testing.T) {
 	fs := &readFailFs{Fs: afero.NewMemMapFs(), target: target}
 	writeSkipErrorTestFile(t, fs, target, "content that cannot be read")
 
-	repos, err := runSkipErrorScan(t, fs, true)
+	repos, err := runSkipErrorScan(context.Background(), t, fs, "/source", true, nil)
 	if err != nil {
 		t.Fatalf("expected scan to complete with --skip-errors, got %v", err)
 	}
@@ -214,3 +268,173 @@ func TestScannerReadErrorSkippedWithSkipErrors(t *testing.T) {
 		t.Fatalf("expected unreadable file skipped, got %d chunks", len(chunks))
 	}
 }
+
+// writeSymlinkSource creates a source directory on disk holding one symlink
+// and returns the directory and the symlink's path.
+func writeSymlinkSource(t *testing.T) (string, string) {
+	t.Helper()
+
+	sourceDir := t.TempDir()
+	linkPath := filepath.Join(sourceDir, "link")
+
+	err := os.Symlink("target.txt", linkPath)
+	if err != nil {
+		t.Fatalf("creating symlink: %v", err)
+	}
+
+	return sourceDir, linkPath
+}
+
+// TestScannerUnreadableSymlinkAbortsWithoutSkipErrors checks that a symlink
+// whose target cannot be read aborts the run when --skip-errors is not set.
+func TestScannerUnreadableSymlinkAbortsWithoutSkipErrors(t *testing.T) {
+	t.Parallel()
+
+	sourceDir, linkPath := writeSymlinkSource(t)
+	fs := &linkRemovedAfterLstatFs{t: t, target: linkPath}
+
+	_, err := runSkipErrorScan(context.Background(), t, fs, sourceDir, false, nil)
+	if !errors.Is(err, os.ErrNotExist) {
+		t.Fatalf("expected scan to fail on the removed symlink, got %v", err)
+	}
+}
+
+// TestScannerUnreadableSymlinkSkippedWithSkipErrors checks that a symlink
+// whose target cannot be read is skipped with an error line, and the run
+// completes, when --skip-errors is set.
+func TestScannerUnreadableSymlinkSkippedWithSkipErrors(t *testing.T) {
+	t.Parallel()
+
+	sourceDir, linkPath := writeSymlinkSource(t)
+	fs := &linkRemovedAfterLstatFs{t: t, target: linkPath}
+	uiw := ui.NewWithColor(io.Discard, false)
+
+	repos, err := runSkipErrorScan(context.Background(), t, fs, sourceDir, true, uiw)
+	if err != nil {
+		t.Fatalf("expected scan to complete with --skip-errors, got %v", err)
+	}
+
+	if uiw.ErrorCount() != 1 {
+		t.Fatalf("expected one error line for the symlink, got %d",
+			uiw.ErrorCount())
+	}
+
+	file, err := repos.Files.GetByPath(context.Background(), linkPath)
+	if err != nil {
+		t.Fatalf("getting %s: %v", linkPath, err)
+	}
+
+	if file != nil {
+		t.Fatalf("expected %s not to be recorded", linkPath)
+	}
+}
+
+// TestScannerCancelStopsSkipErrorsRun checks that a --skip-errors backup
+// cancelled partway through processing returns the cancellation error,
+// opens no further file, and reports no file as failed.
+func TestScannerCancelStopsSkipErrorsRun(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name          string
+		targetContent string
+	}{
+		// The cancellation lands while the target is being read.
+		{name: "while reading a file", targetContent: "first file content"},
+		// An empty target has no chunks, so the cancellation goes unnoticed
+		// until the run moves on to the next file.
+		{name: "between files", targetContent: ""},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			const target = "/source/a.txt"
+
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			fs := &cancelOnOpenFs{
+				Fs: afero.NewMemMapFs(), target: target, cancel: cancel,
+			}
+			writeSkipErrorTestFile(t, fs, target, tt.targetContent)
+			writeSkipErrorTestFile(t, fs, "/source/b.txt", "second file content")
+			writeSkipErrorTestFile(t, fs, "/source/c.txt", "third file content")
+
+			uiw := ui.NewWithColor(io.Discard, false)
+
+			_, err := runSkipErrorScan(ctx, t, fs, "/source", true, uiw)
+			if !errors.Is(err, context.Canceled) {
+				t.Fatalf("expected the cancellation error, got %v", err)
+			}
+
+			if len(fs.openedAfterCancel) != 0 {
+				t.Fatalf("expected no file opened after the cancellation, got %v",
+					fs.openedAfterCancel)
+			}
+
+			if uiw.ErrorCount() != 0 {
+				t.Fatalf("expected no file reported as failed, got %d error lines",
+					uiw.ErrorCount())
+			}
+		})
+	}
+}
+
+// TestScannerRemovesBlobTempFileWhenRunStops checks that a run stopped
+// part-way, by a file it cannot read or by a cancellation, removes the
+// temp file of the blob it was packing.
+func TestScannerRemovesBlobTempFileWhenRunStops(t *testing.T) {
+	t.Parallel()
+
+	// a.txt is backed up first and starts a blob; the run stops at b.txt.
+	const target = "/source/b.txt"
+
+	tests := []struct {
+		name  string
+		newFs func(cancel context.CancelFunc) afero.Fs
+	}{
+		{
+			name: "file cannot be read",
+			newFs: func(context.CancelFunc) afero.Fs {
+				return &readFailFs{Fs: afero.NewMemMapFs(), target: target}
+			},
+		},
+		{
+			name: "cancelled",
+			newFs: func(cancel context.CancelFunc) afero.Fs {
+				return &cancelOnOpenFs{
+					Fs: afero.NewMemMapFs(), target: target, cancel: cancel,
+				}
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			ctx, cancel := context.WithCancel(context.Background())
+			defer cancel()
+
+			fs := tt.newFs(cancel)
+			writeSkipErrorTestFile(t, fs, "/source/a.txt", "first file content")
+			writeSkipErrorTestFile(t, fs, target, "second file content")
+
+			_, err := runSkipErrorScan(ctx, t, fs, "/source", false, nil)
+			if err == nil {
+				t.Fatal("expected the scan to stop with an error, got nil")
+			}
+
+			left, err := afero.Glob(fs, filepath.Join(os.TempDir(), "vaultik-blob-*"))
+			if err != nil {
+				t.Fatalf("listing blob temp files: %v", err)
+			}
+
+			if len(left) != 0 {
+				t.Fatalf("expected no blob temp file left, got %v", left)
+			}
+		})
+	}
+}

internal/snapshot/snapshot.go

@@ -105,26 +105,47 @@ func (sm *SnapshotManager) CreateSnapshot(
 	return sm.CreateSnapshotWithName(ctx, hostname, "", version, gitRevision)
 }
 
+// ShortHostname returns hostname up to its first dot. A snapshot ID starts
+// with this form, while the snapshots table stores the full hostname.
+func ShortHostname(hostname string) string {
+	short, _, _ := strings.Cut(hostname, ".")
+
+	return short
+}
+
 // CreateSnapshotWithName creates a new snapshot record with an optional
 // snapshot name. The snapshot ID format is: hostname_name_timestamp or
-// hostname_timestamp if name is empty.
+// hostname_timestamp if name is empty. The timestamp is in whole seconds.
+// If the local index already has a snapshot with that ID, from a run of the
+// same name that started in the same second, it waits a second and takes a
+// new timestamp.
 func (sm *SnapshotManager) CreateSnapshotWithName(
 	ctx context.Context, hostname, name, version, gitRevision string,
 ) (string, error) {
-	// Use short hostname (strip domain if present)
+	shortHostname := ShortHostname(hostname)
-	shortHostname := hostname
-	if before, _, ok := strings.Cut(hostname, "."); ok {
-		shortHostname = before
-	}
-
-	// Build snapshot ID with optional name
-	timestamp := time.Now().UTC().Format("2006-01-02T15:04:05Z")
 
 	var snapshotID string
-	if name != "" {
+
-		snapshotID = fmt.Sprintf("%s_%s_%s", shortHostname, name, timestamp)
+	for {
-	} else {
+		// Build snapshot ID with optional name
-		snapshotID = fmt.Sprintf("%s_%s", shortHostname, timestamp)
+		timestamp := time.Now().UTC().Format("2006-01-02T15:04:05Z")
+
+		if name != "" {
+			snapshotID = fmt.Sprintf("%s_%s_%s", shortHostname, name, timestamp)
+		} else {
+			snapshotID = fmt.Sprintf("%s_%s", shortHostname, timestamp)
+		}
+
+		existing, err := sm.repos.Snapshots.GetByID(ctx, snapshotID)
+		if err != nil {
+			return "", fmt.Errorf("looking up snapshot %s: %w", snapshotID, err)
+		}
+
+		if existing == nil {
+			break
+		}
+
+		time.Sleep(time.Second)
 	}
 
 	snapshot := &database.Snapshot{
@@ -178,6 +199,7 @@ func (sm *SnapshotManager) UpdateSnapshotStatsExtended(
 			stats.CompressionLevel,
 			stats.BytesUploaded,
 			stats.UploadDurationMs,
+			stats.NewChunkSize,
 		)
 	})
 }
@@ -871,7 +893,7 @@ func (sm *SnapshotManager) getFileSize(path string) int64 {
 // BackupStats contains statistics from a backup operation
 type BackupStats struct {
 	FilesScanned  int
-	TotalSize     int64 // Total size of all files examined
+	TotalSize     int64 // Total size of the files in the snapshot
 	ChunksCreated int
 	BlobsCreated  int
 	BytesUploaded int64
@@ -885,6 +907,7 @@ type ExtendedBackupStats struct {
 	BlobUncompressedSize int64 // Total uncompressed size of all referenced blobs
 	CompressionLevel     int   // Compression level used for this snapshot
 	UploadDurationMs     int64 // Total milliseconds spent uploading to S3
+	NewChunkSize         int64 // Total size of the chunks counted in ChunksCreated
 }
 
 // deleteOtherSnapshots deletes all snapshots except the current one

internal/snapshot/snapshot_test.go

@@ -386,3 +386,36 @@ func TestCleanSnapshotDBNonExistentSnapshot(t *testing.T) {
 		t.Fatalf("unexpected error: %v", err)
 	}
 }
+
+// Two creates of one snapshot name back to back start within one second,
+// the resolution of the timestamp in a snapshot ID. See
+// https://git.eeqj.de/sneak/vaultik/issues/270.
+func TestCreateSnapshotWithNameTwiceBackToBack(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, filepath.Join(t.TempDir(), "index.sqlite"))
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+
+	defer func() { _ = db.Close() }()
+
+	sm := &SnapshotManager{repos: database.NewRepositories(db)}
+
+	first, err := sm.CreateSnapshotWithName(ctx, "test-host", "data", "v", "g")
+	if err != nil {
+		t.Fatalf("first create failed: %v", err)
+	}
+
+	second, err := sm.CreateSnapshotWithName(ctx, "test-host", "data", "v", "g")
+	if err != nil {
+		t.Fatalf("second create failed: %v", err)
+	}
+
+	if first == second {
+		t.Fatalf("both creates returned snapshot ID %s", first)
+	}
+}

internal/storage/conformance_test.go

@@ -14,8 +14,9 @@ import (
 
 // runStorerConformance is the shared Storer contract. Every backend that
 // can run in-process is expected to pass it: TestFileStorer runs it against
-// file://, TestS3Storer against s3://. A new backend inherits this coverage
+// file://, TestS3Storer against s3://, TestRcloneStorer against rclone's
-// by passing its own constructor, so the contract is defined once.
+// local backend. A new backend inherits this coverage by passing its own
+// constructor, so the contract is defined once.
 //
 // It exercises the public Storer interface: round-trip, stat, list with
 // prefix filtering, overwrite, delete, delete-of-missing, and not-found on

internal/storage/faultstore/faultstore.go

@@ -1,15 +1,16 @@
 // Package faultstore provides a storage.Storer wrapper that injects
 // faults on demand, so tests can reproduce the failure modes a real
 // backend exhibits: an upload that fails partway, a backend that reports
-// success while storing nothing, and reads that return corrupt or
+// success while storing nothing, reads that return corrupt or truncated
-// truncated bytes. It is the seam called for by the fault-injection
+// bytes, and a delete the backend refuses. It is the seam called for by
-// tests (sneak/vaultik issue 72) and is meant to be reused by future
+// the fault-injection tests (sneak/vaultik issue 72) and is meant to be
-// tests rather than re-implemented per case.
+// reused by future tests rather than re-implemented per case.
 //
-// The wrapper delegates every method to the inner Storer. Two hooks
+// The wrapper delegates every method to the inner Storer. Three hooks
-// change that: OnPut decides the fate of each write, and OnGet decides
+// change that: OnPut decides the fate of each write, OnGet decides how
-// how each read's bytes are returned. Both are keyed by the object key,
+// each read's bytes are returned, and OnDelete decides whether each
-// so a test can fault only blobs, only metadata, or a single object.
+// delete fails. All are keyed by the object key, so a test can fault
+// only blobs, only metadata, or a single object.
 package faultstore
 
 import (
@@ -25,6 +26,10 @@ import (
 // ErrInjectedUpload is returned by a Put the OnPut hook chose to fail.
 var ErrInjectedUpload = errors.New("faultstore: injected upload failure")
 
+// ErrInjectedDelete is returned by a Delete the OnDelete hook chose to
+// fail.
+var ErrInjectedDelete = errors.New("faultstore: injected delete failure")
+
 // PutAction is the disposition OnPut assigns to a write.
 type PutAction int
 
@@ -66,6 +71,11 @@ type Storer struct {
 	// OnGet, when set, is consulted for every Get with the object key
 	// and damages the returned bytes accordingly.
 	OnGet func(key string) GetFault
+
+	// OnDelete, when set, is consulted before every Delete with the
+	// object key. Returning true fails the delete and leaves the object
+	// in place.
+	OnDelete func(key string) bool
 }
 
 // New wraps inner. inner must be non-nil.
@@ -154,8 +164,12 @@ func (f *Storer) Stat(ctx context.Context, key string) (*storage.ObjectInfo, err
 	return f.inner.Stat(ctx, key)
 }
 
-// Delete delegates unchanged.
+// Delete removes the object unless OnDelete fails the delete.
 func (f *Storer) Delete(ctx context.Context, key string) error {
+	if f.OnDelete != nil && f.OnDelete(key) {
+		return fmt.Errorf("%w for %q", ErrInjectedDelete, key)
+	}
+
 	return f.inner.Delete(ctx, key)
 }
 
@@ -171,6 +185,11 @@ func (f *Storer) ListStream(
 	return f.inner.ListStream(ctx, prefix)
 }
 
+// DeletePartialUploads delegates unchanged.
+func (f *Storer) DeletePartialUploads(ctx context.Context, prefix string) error {
+	return f.inner.DeletePartialUploads(ctx, prefix)
+}
+
 // Info delegates unchanged.
 func (f *Storer) Info() storage.Info {
 	return f.inner.Info()

internal/storage/file.go

@@ -21,9 +21,13 @@ type FileStorer struct {
 // NewFileStorer creates a new filesystem storage backend.
 //
 // Construction is intentionally cheap and does not touch the filesystem.
-// The basePath is recorded; the directory is created lazily on first
+// The backend never creates basePath: the user does, so that when
-// write. Write operations (Put/PutWithProgress) call MkdirAll for the
+// basePath is a directory on an unplugged volume, a write fails instead
-// per-blob parent directory, which also covers basePath on first use.
+// of landing on the disk under the mount point. This does not hold when
+// basePath is the mount point itself, which stays behind when its volume
+// is unplugged.
+// Put and PutWithProgress fail on a missing basePath and create the
+// directories below it as needed.
 // Get and Stat report a key under a missing basePath as ErrNotFound.
 // List and ListStream fail on a missing basePath, because listing it as
 // an empty store would make `prune` drop every local snapshot record.
@@ -50,9 +54,10 @@ const storageDirPerm = 0o755
 // temp file carrying this suffix and only renames it onto the real key once
 // the whole object is on disk, so an interrupted write can never leave a
 // truncated object at the key a later run would Stat and trust as a complete
-// blob. List and ListStream skip these files, so a leftover from an
+// blob. The rclone backend's upload does the same on remotes with a
-// interrupted write is never listed or trusted as a blob; it is otherwise
+// server-side move. List and ListStream skip these files, so a leftover from
-// harmless and is overwritten when the same key is written again.
+// an interrupted write is never listed or trusted as a blob; it is otherwise
+// harmless.
 const tempSuffix = ".partial"
 
 // Put stores data at the specified key.
@@ -241,6 +246,42 @@ func (f *FileStorer) ListStream(ctx context.Context, prefix string) <-chan Objec
 	return ch
 }
 
+// DeletePartialUploads removes every file under prefix whose name ends in
+// tempSuffix. A missing prefix has none to remove.
+func (f *FileStorer) DeletePartialUploads(ctx context.Context, prefix string) error {
+	basePath := f.fullPath(prefix)
+
+	exists, err := afero.Exists(f.fs, basePath)
+	if err != nil {
+		return fmt.Errorf("checking path: %w", err)
+	}
+
+	if !exists {
+		return nil
+	}
+
+	err = afero.Walk(f.fs, basePath, func(path string, info os.FileInfo, err error) error {
+		if err != nil {
+			return err
+		}
+
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+
+		if info.IsDir() || !strings.HasSuffix(info.Name(), tempSuffix) {
+			return nil
+		}
+
+		return f.fs.Remove(path)
+	})
+	if err != nil {
+		return fmt.Errorf("walking directory: %w", err)
+	}
+
+	return nil
+}
+
 // Info returns human-readable storage location information.
 func (f *FileStorer) Info() Info {
 	return Info{
@@ -260,7 +301,19 @@ func (f *FileStorer) writeAtomic(
 	path := f.fullPath(key)
 	dir := filepath.Dir(path)
 
-	err := f.fs.MkdirAll(dir, storageDirPerm)
+	// Checked on every write, so a destination that disappears part-way
+	// through a backup fails it rather than being recreated by MkdirAll.
+	_, err := f.fs.Stat(f.basePath)
+	if os.IsNotExist(err) {
+		return fmt.Errorf("destination directory does not exist; "+
+			"create it, or mount the volume it is on: %w", err)
+	}
+
+	if err != nil {
+		return fmt.Errorf("checking destination directory: %w", err)
+	}
+
+	err = f.fs.MkdirAll(dir, storageDirPerm)
 	if err != nil {
 		return fmt.Errorf("creating directories: %w", err)
 	}

internal/storage/file_atomic_test.go

@@ -14,6 +14,9 @@ import (
 // errStreamInterrupted stands in for an upload cut off mid-stream.
 var errStreamInterrupted = errors.New("connection reset mid-upload")
 
+// testBlobKey is a key laid out as a blob's key is.
+const testBlobKey = "blobs/aa/bb/aabbccddeeff"
+
 // failingReader yields its data once, then fails.
 type failingReader struct {
 	data []byte
@@ -43,7 +46,7 @@ func TestFileStorer_InterruptedWriteLeavesNoTrustedObject(t *testing.T) {
 	}
 
 	ctx := context.Background()
-	key := "blobs/aa/bb/aabbccddeeff"
+	key := testBlobKey
 
 	err = f.PutWithProgress(ctx, key, &failingReader{data: []byte("partial")}, 4096, nil)
 	if err == nil {
@@ -79,7 +82,7 @@ func TestFileStorer_ListSkipsPartialFiles(t *testing.T) {
 	}
 
 	ctx := context.Background()
-	realKey := "blobs/aa/bb/aabbccddeeff"
+	realKey := testBlobKey
 
 	err = f.Put(ctx, realKey, strings.NewReader("blob-bytes"))
 	if err != nil {
@@ -117,3 +120,45 @@ func TestFileStorer_ListSkipsPartialFiles(t *testing.T) {
 		t.Fatalf("ListStream should return only the real key, got %v", streamed)
 	}
 }
+
+// TestFileStorer_DeletePartialUploads checks that a leftover temp file is
+// removed and the object at the real key is kept.
+func TestFileStorer_DeletePartialUploads(t *testing.T) {
+	t.Parallel()
+
+	base := t.TempDir()
+
+	f, err := storage.NewFileStorer(base)
+	if err != nil {
+		t.Fatalf("NewFileStorer: %v", err)
+	}
+
+	ctx := context.Background()
+
+	err = f.Put(ctx, testBlobKey, strings.NewReader("blob-bytes"))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	leftover := filepath.Join(base, testBlobKey+"-123456.partial")
+
+	err = os.WriteFile(leftover, []byte("half"), 0o600)
+	if err != nil {
+		t.Fatalf("writing leftover temp file: %v", err)
+	}
+
+	err = f.DeletePartialUploads(ctx, "blobs/")
+	if err != nil {
+		t.Fatalf("DeletePartialUploads: %v", err)
+	}
+
+	_, err = os.Stat(leftover)
+	if !os.IsNotExist(err) {
+		t.Errorf("leftover temp file was not removed: %v", err)
+	}
+
+	_, err = f.Stat(ctx, testBlobKey)
+	if err != nil {
+		t.Errorf("Stat of the real key: %v", err)
+	}
+}

internal/storage/file_backend_test.go

@@ -4,7 +4,9 @@ import (
 	"context"
 	"errors"
 	"io/fs"
+	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 
 	"sneak.berlin/go/vaultik/internal/storage"
@@ -60,3 +62,43 @@ func TestFileStorerListMissingDestination(t *testing.T) {
 		t.Errorf("ListStream error = %v, want a not-exist error", streamErr)
 	}
 }
+
+// TestFileStorerPutMissingDestination checks that a write fails when the
+// destination directory has gone, even after earlier writes to it
+// succeeded, and that it does not create the directory again.
+func TestFileStorerPutMissingDestination(t *testing.T) {
+	t.Parallel()
+
+	ctx := context.Background()
+	base := filepath.Join(t.TempDir(), "usbstick")
+
+	err := os.Mkdir(base, 0o750)
+	if err != nil {
+		t.Fatalf("Mkdir: %v", err)
+	}
+
+	s, err := storage.NewFileStorer(base)
+	if err != nil {
+		t.Fatalf("NewFileStorer: %v", err)
+	}
+
+	err = s.Put(ctx, "blobs/aa/bb/first", strings.NewReader("first"))
+	if err != nil {
+		t.Fatalf("first Put: %v", err)
+	}
+
+	err = os.Rename(base, base+"-unplugged")
+	if err != nil {
+		t.Fatalf("Rename: %v", err)
+	}
+
+	err = s.Put(ctx, "blobs/aa/bb/second", strings.NewReader("second"))
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("Put = %v, want a not-exist error", err)
+	}
+
+	_, err = os.Stat(base)
+	if !errors.Is(err, fs.ErrNotExist) {
+		t.Errorf("Put created the destination directory again (Stat = %v)", err)
+	}
+}

internal/storage/module.go

@@ -99,6 +99,7 @@ func storerFromParsedS3URL(parsed *URL, cfg *config.Config) (Storer, error) {
 		AccessKeyID:     cfg.S3.AccessKeyID,
 		SecretAccessKey: cfg.S3.SecretAccessKey,
 		Region:          region,
+		PartSize:        cfg.S3.PartSize.Int64(),
 	})
 	if err != nil {
 		return nil, fmt.Errorf("creating S3 client: %w", err)
@@ -134,6 +135,7 @@ func storerFromLegacyS3Config(cfg *config.Config) (Storer, error) {
 		AccessKeyID:     cfg.S3.AccessKeyID,
 		SecretAccessKey: cfg.S3.SecretAccessKey,
 		Region:          region,
+		PartSize:        cfg.S3.PartSize.Int64(),
 	})
 	if err != nil {
 		return nil, fmt.Errorf("creating S3 client: %w", err)

internal/storage/rclone.go

@@ -3,6 +3,7 @@ package storage
 import (
 	"bytes"
 	"context"
+	"crypto/rand"
 	"errors"
 	"fmt"
 	"io"
@@ -68,14 +69,7 @@ func (r *RcloneStorer) Put(ctx context.Context, key string, data io.Reader) erro
 		return fmt.Errorf("reading data: %w", err)
 	}
 
-	// Upload the object
+	return r.upload(ctx, key, bytes.NewReader(buf))
-	_, err = operations.Rcat(ctx, r.fsys, key,
-		io.NopCloser(bytes.NewReader(buf)), time.Now(), nil)
-	if err != nil {
-		return fmt.Errorf("uploading object: %w", err)
-	}
-
-	return nil
 }
 
 // PutWithProgress stores data with progress reporting.
@@ -89,13 +83,7 @@ func (r *RcloneStorer) PutWithProgress(
 		callback: progress,
 	}
 
-	// Upload the object
+	return r.upload(ctx, key, pr)
-	_, err := operations.Rcat(ctx, r.fsys, key, io.NopCloser(pr), time.Now(), nil)
-	if err != nil {
-		return fmt.Errorf("uploading object: %w", err)
-	}
-
-	return nil
 }
 
 // Get retrieves data from the specified key.
@@ -173,6 +161,10 @@ func (r *RcloneStorer) List(ctx context.Context, prefix string) ([]string, error
 
 	err := operations.ListFn(ctx, r.fsys, func(obj fs.Object) {
 		key := obj.Remote()
+		if strings.HasSuffix(key, tempSuffix) {
+			return
+		}
+
 		if prefix == "" || strings.HasPrefix(key, prefix) {
 			keys = append(keys, key)
 		}
@@ -202,6 +194,10 @@ func (r *RcloneStorer) ListStream(
 			}
 
 			key := obj.Remote()
+			if strings.HasSuffix(key, tempSuffix) {
+				return
+			}
+
 			if prefix == "" || strings.HasPrefix(key, prefix) {
 				ch <- ObjectInfo{
 					Key:  key,
@@ -217,6 +213,31 @@ func (r *RcloneStorer) ListStream(
 	return ch
 }
 
+// DeletePartialUploads removes every object under prefix whose name ends
+// in tempSuffix.
+func (r *RcloneStorer) DeletePartialUploads(ctx context.Context, prefix string) error {
+	var partial []fs.Object
+
+	err := operations.ListFn(ctx, r.fsys, func(obj fs.Object) {
+		key := obj.Remote()
+		if strings.HasPrefix(key, prefix) && strings.HasSuffix(key, tempSuffix) {
+			partial = append(partial, obj)
+		}
+	})
+	if err != nil {
+		return fmt.Errorf("listing objects: %w", err)
+	}
+
+	for _, obj := range partial {
+		err = obj.Remove(ctx)
+		if err != nil {
+			return fmt.Errorf("removing object: %w", err)
+		}
+	}
+
+	return nil
+}
+
 // Info returns human-readable storage location information.
 func (r *RcloneStorer) Info() Info {
 	location := r.remote
@@ -230,6 +251,59 @@ func (r *RcloneStorer) Info() Info {
 	}
 }
 
+// upload writes data to key. Where the remote has a server-side move, it
+// writes under a temporary name ending in tempSuffix and moves the object
+// onto key once it is complete, so a killed upload cannot leave a truncated
+// object at key; a remote without one is written in place. List and
+// ListStream skip a temporary object left behind.
+//
+// rclone's own copy does this only where the remote also sets
+// PartialUploads. That flag is not checked here: hdfs, for one, shows a
+// file while it is written without setting it.
+func (r *RcloneStorer) upload(ctx context.Context, key string, data io.Reader) error {
+	if r.fsys.Features().Move == nil {
+		_, err := operations.Rcat(ctx, r.fsys, key, io.NopCloser(data), time.Now(), nil)
+		if err != nil {
+			return fmt.Errorf("uploading object: %w", err)
+		}
+
+		return nil
+	}
+
+	tempKey := key + "-" + rand.Text() + tempSuffix
+
+	obj, err := operations.Rcat(ctx, r.fsys, tempKey, io.NopCloser(data), time.Now(), nil)
+	if err != nil {
+		// Rcat returns the object it wrote when the written data fails its check.
+		if obj != nil {
+			_ = obj.Remove(ctx)
+		}
+
+		return fmt.Errorf("uploading object: %w", err)
+	}
+
+	// On drive, dropbox, onedrive and others the remote's own move does not
+	// replace an object already at key. operations.Move removes the object
+	// it is given first, and copies where the remote refuses the move.
+	existing, err := r.fsys.NewObject(ctx, key)
+	if errors.Is(err, fs.ErrorObjectNotFound) {
+		existing = nil
+	} else if err != nil {
+		_ = obj.Remove(ctx)
+
+		return fmt.Errorf("looking up existing object: %w", err)
+	}
+
+	_, err = operations.Move(ctx, r.fsys, existing, key, obj)
+	if err != nil {
+		_ = obj.Remove(ctx)
+
+		return fmt.Errorf("moving object into place: %w", err)
+	}
+
+	return nil
+}
+
 // progressReader wraps an io.Reader to track read progress.
 type progressReader struct {
 	reader   io.Reader

internal/storage/rclone_test.go

@@ -1,28 +1,47 @@
 package storage_test
 
 import (
+	"bytes"
 	"context"
 	"errors"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
 	"testing"
 
+	"github.com/rclone/rclone/fs"
+	"github.com/rclone/rclone/fs/config/configmap"
 	"sneak.berlin/go/vaultik/internal/storage"
 )
 
-// The rclone backend is a thin adapter over the rclone library: it turns a
-// (remote, path) pair into rclone's "remote:path" string, hands it to
-// rclone, and maps rclone's own results back to the Storer interface. What
-// can be tested in-process, without a configured remote or network, is that
-// adapter layer — how the arguments are shaped and how construction errors
-// are reported. The data-plane operations (Put/Get/List/Delete) are rclone's
-// own, exercised against a real provider (drive, s3-via-rclone, ...), which
-// needs a configured remote with credentials and network access and so is
-// out of reach of a unit test. The shared Storer conformance suite therefore
-// runs against the in-process file and s3 backends; the rclone backend
-// inherits that contract once a remote is configured.
-//
 // These tests use rclone's ":local:" on-the-fly backend, which addresses the
-// local filesystem directly without any configured remote, so construction
+// local filesystem directly without any configured remote, so they run
-// runs entirely in-process.
+// entirely in-process. A remote that needs credentials and network access
+// (drive, s3 via rclone, ...) is out of reach of a unit test.
+
+// newRcloneStorer builds an rclone backend on rclone's local backend,
+// rooted at a fresh temp directory.
+//
+//nolint:ireturn // conformance runs against the Storer interface by design
+func newRcloneStorer(t *testing.T) storage.Storer {
+	t.Helper()
+
+	s, err := storage.NewRcloneStorer(context.Background(), ":local", t.TempDir())
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	return s
+}
+
+// TestRcloneStorer runs the shared Storer contract against the rclone
+// backend.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorer(t *testing.T) {
+	runStorerConformance(t, newRcloneStorer)
+}
 
 // TestNewRcloneStorerConstruction checks that a valid remote constructs a
 // backend and that Info() reports the shaped "remote:path" location.
@@ -44,6 +63,321 @@ func TestNewRcloneStorerConstruction(t *testing.T) {
 	}
 }
 
+// unbufferedUploadContext returns a context in which rclone neither reads
+// ahead of the write nor holds a small upload in memory. Without it, a
+// progress callback runs before anything is written to the remote.
+func unbufferedUploadContext() context.Context {
+	ctx, ci := fs.AddConfig(context.Background())
+	ci.BufferSize = 0
+	ci.StreamingUploadCutoff = 0
+
+	return ctx
+}
+
+// objectAtKeyDuringUpload uploads data to testBlobKey and reports whether
+// an object was at the key before the upload finished. It fails the test
+// unless the key then reads back as the uploaded data.
+func objectAtKeyDuringUpload(
+	ctx context.Context, t *testing.T, s *storage.RcloneStorer,
+) bool {
+	t.Helper()
+
+	data := bytes.Repeat([]byte("blob-bytes"), 1000)
+	seen := false
+
+	err := s.PutWithProgress(ctx, testBlobKey, bytes.NewReader(data),
+		int64(len(data)), func(int64) error {
+			_, statErr := s.Stat(ctx, testBlobKey)
+			if statErr == nil {
+				seen = true
+			}
+
+			return nil
+		})
+	if err != nil {
+		t.Fatalf("PutWithProgress: %v", err)
+	}
+
+	got := readObject(ctx, t, s, testBlobKey)
+	if !bytes.Equal(got, data) {
+		t.Errorf("object read back as %d bytes, want the %d uploaded",
+			len(got), len(data))
+	}
+
+	return seen
+}
+
+// readObject returns the contents of the object at key.
+func readObject(
+	ctx context.Context, t *testing.T, s *storage.RcloneStorer, key string,
+) []byte {
+	t.Helper()
+
+	rc, err := s.Get(ctx, key)
+	if err != nil {
+		t.Fatalf("Get: %v", err)
+	}
+
+	defer func() { _ = rc.Close() }()
+
+	got, err := io.ReadAll(rc)
+	if err != nil {
+		t.Fatalf("reading object: %v", err)
+	}
+
+	return got
+}
+
+// TestRcloneStorerObjectAppearsOnlyWhenComplete checks that on a remote
+// with a server-side move, such as local, nothing is at the key until the
+// upload has finished, so a killed upload cannot leave a truncated object
+// there.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerObjectAppearsOnlyWhenComplete(t *testing.T) {
+	ctx := unbufferedUploadContext()
+
+	s, err := storage.NewRcloneStorer(ctx, ":local", t.TempDir())
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	if objectAtKeyDuringUpload(ctx, t, s) {
+		t.Error("object was at its key before the upload finished")
+	}
+}
+
+// TestRcloneStorerListSkipsPartialFiles checks that a temporary file left
+// by a killed upload is never listed as a key.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerListSkipsPartialFiles(t *testing.T) {
+	dir := t.TempDir()
+	ctx := context.Background()
+
+	s, err := storage.NewRcloneStorer(ctx, ":local", dir)
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	realKey := testBlobKey
+
+	err = s.Put(ctx, realKey, strings.NewReader("blob-bytes"))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	leftover := filepath.Join(dir, realKey+"-123456.partial")
+
+	err = os.WriteFile(leftover, []byte("half"), 0o600)
+	if err != nil {
+		t.Fatalf("writing leftover temp file: %v", err)
+	}
+
+	keys, err := s.List(ctx, "blobs/")
+	if err != nil {
+		t.Fatalf("List: %v", err)
+	}
+
+	if len(keys) != 1 || keys[0] != realKey {
+		t.Fatalf("List should return only the real key, got %v", keys)
+	}
+
+	var streamed []string
+
+	for obj := range s.ListStream(ctx, "blobs/") {
+		if obj.Err != nil {
+			t.Fatalf("ListStream: %v", obj.Err)
+		}
+
+		streamed = append(streamed, obj.Key)
+	}
+
+	if len(streamed) != 1 || streamed[0] != realKey {
+		t.Fatalf("ListStream should return only the real key, got %v", streamed)
+	}
+}
+
+// TestRcloneStorerDeletePartialUploads checks that a temporary file left
+// by a killed upload is removed and the object at the real key is kept.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerDeletePartialUploads(t *testing.T) {
+	dir := t.TempDir()
+	ctx := context.Background()
+
+	s, err := storage.NewRcloneStorer(ctx, ":local", dir)
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	err = s.Put(ctx, testBlobKey, strings.NewReader("blob-bytes"))
+	if err != nil {
+		t.Fatalf("Put: %v", err)
+	}
+
+	leftover := filepath.Join(dir, testBlobKey+"-123456.partial")
+
+	err = os.WriteFile(leftover, []byte("half"), 0o600)
+	if err != nil {
+		t.Fatalf("writing leftover temp file: %v", err)
+	}
+
+	err = s.DeletePartialUploads(ctx, "blobs/")
+	if err != nil {
+		t.Fatalf("DeletePartialUploads: %v", err)
+	}
+
+	_, err = os.Stat(leftover)
+	if !os.IsNotExist(err) {
+		t.Errorf("leftover temp file was not removed: %v", err)
+	}
+
+	_, err = s.Stat(ctx, testBlobKey)
+	if err != nil {
+		t.Errorf("Stat of the real key: %v", err)
+	}
+}
+
+// newRcloneStorerOnWrappedLocal registers name as rclone's local backend
+// wrapped by wrap, and builds an rclone backend on it rooted at a fresh
+// temp directory. wrap changes the features the local backend reports, so
+// that it behaves like a remote a unit test cannot reach.
+func newRcloneStorerOnWrappedLocal(
+	ctx context.Context, t *testing.T, name string, wrap func(fs.Fs) fs.Fs,
+) *storage.RcloneStorer {
+	t.Helper()
+
+	fs.Register(&fs.RegInfo{
+		Name: name,
+		NewFs: func(
+			ctx context.Context, _, root string, _ configmap.Mapper,
+		) (fs.Fs, error) {
+			local, err := fs.NewFs(ctx, ":local:"+root)
+			if err != nil {
+				return nil, err
+			}
+
+			return wrap(local), nil
+		},
+	})
+
+	s, err := storage.NewRcloneStorer(ctx, ":"+name, t.TempDir())
+	if err != nil {
+		t.Fatalf("NewRcloneStorer: %v", err)
+	}
+
+	return s
+}
+
+// withoutPartialUploads is rclone's local backend with the PartialUploads
+// flag cleared. Like hdfs, it then has a server-side move and shows a file
+// while it is written, without setting that flag.
+type withoutPartialUploads struct {
+	fs.Fs
+}
+
+func (f *withoutPartialUploads) Features() *fs.Features {
+	features := *f.Fs.Features()
+	features.PartialUploads = false
+
+	return &features
+}
+
+// TestRcloneStorerMovesIntoPlaceWithoutPartialUploads checks that on a
+// remote with a server-side move nothing is at the key until the upload has
+// finished, even when rclone does not mark the remote as showing partial
+// uploads.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerMovesIntoPlaceWithoutPartialUploads(t *testing.T) {
+	ctx := unbufferedUploadContext()
+	s := newRcloneStorerOnWrappedLocal(ctx, t, "withoutpartialuploads",
+		func(local fs.Fs) fs.Fs { return &withoutPartialUploads{Fs: local} })
+
+	if objectAtKeyDuringUpload(ctx, t, s) {
+		t.Error("object was at its key before the upload finished")
+	}
+}
+
+// withoutMove is rclone's local backend reporting no server-side move.
+type withoutMove struct {
+	fs.Fs
+}
+
+func (f *withoutMove) Features() *fs.Features {
+	features := *f.Fs.Features()
+	features.Move = nil
+
+	return &features
+}
+
+// TestRcloneStorerWritesInPlaceWithoutMove checks that on a remote with no
+// server-side move an object is written straight to its key and reads back
+// from there.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerWritesInPlaceWithoutMove(t *testing.T) {
+	ctx := unbufferedUploadContext()
+	s := newRcloneStorerOnWrappedLocal(ctx, t, "withoutmove",
+		func(local fs.Fs) fs.Fs { return &withoutMove{Fs: local} })
+
+	if !objectAtKeyDuringUpload(ctx, t, s) {
+		t.Error("object was not at its key while it was uploaded")
+	}
+}
+
+var errNameConflict = errors.New("an object with this name already exists")
+
+// moveRefusesExisting is rclone's local backend with a server-side move
+// that, like dropbox's or onedrive's, refuses to move onto an existing
+// object.
+type moveRefusesExisting struct {
+	fs.Fs
+}
+
+func (f *moveRefusesExisting) Features() *fs.Features {
+	features := *f.Fs.Features()
+	features.Move = f.move
+
+	return &features
+}
+
+//nolint:ireturn // the signature is rclone's
+func (f *moveRefusesExisting) move(
+	ctx context.Context, src fs.Object, remote string,
+) (fs.Object, error) {
+	_, err := f.NewObject(ctx, remote)
+	if err == nil {
+		return nil, errNameConflict
+	}
+
+	return f.Fs.Features().Move(ctx, src, remote)
+}
+
+// TestRcloneStorerOverwritesWhereMoveRefusesExisting checks that writing a
+// key twice replaces the object on a remote whose server-side move will not
+// replace an existing object.
+//
+//nolint:paralleltest // NewRcloneStorer installs the process-global rclone config
+func TestRcloneStorerOverwritesWhereMoveRefusesExisting(t *testing.T) {
+	ctx := context.Background()
+	s := newRcloneStorerOnWrappedLocal(ctx, t, "moverefusesexisting",
+		func(local fs.Fs) fs.Fs { return &moveRefusesExisting{Fs: local} })
+
+	for _, content := range []string{"first", "second"} {
+		err := s.Put(ctx, testBlobKey, strings.NewReader(content))
+		if err != nil {
+			t.Fatalf("Put %q: %v", content, err)
+		}
+	}
+
+	got := readObject(ctx, t, s, testBlobKey)
+	if string(got) != "second" {
+		t.Errorf("object = %q, want %q", got, "second")
+	}
+}
+
 // TestNewRcloneStorerUnknownRemote checks that a remote that is not in the
 // rclone config fails construction with the ErrRemoteNotFound sentinel,
 // rather than silently returning a backend pointed nowhere.

internal/storage/s3.go

@@ -99,6 +99,12 @@ func (s *S3Storer) ListStream(ctx context.Context, prefix string) <-chan ObjectI
 	return ch
 }
 
+// DeletePartialUploads has nothing to remove: S3 shows an object only once
+// its upload has completed, so an upload cut off part-way leaves none.
+func (s *S3Storer) DeletePartialUploads(_ context.Context, _ string) error {
+	return nil
+}
+
 // Info returns human-readable storage location information.
 func (s *S3Storer) Info() Info {
 	return Info{

internal/storage/s3_test.go

@@ -1,11 +1,15 @@
 package storage_test
 
 import (
+	"bytes"
 	"context"
 	"errors"
+	"io"
+	"net/http"
 	"net/http/httptest"
 	"slices"
 	"strings"
+	"sync/atomic"
 	"testing"
 
 	"github.com/johannesboyne/gofakes3"
@@ -19,6 +23,16 @@ import (
 // s3TestBucket is the bucket created for each in-process S3 server.
 const s3TestBucket = "test-bucket"
 
+// Credentials for the tests' S3 clients. The test servers accept any.
+const (
+	s3TestAccessKeyID     = "key"
+	s3TestSecretAccessKey = "secret"
+)
+
+// s3TestRegion is the region of the S3 clients built here with
+// s3.NewClient. The test servers ignore it.
+const s3TestRegion = "us-east-1"
+
 // newS3Storer builds an s3:// backend backed by a fresh in-process
 // S3 server. It reuses the same in-memory S3 harness (gofakes3 + s3mem
 // over httptest) that internal/s3 and the not-found regression test use,
@@ -45,7 +59,7 @@ func newS3Storer(t *testing.T) storage.Storer {
 		Bucket:          s3TestBucket,
 		AccessKeyID:     "test",
 		SecretAccessKey: "test",
-		Region:          "us-east-1",
+		Region:          s3TestRegion,
 	})
 	if err != nil {
 		t.Fatalf("new client: %v", err)
@@ -83,6 +97,68 @@ func TestS3StorerMissingKeyMapsToErrNotFound(t *testing.T) {
 	}
 }
 
+// TestS3ListingCutShortIsAnError checks that List and ListStream fail on a
+// listing page marked truncated that carries no continuation token, as a
+// server answering V2 listings V1-style sends it. Taken as the last page,
+// it would hide every later key, and prune would treat the blobs of the
+// snapshots listed there as unreferenced.
+func TestS3ListingCutShortIsAnError(t *testing.T) {
+	t.Parallel()
+
+	const (
+		prefix        = "metadata/"
+		truncatedPage = `<?xml version="1.0" encoding="UTF-8"?>
+<ListBucketResult xmlns="http://s3.amazonaws.com/doc/2006-03-01/">
+  <Name>test-bucket</Name>
+  <Prefix>metadata/</Prefix>
+  <KeyCount>1</KeyCount>
+  <MaxKeys>1</MaxKeys>
+  <IsTruncated>true</IsTruncated>
+  <Contents>
+    <Key>metadata/snap/manifest.json.zst</Key>
+    <Size>8</Size>
+  </Contents>
+</ListBucketResult>`
+	)
+
+	srv := httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, _ *http.Request) {
+			_, _ = io.WriteString(w, truncatedPage)
+		}))
+	t.Cleanup(srv.Close)
+
+	client, err := s3.NewClient(context.Background(), s3.Config{
+		Endpoint:        srv.URL,
+		Bucket:          s3TestBucket,
+		AccessKeyID:     s3TestAccessKeyID,
+		SecretAccessKey: s3TestSecretAccessKey,
+		Region:          s3TestRegion,
+	})
+	if err != nil {
+		t.Fatalf("new client: %v", err)
+	}
+
+	storer := storage.NewS3Storer(client)
+	ctx := context.Background()
+
+	keys, err := storer.List(ctx, prefix)
+	if !errors.Is(err, s3.ErrNoContinuationToken) {
+		t.Errorf("List = %q, %v; want ErrNoContinuationToken", keys, err)
+	}
+
+	var streamErr error
+
+	for obj := range storer.ListStream(ctx, prefix) {
+		if obj.Err != nil {
+			streamErr = obj.Err
+		}
+	}
+
+	if !errors.Is(streamErr, s3.ErrNoContinuationToken) {
+		t.Errorf("ListStream error = %v, want ErrNoContinuationToken", streamErr)
+	}
+}
+
 // TestS3URLPrefixKeyLayout pins the bucket keys an s3:// URL reads and
 // writes: the README's remote storage layout, with the prefix joined to
 // each key by one "/". s3://b/p and s3://b/p/ must be the same
@@ -128,8 +204,8 @@ func TestS3URLPrefixKeyLayout(t *testing.T) {
 			storer, err := storage.NewStorer(&config.Config{
 				StorageURL: storageURL + "?endpoint=" + srv.URL,
 				S3: config.S3Config{
-					AccessKeyID:     "key",
+					AccessKeyID:     s3TestAccessKeyID,
-					SecretAccessKey: "secret",
+					SecretAccessKey: s3TestSecretAccessKey,
 				},
 			})
 			if err != nil {
@@ -171,6 +247,90 @@ func TestS3URLPrefixKeyLayout(t *testing.T) {
 	}
 }
 
+// TestS3UploadUsesConfiguredPartSize checks that s3.part_size reaches the
+// multipart uploader, through storage_url and through the s3.* fields. An
+// object three parts long must arrive as three parts; at the SDK's default
+// of 5 MiB it would arrive as four.
+func TestS3UploadUsesConfiguredPartSize(t *testing.T) {
+	t.Parallel()
+
+	const (
+		partSize  = 6 * 1024 * 1024
+		wantParts = 3
+	)
+
+	backend := s3mem.New()
+
+	err := backend.CreateBucket(s3TestBucket)
+	if err != nil {
+		t.Fatalf("create bucket: %v", err)
+	}
+
+	// Every part of a multipart upload is one request with a partNumber.
+	var parts atomic.Int32
+
+	fake := gofakes3.New(backend).Server()
+	srv := httptest.NewServer(http.HandlerFunc(
+		func(w http.ResponseWriter, r *http.Request) {
+			if r.URL.Query().Has("partNumber") {
+				parts.Add(1)
+			}
+
+			fake.ServeHTTP(w, r)
+		}))
+	t.Cleanup(srv.Close)
+
+	cases := []struct {
+		name string
+		cfg  *config.Config
+	}{
+		{
+			name: "storage_url",
+			cfg: &config.Config{
+				StorageURL: "s3://" + s3TestBucket + "?endpoint=" + srv.URL,
+				S3: config.S3Config{
+					AccessKeyID:     s3TestAccessKeyID,
+					SecretAccessKey: s3TestSecretAccessKey,
+					PartSize:        partSize,
+				},
+			},
+		},
+		{
+			name: "s3.endpoint",
+			cfg: &config.Config{
+				S3: config.S3Config{
+					Endpoint:        srv.URL,
+					Bucket:          s3TestBucket,
+					AccessKeyID:     s3TestAccessKeyID,
+					SecretAccessKey: s3TestSecretAccessKey,
+					PartSize:        partSize,
+				},
+			},
+		},
+	}
+
+	for _, tc := range cases {
+		parts.Store(0)
+
+		storer, err := storage.NewStorer(tc.cfg)
+		if err != nil {
+			t.Fatalf("%s: NewStorer: %v", tc.name, err)
+		}
+
+		data := bytes.NewReader(make([]byte, wantParts*partSize))
+
+		err = storer.PutWithProgress(
+			context.Background(), "blob", data, data.Size(), nil)
+		if err != nil {
+			t.Fatalf("%s: PutWithProgress: %v", tc.name, err)
+		}
+
+		if got := parts.Load(); got != wantParts {
+			t.Errorf("%s: uploaded in %d parts, want %d", tc.name, got, wantParts)
+		}
+	}
+}
+
 // listStreamKeys returns the keys ListStream yields under a prefix, and
 // fails the test on a listing error.
 func listStreamKeys(t *testing.T, s storage.Storer, prefix string) []string {

internal/storage/storer.go

@@ -41,7 +41,8 @@ type Info struct {
 // directory for filesystem).
 type Storer interface {
 	// Put stores data at the specified key.
-	// Parent directories are created automatically for filesystem backends.
+	// The filesystem backend fails when its destination directory does
+	// not exist, and creates the directories below it as needed.
 	Put(ctx context.Context, key string, data io.Reader) error
 
 	// PutWithProgress stores data with progress reporting.
@@ -71,6 +72,12 @@ type Storer interface {
 	// If an error occurs during listing, the final item will have Err set.
 	ListStream(ctx context.Context, prefix string) <-chan ObjectInfo
 
+	// DeletePartialUploads removes every object under prefix that an
+	// upload cut off part-way left under a temporary name ending in
+	// `.partial`. The file and rclone backends' List and ListStream skip
+	// such an object.
+	DeletePartialUploads(ctx context.Context, prefix string) error
+
 	// Info returns human-readable storage location information.
 	Info() Info
 }

internal/storage/url.go

@@ -17,8 +17,10 @@ const (
 
 // Sentinel errors for storage URL parsing.
 var (
-	ErrEmptyStorageURL   = errors.New("storage URL is empty")
+	ErrEmptyStorageURL  = errors.New("storage URL is empty")
-	ErrEmptyFilePath     = errors.New("file URL path is empty")
+	ErrEmptyFilePath    = errors.New("file URL path is empty")
+	ErrRelativeFilePath = errors.New(
+		"file URL needs an absolute path, as in file:///path/to/backup")
 	ErrMissingBucket     = errors.New("s3 URL missing bucket name")
 	ErrMissingRemote     = errors.New("rclone URL missing remote name")
 	ErrUnsupportedScheme = errors.New(
@@ -51,17 +53,8 @@ func ParseStorageURL(rawURL string) (*URL, error) {
 		return nil, ErrEmptyStorageURL
 	}
 
-	// Handle file:// URLs
 	if after, ok := strings.CutPrefix(rawURL, "file://"); ok {
-		path := after
+		return parseFileURL(after)
-		if path == "" {
-			return nil, ErrEmptyFilePath
-		}
-
-		return &URL{
-			Scheme: schemeFile,
-			Prefix: path,
-		}, nil
 	}
 
 	if strings.HasPrefix(rawURL, "s3://") {
@@ -75,6 +68,27 @@ func ParseStorageURL(rawURL string) (*URL, error) {
 	return nil, ErrUnsupportedScheme
 }
 
+// parseFileURL parses the path that follows file://. The path must be
+// absolute: a relative one would resolve against the working directory,
+// so the destination would move with it. Percent-escapes are decoded, so
+// my%20stick names the directory "my stick"; a literal space is kept.
+func parseFileURL(rawPath string) (*URL, error) {
+	if rawPath == "" {
+		return nil, ErrEmptyFilePath
+	}
+
+	if !strings.HasPrefix(rawPath, "/") {
+		return nil, ErrRelativeFilePath
+	}
+
+	path, err := url.PathUnescape(rawPath)
+	if err != nil {
+		return nil, fmt.Errorf("invalid file URL path: %w", err)
+	}
+
+	return &URL{Scheme: schemeFile, Prefix: path}, nil
+}
+
 // parseS3URL parses an s3://bucket/prefix URL. It rejects credentials in
 // the userinfo and any query parameter other than endpoint, region and
 // ssl, so a credential-bearing URL is never stored or echoed.

internal/storage/url_parse_test.go

@@ -14,7 +14,10 @@ import (
 func TestParseStorageURLValid(t *testing.T) {
 	t.Parallel()
 
-	const bucket = "mybucket"
+	const (
+		bucket     = "mybucket"
+		fileScheme = "file"
+	)
 
 	cases := []struct {
 		name string
@@ -24,7 +27,17 @@ func TestParseStorageURLValid(t *testing.T) {
 		{
 			name: "file absolute path",
 			raw:  "file:///var/backups/vaultik",
-			want: &storage.URL{Scheme: "file", Prefix: "/var/backups/vaultik"},
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/var/backups/vaultik"},
+		},
+		{
+			name: "file percent-escaped path is decoded",
+			raw:  "file:///Volumes/my%20stick/backup",
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/Volumes/my stick/backup"},
+		},
+		{
+			name: "file path with a literal space",
+			raw:  "file:///Volumes/my stick/backup",
+			want: &storage.URL{Scheme: fileScheme, Prefix: "/Volumes/my stick/backup"},
 		},
 		{
 			name: "s3 bucket and prefix, ssl defaults on",
@@ -90,6 +103,7 @@ func TestParseStorageURLErrors(t *testing.T) {
 	}{
 		{"empty url", "", storage.ErrEmptyStorageURL},
 		{"file empty path", "file://", storage.ErrEmptyFilePath},
+		{"file relative path", "file://dest", storage.ErrRelativeFilePath},
 		{"s3 missing bucket", "s3://", storage.ErrMissingBucket},
 		{"s3 missing bucket with path", "s3:///justprefix", storage.ErrMissingBucket},
 		{"rclone missing remote", "rclone://", storage.ErrMissingRemote},

internal/types/types.go

@@ -155,8 +155,8 @@ type BlobHash string
 // FilePath represents an absolute path to a file or directory.
 type FilePath string
 
-// SourcePath represents the root directory from which files are backed up.
+// SourcePath is the source directory a scan found a file under, made
-// Used during restore to strip the source prefix from paths.
+// absolute and with symlinks resolved.
 type SourcePath string
 
 // Hostname identifies a host machine.

internal/vaultik/blob_fetch.go

@@ -103,6 +103,8 @@ func (v *Vaultik) FetchAndDecryptBlob(
 }
 
 // FetchBlob downloads a blob and returns a reader for the encrypted data.
+// The reader is closed as soon as ctx is cancelled, which ends a read
+// blocked inside it.
 func (v *Vaultik) FetchBlob(
 	ctx context.Context, blobHash string,
 ) (io.ReadCloser, error) {
@@ -120,5 +122,47 @@ func (v *Vaultik) FetchBlob(
 		return nil, fmt.Errorf("downloading blob %s: %w", shortHash(blobHash), err)
 	}
 
-	return rc, nil
+	return newCloseOnCancelReader(ctx, rc), nil
+}
+
+// closeOnCancelReader closes the reader it wraps when ctx is cancelled.
+// Not every backend's reader stops on cancel by itself (the file
+// backend's does not), and a read stalled on a slow or hung source would
+// otherwise outlast the shutdown timeout, so the process would exit
+// before restore or deep verify removed its decrypted temporary files.
+type closeOnCancelReader struct {
+	io.ReadCloser
+
+	ctx  context.Context //nolint:containedctx // Read reports its error
+	stop func() bool
+}
+
+func newCloseOnCancelReader(
+	ctx context.Context, rc io.ReadCloser,
+) *closeOnCancelReader {
+	return &closeOnCancelReader{
+		ReadCloser: rc,
+		ctx:        ctx,
+		stop:       context.AfterFunc(ctx, func() { _ = rc.Close() }),
+	}
+}
+
+// Read returns the context's error, not the closed reader's, once ctx
+// has been cancelled.
+func (r *closeOnCancelReader) Read(p []byte) (int, error) {
+	n, err := r.ReadCloser.Read(p)
+	if err != nil && r.ctx.Err() != nil {
+		return n, r.ctx.Err()
+	}
+
+	return n, err
+}
+
+// Close closes the wrapped reader, unless the cancel already closed it.
+func (r *closeOnCancelReader) Close() error {
+	if !r.stop() {
+		return nil
+	}
+
+	return r.ReadCloser.Close()
 }

internal/vaultik/changed_file_restore_test.go

@@ -130,15 +130,14 @@ func assertThirdSnapshotRestores(
 // up, and that snapshot is removed. The first snapshot keeps the file row,
 // which now lists the appended content's chunks, while removal drops the
 // blob that held them.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestBackupAfterRemovingNewestSnapshotRestoresChangedFile(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 	changedPath := filepath.Join(dataDir, "appended.bin")
@@ -178,15 +177,14 @@ func TestBackupAfterRemovingNewestSnapshotRestoresChangedFile(t *testing.T) {
 // The next run's prune drops that incomplete snapshot and its blob, while
 // the first snapshot keeps the file row, which now lists the appended
 // content's chunks.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestBackupAfterInterruptedRunRestoresChangedFile(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 	changedPath := filepath.Join(dataDir, "appended.bin")

internal/vaultik/deep_verify_test.go

@@ -30,7 +30,7 @@ func TestDeepVerifyAcceptsHealthyAndRejectsCorruptBlob(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	chunkSize := int64(64 * 1024)

internal/vaultik/destination_marker.go

@@ -0,0 +1,74 @@
+package vaultik
+
+import (
+	"errors"
+	"fmt"
+	"strings"
+
+	"sneak.berlin/go/vaultik/internal/storage"
+)
+
+// destinationMarkerKey is the object every backup writes at the root of
+// the backup destination store. While the local index holds a snapshot,
+// a destination without it is treated as not present: an unplugged
+// volume leaves an empty directory at its mount point, which would
+// otherwise read as a store whose snapshots are all gone.
+const destinationMarkerKey = "vaultik-destination.txt"
+
+// destinationMarkerText is the whole content of the marker. It names no
+// host, path or time.
+const destinationMarkerText = "This is a vaultik backup destination store.\n"
+
+// errDestinationNotPresent is returned when the local index holds a
+// snapshot and the backup destination store has no marker.
+var errDestinationNotPresent = errors.New("backup destination store is not present")
+
+// checkDestinationMarker returns errDestinationNotPresent when the local
+// index holds a snapshot and the backup destination store has no marker.
+// With no snapshot in the local index, as on a first backup or on
+// another machine, there is nothing to check.
+func (v *Vaultik) checkDestinationMarker() error {
+	ids, err := v.localSnapshotIDs()
+	if err != nil {
+		return fmt.Errorf("listing local snapshots: %w", err)
+	}
+
+	if len(ids) == 0 {
+		return nil
+	}
+
+	_, err = v.Storage.Stat(v.ctx, destinationMarkerKey)
+	if errors.Is(err, storage.ErrNotFound) {
+		return fmt.Errorf("%w: %s has no %s, but the local index holds "+
+			"snapshots stored there; mount the volume if it is removable, "+
+			"or if the destination was emptied on purpose, run "+
+			"'vaultik database delete' and back up again",
+			errDestinationNotPresent, v.Storage.Info().Location,
+			destinationMarkerKey)
+	}
+
+	if err != nil {
+		return fmt.Errorf("checking for %s: %w", destinationMarkerKey, err)
+	}
+
+	return nil
+}
+
+// markDestination checks the backup destination store with
+// checkDestinationMarker and then writes the marker, so a first backup
+// marks its destination and a later one writes the same text again.
+func (v *Vaultik) markDestination() error {
+	err := v.checkDestinationMarker()
+	if err != nil {
+		return err
+	}
+
+	err = v.Storage.Put(v.ctx, destinationMarkerKey,
+		strings.NewReader(destinationMarkerText))
+	if err != nil {
+		return fmt.Errorf("writing %s to the backup destination store: %w",
+			destinationMarkerKey, err)
+	}
+
+	return nil
+}

internal/vaultik/destination_marker_test.go

@@ -0,0 +1,184 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// These tests cover https://git.eeqj.de/sneak/vaultik/issues/239: a
+// file:// destination at a volume's mount point leaves an empty directory
+// behind when the volume is unplugged. While the local index holds
+// snapshots, a destination without vaultik-destination.txt is not present,
+// so no command may write to it or conclude from it that the local
+// snapshots are gone.
+
+// markerFile is the name of the file a backup writes at the root of the
+// destination.
+const markerFile = "vaultik-destination.txt"
+
+// backUpThenUnplugMountPoint backs up to a file:// destination, then
+// moves the destination directory away and leaves an empty directory in
+// its place, as unplugging a volume mounted there would. It returns the
+// path of that empty directory.
+func backUpThenUnplugMountPoint(
+	ctx context.Context, t *testing.T,
+) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer, string) {
+	t.Helper()
+
+	storeDir := filepath.Join(t.TempDir(), "mnt")
+	v, repos, out := backUpToFileDestination(ctx, t, storeDir)
+
+	require.NoError(t, os.Rename(storeDir, storeDir+"-unplugged"))
+	require.NoError(t, os.Mkdir(storeDir, 0o750))
+
+	return v, repos, out, storeDir
+}
+
+// assertUntouched checks that the local index still holds its one
+// snapshot and that nothing was written to the empty mount point.
+func assertUntouched(
+	ctx context.Context, t *testing.T,
+	repos *database.Repositories, storeDir string,
+) {
+	t.Helper()
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Len(t, snapshots, 1, "no local snapshot record may be deleted")
+
+	entries, err := os.ReadDir(storeDir)
+	require.NoError(t, err)
+	assert.Empty(t, entries, "nothing may be written to the empty mount point")
+}
+
+// TestFirstBackupWritesMarker checks that a backup writes the marker
+// file at the root of the destination.
+func TestFirstBackupWritesMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	storeDir := t.TempDir()
+	backUpToFileDestination(context.Background(), t, storeDir)
+
+	assert.FileExists(t, filepath.Join(storeDir, markerFile))
+}
+
+// TestBackupRefusesDestinationWithoutMarker checks that snapshot create
+// --prune writes nothing and drops no local snapshot record when the
+// destination has no marker.
+func TestBackupRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		Cron:  true,
+		Prune: true,
+	})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestPruneRefusesDestinationWithoutMarker checks that prune fails and
+// deletes no local snapshot record when the destination has no marker.
+func TestPruneRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.Prune(&vaultik.PruneOptions{Force: true})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestPurgeRefusesDestinationWithoutMarker checks that snapshot purge
+// fails and deletes no local snapshot record when the destination has no
+// marker.
+func TestPurgeRefusesDestinationWithoutMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, _, storeDir := backUpThenUnplugMountPoint(ctx, t)
+
+	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+		Force:      true,
+	})
+	require.ErrorContains(t, err, markerFile)
+
+	assertUntouched(ctx, t, repos, storeDir)
+}
+
+// TestRemoveSnapshotWarnsWhenDestinationHasNoMarker checks that snapshot
+// remove warns that the metadata could not be removed from a destination
+// without a marker, instead of reporting that it was.
+func TestRemoveSnapshotWarnsWhenDestinationHasNoMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out, _ := backUpThenUnplugMountPoint(ctx, t)
+
+	result, err := v.RemoveSnapshot(localSnapshotID(ctx, t, repos, "first"),
+		&vaultik.RemoveOptions{Force: true})
+	require.NoError(t, err)
+
+	assert.False(t, result.RemoteRemoved)
+	assert.Contains(t, out.String(),
+		"Could not remove snapshot metadata from remote")
+	assert.Contains(t, out.String(), markerFile)
+	assert.NotContains(t, out.String(),
+		"Removed snapshot metadata from remote storage")
+}
+
+// TestListSnapshotsWarnsWhenDestinationHasNoMarker checks that snapshot
+// list warns and shows the local index alone when the destination has no
+// marker, without reporting the local snapshot as missing from it.
+func TestListSnapshotsWarnsWhenDestinationHasNoMarker(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, repos, out, _ := backUpThenUnplugMountPoint(ctx, t)
+	id := localSnapshotID(ctx, t, repos, "first")
+
+	require.NoError(t, v.ListSnapshots(false))
+
+	assert.Contains(t, out.String(), "Could not list backup destination store")
+	assert.Contains(t, out.String(), markerFile)
+	assert.Contains(t, out.String(), id)
+	assert.NotContains(t, out.String(), "not found in backup destination store")
+}
+
+// TestBackupAfterNukeRemoteSucceeds checks that remote nuke leaves the
+// destination ready for a new first backup, which writes the marker
+// again.
+func TestBackupAfterNukeRemoteSucceeds(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	storeDir := t.TempDir()
+	v, _, _ := backUpToFileDestination(context.Background(), t, storeDir)
+
+	require.NoError(t, v.NukeRemote(true))
+	assert.NoFileExists(t, filepath.Join(storeDir, markerFile))
+
+	require.NoError(t, backUp(v, "second"))
+	assert.FileExists(t, filepath.Join(storeDir, markerFile))
+}

internal/vaultik/destination_validation_test.go

@@ -161,13 +161,14 @@ func TestVerifyManifestBlobsRejectsShortHash(t *testing.T) {
 		Blobs: []snapshot.BlobInfo{{Hash: "abc", CompressedSize: 1}},
 	}
 
-	verified, missing, mismatched, missingSize, err :=
+	verified, missing, mismatched, missingSize, unchecked, err :=
 		v.verifyManifestBlobs(manifest, &VerifyOptions{JSON: true})
 	require.ErrorIs(t, err, errInvalidBlobHash)
 	require.Zero(t, verified)
 	require.Zero(t, missing)
 	require.Zero(t, mismatched)
 	require.Zero(t, missingSize)
+	require.Empty(t, unchecked)
 }
 
 // TestVerifyBlobChunksRejectsNegativeLength proves a blob_chunks row with a
@@ -257,6 +258,10 @@ func (s *stubLister) List(_ context.Context, _ string) ([]string, error) {
 	return nil, errStubUnused
 }
 
+func (s *stubLister) DeletePartialUploads(_ context.Context, _ string) error {
+	return errStubUnused
+}
+
 func (s *stubLister) Info() storage.Info {
 	return storage.Info{}
 }

internal/vaultik/fault_injection_test.go

@@ -38,11 +38,9 @@ import (
 // (https://git.eeqj.de/sneak/vaultik/issues/130) and is not re-tested
 // here; these tests target the layers above the backend.
 //
-// The tests run serially, not with t.Parallel: each calls
+// log.Initialize replaces the package-global logger that a running
-// log.Initialize, which replaces the package-global logger, and a
+// backup or restore reads, so each test calls it before t.Parallel,
-// backup or restore running concurrently reads that same logger. Under
+// while no parallel test is running yet.
-// -race the two collide. Running one at a time is the same choice
-// prune_count_test.go already makes for the same reason.
 
 const (
 	faultChunkSize   = int64(64 * 1024)
@@ -165,17 +163,19 @@ func newReaderVaultik(
 // Scenario 3: a stored blob's bytes are flipped before restore reads
 // them. Restore must fail loudly, and no file must be left on the
 // restore target holding corrupt content.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestRestoreRejectsCorruptBlob(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
 	assertRestoreRejectsDamagedBlob(t, faultstore.GetCorrupt, "corrupt")
 }
 
 // Scenario 4: a stored blob is truncated before restore reads it. Same
 // contract as the corrupt case.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestRestoreRejectsTruncatedBlob(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
 	assertRestoreRejectsDamagedBlob(t, faultstore.GetTruncate, "truncated")
 }
 
@@ -188,12 +188,11 @@ func assertRestoreRejectsDamagedBlob(
 	t *testing.T, fault faultstore.GetFault, name string,
 ) {
 	t.Helper()
-	log.Initialize(log.Config{})
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -232,15 +231,14 @@ func assertRestoreRejectsDamagedBlob(
 
 // Scenario 6: the backend accepts blob uploads and reports success but
 // stores nothing. verify --deep must catch it.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestDeepVerifyCatchesLyingBackend(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -285,15 +283,14 @@ func TestDeepVerifyCatchesLyingBackend(t *testing.T) {
 // Scenario 1a: a blob upload fails partway through. The interrupted run
 // must not record the blob as uploaded, must not reference it from the
 // snapshot, and must leave no blob object at the destination.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestInterruptedBlobUploadRecordsNoUploadedBlob(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -301,10 +298,6 @@ func TestInterruptedBlobUploadRecordsNoUploadedBlob(t *testing.T) {
 
 	writeFaultSourceTree(t, fs, dataDir)
 
-	// No upload succeeds, so nothing creates the destination directory.
-	// It must exist for the listing below to show that no blob survived.
-	require.NoError(t, os.Mkdir(storeDir, 0o750))
-
 	inner, err := storage.NewFileStorer(storeDir)
 	require.NoError(t, err)
 
@@ -361,15 +354,14 @@ func TestInterruptedBlobUploadRecordsNoUploadedBlob(t *testing.T) {
 // chunks in a blob that was actually uploaded, so the retry re-chunks and
 // re-uploads the affected data instead of silently referencing data that
 // never reached storage.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestBackupRetryAfterInterruptedUploadIsRestorable(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -426,15 +418,14 @@ func TestBackupRetryAfterInterruptedUploadIsRestorable(t *testing.T) {
 // covered by TestBackupCompletesOnlyAfterMetadataExport
 // (https://git.eeqj.de/sneak/vaultik/issues/177); this test exercises the
 // lower-level export path in isolation.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestBackupSurvivesMetadataExportInterruption(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
 	ctx := context.Background()
@@ -509,15 +500,14 @@ func TestBackupSurvivesMetadataExportInterruption(t *testing.T) {
 // destination. Rerunning the backup must then prune the incomplete
 // snapshot, produce a snapshot whose destination metadata and local index
 // agree, and restore. See https://git.eeqj.de/sneak/vaultik/issues/177.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestBackupCompletesOnlyAfterMetadataExport(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -685,15 +675,14 @@ func faultScannerFactory(
 // Scenario 5: the restore target runs out of space mid-file. Restore
 // must fail with an out-of-space error, and must not leave a truncated
 // file at the target path presenting as a complete restore.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestRestoreReportsDiskFull(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	osFS := afero.NewOsFs()
 	tempDir := t.TempDir()
 	dataDir := filepath.Join(tempDir, "src")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/helpers.go

@@ -3,12 +3,14 @@ package vaultik
 import (
 	"errors"
 	"fmt"
+	"math"
 	"regexp"
 	"strconv"
 	"strings"
 	"time"
 
 	"github.com/dustin/go-humanize"
+	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/types"
 )
 
@@ -46,12 +48,13 @@ const (
 	year  = 365 * day
 )
 
-// Snapshot IDs split on "_" into hostname, optional name parts, and a
+// maxDuration is the largest value a time.Duration can hold, about 292
-// trailing timestamp.
+// years.
-const (
+const maxDuration = time.Duration(math.MaxInt64)
-	minSnapshotIDParts     = 2
+
-	minSnapshotIDNameParts = 3
+// A snapshot ID split on "_" has at least a hostname and a trailing
-)
+// timestamp.
+const minSnapshotIDParts = 2
 
 // SnapshotInfo contains information about a snapshot.
 //
@@ -86,22 +89,6 @@ type SnapshotInfo struct {
 	RemotePresent    *bool            `json:"remote_present"`
 }
 
-// formatBytes formats bytes in a human-readable format
-func formatBytes(bytes int64) string {
-	const unit = 1024
-	if bytes < unit {
-		return fmt.Sprintf("%d B", bytes)
-	}
-
-	div, exp := int64(unit), 0
-	for n := bytes / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(bytes)/float64(div), "KMGTPE"[exp])
-}
-
 // parseSnapshotTimestamp extracts the timestamp from a snapshot ID
 // Format: hostname_snapshotname_2026-01-12T14:41:15Z
 func parseSnapshotTimestamp(snapshotID string) (time.Time, error) {
@@ -121,29 +108,36 @@ func parseSnapshotTimestamp(snapshotID string) (time.Time, error) {
 	return timestamp.UTC(), nil
 }
 
-// parseSnapshotName extracts the snapshot name from a snapshot ID.
+// parseSnapshotName extracts the snapshot name from a snapshot ID of the
-// Format: hostname_snapshotname_timestamp — the middle part(s) between hostname
+// form hostname_name_timestamp, given the hostname stored with that
-// and the RFC3339 timestamp are the snapshot name (may contain underscores).
+// snapshot. The hostname and the name may both contain underscores, so the
-// Returns the snapshot name, or empty string if the ID is malformed.
+// name is what is left after removing the short hostname and its "_" from
-func parseSnapshotName(snapshotID string) string {
+// the front and the last "_" and the timestamp from the end. Returns "" for
-	parts := strings.Split(snapshotID, "_")
+// an ID with no name (hostname_timestamp), and for an ID that does not start
-	if len(parts) < minSnapshotIDNameParts {
+// with that hostname, which CreateSnapshotWithName never writes.
-		// Format: hostname_timestamp — no snapshot name
+func parseSnapshotName(snapshotID, hostname string) string {
+	prefix := snapshot.ShortHostname(hostname) + "_"
+
+	rest, ok := strings.CutPrefix(snapshotID, prefix)
+	if !ok {
 		return ""
 	}
-	// Format: hostname_name_timestamp — middle parts are the name.
+
-	// The last part is the RFC3339 timestamp, the first part is the hostname,
+	end := strings.LastIndex(rest, "_")
-	// everything in between is the snapshot name (which may itself contain underscores).
+	if end < 0 {
-	return strings.Join(parts[1:len(parts)-1], "_")
+		return ""
+	}
+
+	return rest[:end]
 }
 
 // parseDuration parses a duration string with support for human-friendly units:
 // d/day/days, w/week/weeks, mo/month/months, y/year/years, plus standard Go
 // duration units. Following Go, m is minutes and mo is months. A bare number,
-// an unknown unit, and a negative value are all rejected. Outside the Go
+// an unknown unit, a negative value, and a total longer than maxDuration are
-// units the input must be whole numbers each followed directly by a unit
+// all rejected. Outside the Go units the input must be whole numbers each
-// (2w3d), so 1.0y and 30 days are errors; decimals work only in Go units
+// followed directly by a unit (2w3d), so 1.0y and 30 days are errors;
-// (1.5h).
+// decimals work only in Go units (1.5h).
 func parseDuration(s string) (time.Duration, error) {
 	if strings.HasPrefix(strings.TrimSpace(s), "-") {
 		return 0, errNegativeDuration
@@ -176,19 +170,30 @@ func parseDuration(s string) (time.Duration, error) {
 			return 0, fmt.Errorf("invalid number %q: %w", match[1], err)
 		}
 
+		var unitLength time.Duration
+
 		unit := strings.ToLower(match[2])
 		switch unit {
 		case "d", "day", "days":
-			total += time.Duration(n) * day
+			unitLength = day
 		case "w", "week", "weeks":
-			total += time.Duration(n) * week
+			unitLength = week
 		case "mo", "month", "months":
-			total += time.Duration(n) * month
+			unitLength = month
 		case "y", "year", "years":
-			total += time.Duration(n) * year
+			unitLength = year
 		default:
 			return 0, fmt.Errorf("%w %q", errUnknownTimeUnit, unit)
 		}
+
+		// Past maxDuration the sum wraps to a negative value, which would
+		// put the purge cutoff in the future.
+		if time.Duration(n) > (maxDuration-total)/unitLength {
+			return 0, fmt.Errorf("%w: %q is longer than about 292 years",
+				errInvalidDuration, s)
+		}
+
+		total += time.Duration(n) * unitLength
 	}
 
 	return total, nil

internal/vaultik/helpers_test.go

@@ -11,33 +11,55 @@ func TestParseSnapshotName(t *testing.T) {
 	tests := []struct {
 		name       string
 		snapshotID string
+		hostname   string
 		want       string
 	}{
 		{
 			name:       "standard format with name",
 			snapshotID: "myhost_home_2026-01-12T14:41:15Z",
+			hostname:   "myhost",
 			want:       "home",
 		},
 		{
 			name:       "standard format with different name",
 			snapshotID: "server1_system_2026-02-15T09:30:00Z",
+			hostname:   "server1",
 			want:       "system",
 		},
 		{
 			name:       "name with underscores",
 			snapshotID: "myhost_my_special_backup_2026-03-01T00:00:00Z",
+			hostname:   "myhost",
 			want:       "my_special_backup",
 		},
+		{
+			name:       "hostname with underscores",
+			snapshotID: "my_host_docs_2026-03-01T00:00:00Z",
+			hostname:   "my_host",
+			want:       "docs",
+		},
+		{
+			name:       "stored hostname with domain",
+			snapshotID: "my_host_mail_2026-03-01T00:00:00Z",
+			hostname:   "my_host.example.com",
+			want:       "mail",
+		},
+		{
+			name:       "no name",
+			snapshotID: "my_host_2026-03-01T00:00:00Z",
+			hostname:   "my_host",
+			want:       "",
+		},
 	}
 
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {
 			t.Parallel()
 
-			got := parseSnapshotName(tt.snapshotID)
+			got := parseSnapshotName(tt.snapshotID, tt.hostname)
 			if got != tt.want {
-				t.Errorf("parseSnapshotName(%q) = %q, want %q",
+				t.Errorf("parseSnapshotName(%q, %q) = %q, want %q",
-					tt.snapshotID, got, tt.want)
+					tt.snapshotID, tt.hostname, got, tt.want)
 			}
 		})
 	}
@@ -89,6 +111,17 @@ func TestParseDuration(t *testing.T) {
 		{"30 days", 0, true},
 		{"30 days ago", 0, true},
 		{"x7d", 0, true},
+		// The largest time.Duration is about 292 years. A larger value
+		// must be rejected, not wrapped to a negative one, which would
+		// put the cutoff in the future and select every snapshot.
+		{"292y", 292 * 365 * 24 * time.Hour, false},
+		{"293y", 0, true},
+		{"3559mo", 0, true},
+		{"15251w", 0, true},
+		{"106752d", 0, true},
+		{"2562048h", 0, true},
+		{"200y200y", 0, true},
+		{"292y6mo", 0, true},
 	}
 
 	for _, tt := range tests {

internal/vaultik/info.go

@@ -167,10 +167,6 @@ const (
 	// metadataKeyParts is the minimum "/"-separated segment count of a
 	// metadata object key (metadata/<snapshot-id>/<filename>).
 	metadataKeyParts = 3
-
-	// blobKeyParts is the minimum "/"-separated segment count of a blob
-	// object key (blobs/<aa>/<bb>/<hash>).
-	blobKeyParts = 4
 )
 
 // SnapshotMetadataInfo contains information about a single snapshot's metadata
@@ -181,8 +177,16 @@ type SnapshotMetadataInfo struct {
 	ManifestSize int64  `json:"manifest_size"`
 	DatabaseSize int64  `json:"database_size"`
 	TotalSize    int64  `json:"total_size"`
-	BlobCount    int    `json:"blob_count"`
+
-	BlobsSize    int64  `json:"blobs_size"`
+	// Both stay nil (null in the JSON) when the snapshot's manifest was
+	// listed but could not be read.
+	BlobCount *int   `json:"blob_count"`
+	BlobsSize *int64 `json:"blobs_size"`
+
+	// Set when the listing holds this snapshot's manifest.json.zst. A
+	// backup interrupted before its manifest upload leaves a directory
+	// without one, which prune does not treat as a snapshot.
+	hasManifest bool
 }
 
 // RemoteInfoResult contains all remote storage information
@@ -206,9 +210,20 @@ type RemoteInfoResult struct {
 	ReferencedBlobCount int   `json:"referenced_blob_count"`
 	ReferencedBlobSize  int64 `json:"referenced_blob_size"`
 
-	// Orphaned blobs
+	// Orphaned blobs. Both stay nil (null in the JSON) when a manifest
-	OrphanedBlobCount int   `json:"orphaned_blob_count"`
+	// was listed but not read, since that snapshot's blobs would be
-	OrphanedBlobSize  int64 `json:"orphaned_blob_size"`
+	// counted as orphaned.
+	OrphanedBlobCount *int   `json:"orphaned_blob_count"`
+	OrphanedBlobSize  *int64 `json:"orphaned_blob_size"`
+
+	// Remote key of each snapshot whose manifest could not be read
+	UnreadableManifests []string `json:"unreadable_manifests,omitempty"`
+
+	// Number of manifests not read because the name above them under
+	// metadata/ is not a remote key. The names themselves are not
+	// reported: they come from the destination store and may hold
+	// control characters.
+	SkippedManifestCount int `json:"skipped_manifest_count,omitempty"`
 }
 
 // RemoteInfo displays information about remote storage
@@ -234,16 +249,28 @@ func (v *Vaultik) RemoteInfo(jsonOutput bool) error {
 		v.stdoutf("Scanning snapshot metadata...\n")
 	}
 
-	snapshotMetadata, snapshotIDs, err := v.collectSnapshotMetadata()
+	snapshotMetadata, snapshotIDs, skippedManifestCount, err := v.collectSnapshotMetadata()
 	if err != nil {
 		return err
 	}
 
+	result.SkippedManifestCount = skippedManifestCount
+
 	if showText {
-		v.stdoutf("Downloading %d manifest(s)...\n", len(snapshotIDs))
+		manifestCount := 0
+
+		for _, info := range snapshotMetadata {
+			if info.hasManifest {
+				manifestCount++
+			}
+		}
+
+		v.stdoutf("Downloading %d manifest(s)...\n", manifestCount)
 	}
 
-	referencedBlobs := v.collectReferencedBlobsFromManifests(snapshotIDs, snapshotMetadata)
+	referencedBlobs, unreadableManifests := v.collectReferencedBlobsFromManifests(
+		snapshotIDs, snapshotMetadata)
+	result.UnreadableManifests = unreadableManifests
 
 	v.populateRemoteInfoResult(result, snapshotMetadata, snapshotIDs, referencedBlobs)
 
@@ -256,7 +283,7 @@ func (v *Vaultik) RemoteInfo(jsonOutput bool) error {
 		"snapshots", result.TotalMetadataCount,
 		"total_blobs", result.TotalBlobCount,
 		"referenced_blobs", result.ReferencedBlobCount,
-		"orphaned_blobs", result.OrphanedBlobCount)
+		"unreadable_manifests", len(result.UnreadableManifests))
 
 	if jsonOutput {
 		enc := json.NewEncoder(v.Stdout)
@@ -273,16 +300,18 @@ func (v *Vaultik) RemoteInfo(jsonOutput bool) error {
 }
 
 // collectSnapshotMetadata scans remote metadata and returns
-// per-snapshot info and sorted IDs.
+// per-snapshot info, sorted IDs and the number of manifests it skipped
+// because the name above them is not a remote key.
 func (v *Vaultik) collectSnapshotMetadata() (
-	map[string]*SnapshotMetadataInfo, []string, error,
+	map[string]*SnapshotMetadataInfo, []string, int, error,
 ) {
 	snapshotMetadata := make(map[string]*SnapshotMetadataInfo)
+	skippedManifestCount := 0
 
 	metadataCh := v.Storage.ListStream(v.ctx, "metadata/")
 	for obj := range metadataCh {
 		if obj.Err != nil {
-			return nil, nil, fmt.Errorf("listing metadata: %w", obj.Err)
+			return nil, nil, 0, fmt.Errorf("listing metadata: %w", obj.Err)
 		}
 
 		parts := strings.Split(obj.Key, "/")
@@ -291,6 +320,22 @@ func (v *Vaultik) collectSnapshotMetadata() (
 		}
 
 		snapshotID := parts[1]
+		filename := parts[2]
+		isManifest := filename == "manifest.json.zst"
+
+		// The name comes from the destination store, which is not
+		// trusted, and is printed in the report. Accept it only in the
+		// form of a remote key.
+		if !isBlobHash(snapshotID) {
+			log.Warn("Skipping non-conforming key under metadata/",
+				"key", obj.Key)
+
+			if isManifest {
+				skippedManifestCount++
+			}
+
+			continue
+		}
 
 		if _, exists := snapshotMetadata[snapshotID]; !exists {
 			snapshotMetadata[snapshotID] = &SnapshotMetadataInfo{SnapshotID: snapshotID}
@@ -298,7 +343,10 @@ func (v *Vaultik) collectSnapshotMetadata() (
 
 		info := snapshotMetadata[snapshotID]
 
-		filename := parts[2]
+		if isManifest {
+			info.hasManifest = true
+		}
+
 		if strings.HasPrefix(filename, "manifest") {
 			info.ManifestSize = obj.Size
 		} else if strings.HasPrefix(filename, "db") {
@@ -315,17 +363,29 @@ func (v *Vaultik) collectSnapshotMetadata() (
 
 	sort.Strings(snapshotIDs)
 
-	return snapshotMetadata, snapshotIDs, nil
+	return snapshotMetadata, snapshotIDs, skippedManifestCount, nil
 }
 
-// collectReferencedBlobsFromManifests downloads manifests and returns
+// collectReferencedBlobsFromManifests downloads the listed manifests
-// referenced blob hashes with sizes.
+// and returns referenced blob hashes with sizes, and the remote keys
+// of the manifests it could not read.
 func (v *Vaultik) collectReferencedBlobsFromManifests(
 	snapshotIDs []string, snapshotMetadata map[string]*SnapshotMetadataInfo,
-) map[string]int64 {
+) (map[string]int64, []string) {
 	referencedBlobs := make(map[string]int64)
 
+	var unreadable []string
+
 	for _, snapshotID := range snapshotIDs {
+		info := snapshotMetadata[snapshotID]
+		if !info.hasManifest {
+			// The orphan figures count this directory's blobs as
+			// orphaned, so it references none.
+			info.BlobCount, info.BlobsSize = new(int), new(int64)
+
+			continue
+		}
+
 		// snapshotIDs here are remote keys, taken straight from the
 		// metadata/ listing. downloadManifestByKey is the single reader
 		// for remote manifests; see its doc comment.
@@ -333,11 +393,12 @@ func (v *Vaultik) collectReferencedBlobsFromManifests(
 		if err != nil {
 			log.Warn("Failed to read manifest", "snapshot", snapshotID, "error", err)
 
+			unreadable = append(unreadable, snapshotID)
+
 			continue
 		}
 
-		info := snapshotMetadata[snapshotID]
+		blobCount := manifest.BlobCount
-		info.BlobCount = manifest.BlobCount
 
 		var blobsSize int64
 
@@ -346,10 +407,11 @@ func (v *Vaultik) collectReferencedBlobsFromManifests(
 			blobsSize += blob.CompressedSize
 		}
 
-		info.BlobsSize = blobsSize
+		info.BlobCount = &blobCount
+		info.BlobsSize = &blobsSize
 	}
 
-	return referencedBlobs
+	return referencedBlobs, unreadable
 }
 
 // populateRemoteInfoResult fills in the result's snapshot and
@@ -378,8 +440,9 @@ func (v *Vaultik) populateRemoteInfoResult(
 }
 
 // scanRemoteBlobStorage lists all blobs on remote and computes orphan
-// stats. showText is true only when the human report is being printed
+// stats when every listed manifest was read. showText is true only
-// (not --json, not --quiet), gating the progress line.
+// when the human report is being printed (not --json, not --quiet),
+// gating the progress line.
 func (v *Vaultik) scanRemoteBlobStorage(
 	result *RemoteInfoResult, referencedBlobs map[string]int64, showText bool,
 ) error {
@@ -387,32 +450,40 @@ func (v *Vaultik) scanRemoteBlobStorage(
 		v.stdoutf("Scanning blobs...\n")
 	}
 
-	blobCh := v.Storage.ListStream(v.ctx, "blobs/")
+	// Use prune's listing, so a file under blobs/ that prune would never
-	allBlobs := make(map[string]int64)
+	// delete is not counted as a blob, and so not as an orphaned one.
-
+	allBlobs, err := v.listAllRemoteBlobs()
-	for obj := range blobCh {
+	if err != nil {
-		if obj.Err != nil {
+		return err
-			return fmt.Errorf("listing blobs: %w", obj.Err)
-		}
-
-		parts := strings.Split(obj.Key, "/")
-		if len(parts) < blobKeyParts {
-			continue
-		}
-
-		hash := parts[3]
-		allBlobs[hash] = obj.Size
-		result.TotalBlobCount++
-		result.TotalBlobSize += obj.Size
 	}
 
+	for _, size := range allBlobs {
+		result.TotalBlobCount++
+		result.TotalBlobSize += size
+	}
+
+	// A blob named only by a manifest that could not be read, or by one
+	// under a skipped name, would be counted as orphaned, so the orphan
+	// figures stay unknown.
+	if len(result.UnreadableManifests) > 0 || result.SkippedManifestCount > 0 {
+		return nil
+	}
+
+	var (
+		orphanedCount int
+		orphanedSize  int64
+	)
+
 	for hash, size := range allBlobs {
 		if _, referenced := referencedBlobs[hash]; !referenced {
-			result.OrphanedBlobCount++
+			orphanedCount++
-			result.OrphanedBlobSize += size
+			orphanedSize += size
 		}
 	}
 
+	result.OrphanedBlobCount = &orphanedCount
+	result.OrphanedBlobSize = &orphanedSize
+
 	return nil
 }
 
@@ -442,13 +513,23 @@ func (v *Vaultik) printRemoteInfoTable(result *RemoteInfoResult) {
 		v.stdoutf("%s", separator)
 
 		for _, info := range result.Snapshots {
+			blobCount := unknownText
+			if info.BlobCount != nil {
+				blobCount = humanize.Comma(int64(*info.BlobCount))
+			}
+
+			blobsSize := unknownText
+			if info.BlobsSize != nil {
+				blobsSize = ubytes(*info.BlobsSize)
+			}
+
 			v.stdoutf(rowFormat,
 				truncateString(info.SnapshotID, snapshotIDColWidth),
 				ubytes(info.ManifestSize),
 				ubytes(info.DatabaseSize),
 				ubytes(info.TotalSize),
-				humanize.Comma(int64(info.BlobCount)),
+				blobCount,
-				ubytes(info.BlobsSize),
+				blobsSize,
 			)
 		}
 
@@ -465,11 +546,21 @@ func (v *Vaultik) printRemoteInfoTable(result *RemoteInfoResult) {
 	v.stdoutf("Referenced by snapshots:    %s (%s)\n",
 		humanize.Comma(int64(result.ReferencedBlobCount)),
 		ubytes(result.ReferencedBlobSize))
-	v.stdoutf("Orphaned (unreferenced):    %s (%s)\n",
-		humanize.Comma(int64(result.OrphanedBlobCount)),
-		ubytes(result.OrphanedBlobSize))
 
-	if result.OrphanedBlobCount > 0 {
+	if result.OrphanedBlobCount == nil {
+		v.stdoutf("Orphaned (unreferenced):    unknown "+
+			"(%d manifest(s) could not be read, "+
+			"%d manifest(s) under a non-conforming name skipped)\n",
+			len(result.UnreadableManifests), result.SkippedManifestCount)
+
+		return
+	}
+
+	v.stdoutf("Orphaned (unreferenced):    %s (%s)\n",
+		humanize.Comma(int64(*result.OrphanedBlobCount)),
+		ubytes(*result.OrphanedBlobSize))
+
+	if *result.OrphanedBlobCount > 0 {
 		v.stdoutf("\nRun 'vaultik prune' to remove orphaned blobs.\n")
 	}
 }

internal/vaultik/integration_test.go

@@ -155,6 +155,12 @@ func (m *MockStorer) ListStream(
 	return ch
 }
 
+// DeletePartialUploads has nothing to remove: Put stores each object
+// under its key at once.
+func (m *MockStorer) DeletePartialUploads(_ context.Context, _ string) error {
+	return nil
+}
+
 func (m *MockStorer) Info() storage.Info {
 	return storage.Info{
 		Type:     "mock",
@@ -765,7 +771,7 @@ func TestEndToEndFileStorage(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -928,17 +934,17 @@ func setupDedupBackupEnv(
 	}
 }
 
-// runDedupSnapshot creates a "dedup" snapshot, scans dataDir into it,
+// runDedupSnapshot creates a snapshot with the given name, scans dataDir
-// completes it, and exports its metadata, returning the snapshot ID and
+// into it, completes it, and exports its metadata, returning the snapshot
-// scan result.
+// ID and scan result.
 func runDedupSnapshot(
 	ctx context.Context, t *testing.T,
 	sm *snapshot.SnapshotManager, scanner *snapshot.Scanner,
-	hostname, dataDir, dbPath string,
+	hostname, name, dataDir, dbPath string,
 ) (string, *snapshot.ScanResult) {
 	t.Helper()
 
-	id, err := sm.CreateSnapshotWithName(ctx, hostname, "dedup", "v", "g")
+	id, err := sm.CreateSnapshotWithName(ctx, hostname, name, "v", "g")
 	require.NoError(t, err)
 
 	result, err := scanner.Scan(ctx, dataDir, id)
@@ -958,7 +964,7 @@ func TestDedupOnlySnapshotRestores(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 
@@ -980,16 +986,15 @@ func TestDedupOnlySnapshotRestores(t *testing.T) {
 
 	// First snapshot — uploads all blobs.
 	_, r1 := runDedupSnapshot(ctx, t, sm, makeScanner(),
-		cfg.Hostname, dataDir, dbPath)
+		cfg.Hostname, "first", dataDir, dbPath)
 	require.Positive(t, r1.BlobsCreated,
 		"first snapshot should upload at least one blob")
 
-	// Second snapshot — same data, every chunk dedups. Sleep past the
+	// Second snapshot — same data, every chunk dedups. Its own name gives
-	// second-precision timestamp so the snapshot IDs differ.
+	// it a different snapshot ID without waiting for the one-second
-	time.Sleep(1100 * time.Millisecond)
+	// timestamp in the ID to tick over.
-
 	id2, r2 := runDedupSnapshot(ctx, t, sm, makeScanner(),
-		cfg.Hostname, dataDir, dbPath)
+		cfg.Hostname, "second", dataDir, dbPath)
 	require.Equal(t, 0, r2.BlobsCreated,
 		"second snapshot should upload zero new blobs (fully dedup'd)")
 

internal/vaultik/missing_destination_test.go

@@ -6,6 +6,7 @@ import (
 	"io/fs"
 	"os"
 	"path/filepath"
+	"strings"
 	"testing"
 
 	"github.com/spf13/afero"
@@ -23,13 +24,11 @@ import (
 // that is not plugged in, cannot be listed. It is not an empty store, so
 // no command may conclude from it that the local snapshots are gone.
 
-// backUpToFileDestination backs up the snapshot named "first" to a
+// newFileDestinationVaultik returns a Vaultik that backs up the
-// file:// destination at storeDir, which need not exist yet. Everything
+// snapshot named "first" to a file:// destination at storeDir.
-// the returned Vaultik prints after the backup goes to the returned
+func newFileDestinationVaultik(
-// buffer.
-func backUpToFileDestination(
 	ctx context.Context, t *testing.T, storeDir string,
-) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer) {
+) (*vaultik.Vaultik, *database.Repositories) {
 	t.Helper()
 
 	osFs := afero.NewOsFs()
@@ -48,8 +47,22 @@ func backUpToFileDestination(
 
 	repos := database.NewRepositories(db)
 	cfg := changedFileConfig(dataDir, dbPath)
-	v := newBackupVaultik(ctx, cfg, store, repos, db, osFs)
 
+	return newBackupVaultik(ctx, cfg, store, repos, db, osFs), repos
+}
+
+// backUpToFileDestination creates the directory storeDir, as a user must
+// before the first backup, and backs up the snapshot named "first" to a
+// file:// destination there. Everything the returned Vaultik prints
+// after the backup goes to the returned buffer.
+func backUpToFileDestination(
+	ctx context.Context, t *testing.T, storeDir string,
+) (*vaultik.Vaultik, *database.Repositories, *bytes.Buffer) {
+	t.Helper()
+
+	require.NoError(t, os.MkdirAll(storeDir, 0o750))
+
+	v, repos := newFileDestinationVaultik(ctx, t, storeDir)
 	require.NoError(t, backUp(v, "first"))
 
 	out := &bytes.Buffer{}
@@ -74,31 +87,36 @@ func backUpThenUnplug(
 	return v, repos, out
 }
 
-// TestFirstBackupCreatesDestinationDirectory checks that a first backup
+// TestBackupFailsWhenDestinationDirectoryMissing checks that a backup to
-// to a destination directory that does not exist yet creates it, and
+// a file:// destination whose directory does not exist fails before it
-// that the destination can be listed afterwards.
+// records a snapshot, and does not create the directory.
-//
+func TestBackupFailsWhenDestinationDirectoryMissing(t *testing.T) {
-//nolint:paralleltest // installs the global logger via log.Initialize
-func TestFirstBackupCreatesDestinationDirectory(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	ctx := context.Background()
-	storeDir := filepath.Join(t.TempDir(), "volume", "backup")
+	volume := filepath.Join(t.TempDir(), "volume")
-	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+	v, repos := newFileDestinationVaultik(ctx, t, filepath.Join(volume, "backup"))
 
-	require.NoError(t, v.ListSnapshots(false))
+	err := backUp(v, "first")
+	require.ErrorIs(t, err, fs.ErrNotExist)
+	require.ErrorContains(t, err, "create it, or mount the volume it is on")
 
-	assert.NotContains(t, out.String(), "Could not list backup destination store")
+	_, err = os.Stat(volume)
-	assert.NotContains(t, out.String(), "not found in backup destination store")
+	require.ErrorIs(t, err, fs.ErrNotExist,
+		"the backup must not create the destination directory")
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	assert.Empty(t, snapshots)
 }
 
 // TestListSnapshotsWarnsWhenDestinationMissing checks that snapshot list
 // warns and shows the local index alone, without reporting the local
 // snapshot as missing from the destination.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestListSnapshotsWarnsWhenDestinationMissing(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	ctx := context.Background()
 	v, repos, out := backUpThenUnplug(ctx, t)
@@ -115,10 +133,9 @@ func TestListSnapshotsWarnsWhenDestinationMissing(t *testing.T) {
 // TestRemoveSnapshotWarnsWhenDestinationMissing checks that snapshot
 // remove warns that the metadata could not be removed from the
 // destination, instead of reporting that it was.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestRemoveSnapshotWarnsWhenDestinationMissing(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	ctx := context.Background()
 	v, repos, out := backUpThenUnplug(ctx, t)
@@ -137,10 +154,9 @@ func TestRemoveSnapshotWarnsWhenDestinationMissing(t *testing.T) {
 // TestPruneKeepsLocalRecordsWhenDestinationMissing checks that prune
 // fails on a destination it cannot list and deletes no local snapshot
 // record.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestPruneKeepsLocalRecordsWhenDestinationMissing(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	ctx := context.Background()
 	v, repos, _ := backUpThenUnplug(ctx, t)
@@ -153,3 +169,22 @@ func TestPruneKeepsLocalRecordsWhenDestinationMissing(t *testing.T) {
 	require.NoError(t, err)
 	assert.Len(t, snapshots, 1, "prune must delete no local snapshot record")
 }
+
+// TestPurgeSaysListingFailedOnceWhenDestinationMissing checks that
+// snapshot purge fails on a destination it cannot list, with an error
+// that says "listing remote snapshots" once.
+func TestPurgeSaysListingFailedOnceWhenDestinationMissing(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	v, _, _ := backUpThenUnplug(ctx, t)
+
+	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+		Force:      true,
+	})
+	require.ErrorIs(t, err, fs.ErrNotExist)
+	assert.Equal(t, 1, strings.Count(err.Error(), "listing remote snapshots"),
+		err.Error())
+}

internal/vaultik/new_chunk_size_test.go

@@ -0,0 +1,133 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"maps"
+	"path/filepath"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/chunker"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestNewChunkSizeOfChangedFileStaysFixed covers
+// https://git.eeqj.de/sneak/vaultik/issues/311. A backup whose only change
+// is a modified file lists the size of the chunks that change added as its
+// NEW CHUNK SIZE, and a later backup that changes the file again leaves
+// every earlier snapshot's figure as it was. The file's row in the local
+// index is shared by all three snapshots and rewritten by each backup.
+func TestNewChunkSizeOfChangedFileStaysFixed(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+	dataDir := filepath.Join(tempDir, "src")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+	changedPath := filepath.Join(dataDir, "changed.bin")
+
+	ctx := context.Background()
+	files := writeFaultSourceTree(t, fs, dataDir)
+	cfg := changedFileConfig(dataDir, dbPath)
+
+	appendRandomBytes(t, fs, files, changedPath, appendedFileSize)
+
+	store, err := storage.NewFileStorer(t.TempDir())
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, dbPath)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+	v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+
+	require.NoError(t, backUp(v, "first"))
+
+	firstFiles := maps.Clone(files)
+	firstID := localSnapshotID(ctx, t, repos, "first")
+
+	appendRandomBytes(t, fs, files, changedPath, faultChunkSize)
+	require.NoError(t, backUp(v, "second"))
+
+	secondID := localSnapshotID(ctx, t, repos, "second")
+	secondAdded := sizeOfNewChunks(t, files, firstFiles)
+	require.Positive(t, secondAdded)
+	assert.Equal(t, secondAdded, listedNewChunkSizes(t, v)[secondID],
+		"the second snapshot must count the chunks the change added")
+
+	appendRandomBytes(t, fs, files, changedPath, faultChunkSize)
+	require.NoError(t, backUp(v, "third"))
+
+	sizes := listedNewChunkSizes(t, v)
+	assert.Equal(t, sizeOfNewChunks(t, firstFiles, nil), sizes[firstID],
+		"a later backup must not change the first snapshot's figure")
+	assert.Equal(t, secondAdded, sizes[secondID],
+		"a later backup must not change the second snapshot's figure")
+}
+
+// sizeOfNewChunks returns the total size of the distinct chunks of the
+// files in after that no file in before has, cut the way the backup cuts
+// them.
+func sizeOfNewChunks(t *testing.T, after, before map[string][]byte) int64 {
+	t.Helper()
+
+	known := make(map[string]bool)
+	for _, chunk := range chunksOf(t, before) {
+		known[chunk.Hash] = true
+	}
+
+	var total int64
+
+	for _, chunk := range chunksOf(t, after) {
+		if !known[chunk.Hash] {
+			known[chunk.Hash] = true
+			total += chunk.Size
+		}
+	}
+
+	return total
+}
+
+// chunksOf returns the chunks of every file in files.
+func chunksOf(t *testing.T, files map[string][]byte) []chunker.Chunk {
+	t.Helper()
+
+	c := chunker.NewChunker(faultChunkSize)
+
+	var all []chunker.Chunk
+
+	for _, content := range files {
+		chunks, err := c.ChunkReader(bytes.NewReader(content))
+		require.NoError(t, err)
+
+		all = append(all, chunks...)
+	}
+
+	return all
+}
+
+// listedNewChunkSizes runs `snapshot list --json` and returns each listed
+// snapshot's new_chunk_size by snapshot ID.
+func listedNewChunkSizes(t *testing.T, v *vaultik.Vaultik) map[string]int64 {
+	t.Helper()
+
+	var out bytes.Buffer
+
+	v.Stdout = &out
+	require.NoError(t, v.ListSnapshots(true))
+
+	sizes := make(map[string]int64)
+	for _, row := range decodeListJSON(t, out.String()) {
+		sizes[row.ID] = row.NewChunkSize
+	}
+
+	return sizes
+}

internal/vaultik/nuke_remote_test.go

@@ -0,0 +1,144 @@
+package vaultik_test
+
+import (
+	"context"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
+)
+
+// TestNukeRemoteLeavesNoFiles checks that remote nuke leaves no file
+// under a file:// destination, including the `.partial` files uploads
+// killed part-way leave next to a snapshot's metadata and next to where a
+// blob would have been.
+func TestNukeRemoteLeavesNoFiles(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, repos, out := backUpToFileDestination(ctx, t, storeDir)
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	require.Len(t, snapshots, 1)
+
+	snapshotKey := snapshot.RemoteSnapshotKey(snapshots[0].ID.String())
+	hash := testBlobHashA
+	leftovers := []string{
+		filepath.Join(storeDir, "metadata", snapshotKey,
+			"db.zst.age-123456.partial"),
+		filepath.Join(storeDir, "blobs", hash[:2], hash[2:4],
+			hash+"-123456.partial"),
+	}
+
+	for _, leftover := range leftovers {
+		require.NoError(t, os.MkdirAll(filepath.Dir(leftover), 0o750))
+		require.NoError(t, os.WriteFile(leftover, []byte("half an upload"), 0o600))
+	}
+
+	require.NoError(t, v.NukeRemote(true))
+
+	var files []string
+
+	err = filepath.WalkDir(storeDir,
+		func(path string, entry fs.DirEntry, err error) error {
+			if err != nil {
+				return err
+			}
+
+			if !entry.IsDir() {
+				files = append(files, path)
+			}
+
+			return nil
+		})
+	require.NoError(t, err)
+	assert.Empty(t, files)
+
+	// Nuke deletes the blobs itself, so advice to run prune is wrong.
+	assert.Contains(t, out.String(), "Backup destination store is now empty.")
+	assert.NotContains(t, out.String(), "vaultik prune")
+}
+
+// TestNukeRemoteFailsWhenADeleteFails checks that remote nuke returns an
+// error when a snapshot's metadata or a blob cannot be deleted, does not
+// report the store empty, and advises running nuke again: prune deletes
+// no metadata.
+func TestNukeRemoteFailsWhenADeleteFails(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for _, refusedPrefix := range []string{"metadata/", "blobs/"} {
+		t.Run(refusedPrefix, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+			storeDir := filepath.Join(t.TempDir(), "store")
+			v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+			store := faultstore.New(v.Storage)
+			store.OnDelete = func(key string) bool {
+				return strings.HasPrefix(key, refusedPrefix)
+			}
+			v.Storage = store
+
+			require.Error(t, v.NukeRemote(true))
+
+			assert.NotContains(t, out.String(), "now empty")
+			assert.Contains(t, out.String(), "vaultik remote nuke --force")
+			assert.NotContains(t, out.String(), "vaultik prune")
+		})
+	}
+}
+
+// TestNukeRemoteRefusesAnotherDestinationWithoutRetryAdvice checks that
+// remote nuke, when the local index belongs to a different destination
+// than the configured one, fails without advising to run nuke again:
+// that run would fail the same way.
+func TestNukeRemoteRefusesAnotherDestinationWithoutRetryAdvice(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+	v.Config.StorageURL = "file://" + filepath.Join(t.TempDir(), "elsewhere")
+
+	err := v.NukeRemote(true)
+
+	require.ErrorContains(t, err, "bound to a different backup destination")
+	assert.NotContains(t, out.String(), "vaultik remote nuke --force")
+}
+
+// TestNukeRemoteFailsWhenALocalSnapshotCannotBeRemoved checks that a
+// snapshot remote nuke cannot remove from the local index fails the
+// run: nuke then leaves that snapshot's metadata on the destination
+// store.
+func TestNukeRemoteFailsWhenALocalSnapshotCannotBeRemoved(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, out := backUpToFileDestination(ctx, t, storeDir)
+
+	_, err := v.DB.Conn().ExecContext(ctx, `
+		CREATE TRIGGER refuse_snapshot_delete BEFORE DELETE ON snapshots
+		BEGIN SELECT RAISE(ABORT, 'simulated index error'); END`)
+	require.NoError(t, err)
+
+	require.Error(t, v.NukeRemote(true))
+
+	assert.NotContains(t, out.String(), "now empty")
+	assert.Contains(t, out.String(), "vaultik remote nuke --force")
+}

internal/vaultik/prune.go

@@ -19,13 +19,27 @@ type PruneOptions struct {
 var errNukeRequiresForce = errors.New(
 	"nuke requires --force (this deletes ALL remote snapshots and blobs)")
 
+// errBlobDeleteFailed reports unreferenced blobs prune could not delete.
+var errBlobDeleteFailed = errors.New("blob(s) could not be deleted")
+
+// nukeCommandHint is the command suggested when a remote nuke could not
+// delete everything: running it again deletes what is left, which
+// `vaultik prune` does not do for snapshot metadata.
+const nukeCommandHint = "vaultik remote nuke --force"
+
 // metadataDirName is the top-level remote directory holding snapshot
 // metadata.
 const metadataDirName = "metadata"
 
+// blobKeyParts is the "/"-separated segment count of a blob object key
+// (blobs/<aa>/<bb>/<hash>).
+const blobKeyParts = 4
+
 // NukeRemote deletes every snapshot's metadata and every blob from remote
-// storage. After this returns successfully the bucket prefix is empty and
+// storage, along with any object an upload cut off part-way left under a
-// the next backup starts from scratch.
+// temporary `.partial` name, and the destination marker. After this
+// returns successfully the bucket prefix is empty and the next backup
+// starts from scratch.
 //
 // Refuses to run unless force is true. The caller is responsible for
 // confirming with the user.
@@ -34,6 +48,29 @@ func (v *Vaultik) NukeRemote(force bool) error {
 		return errNukeRequiresForce
 	}
 
+	// Checked before anything is deleted: running nuke again would fail
+	// the same way, so the advice below must not follow this error.
+	err := v.EnsureStorageBinding()
+	if err != nil {
+		return err
+	}
+
+	err = v.emptyDestinationStore()
+	if err != nil {
+		v.UI.Warningf("Run '%s' again to delete what is left on the "+
+			"backup destination store.", nukeCommandHint)
+
+		return err
+	}
+
+	v.UI.Completef("Backup destination store is now empty.")
+
+	return nil
+}
+
+// emptyDestinationStore does the deleting for NukeRemote and stops at
+// the first step that fails.
+func (v *Vaultik) emptyDestinationStore() error {
 	v.UI.Beginf("Removing all snapshot metadata from backup destination store.")
 
 	_, err := v.RemoveAllSnapshots(&RemoveOptions{Force: true})
@@ -48,7 +85,21 @@ func (v *Vaultik) NukeRemote(force bool) error {
 		return fmt.Errorf("pruning blobs: %w", err)
 	}
 
-	v.UI.Completef("Backup destination store is now empty.")
+	// The file and rclone listings skip `.partial` objects, so the two
+	// steps above never delete them.
+	for _, prefix := range []string{"metadata/", "blobs/"} {
+		err = v.Storage.DeletePartialUploads(v.ctx, prefix)
+		if err != nil {
+			return fmt.Errorf("deleting partial uploads: %w", err)
+		}
+	}
+
+	// Last, once the local snapshot records are gone: the next backup
+	// then has none and writes the marker again.
+	err = v.Storage.Delete(v.ctx, destinationMarkerKey)
+	if err != nil {
+		return fmt.Errorf("deleting %s: %w", destinationMarkerKey, err)
+	}
 
 	return nil
 }
@@ -152,14 +203,17 @@ func (v *Vaultik) PruneBlobs(opts *PruneOptions) error {
 	v.deleteUnreferencedBlobs(unreferencedBlobs, allBlobs, result)
 
 	if opts.JSON {
-		return v.outputPruneBlobsJSON(result)
+		err = v.outputPruneBlobsJSON(result)
+		if err != nil {
+			return err
+		}
+	} else {
+		v.stdoutf("\nDeleted %d blob(s) totaling %s\n",
+			result.BlobsDeleted, ubytes(result.BytesFreed))
 	}
 
-	v.stdoutf("\nDeleted %d blob(s) totaling %s\n",
-		result.BlobsDeleted, ubytes(result.BytesFreed))
-
 	if result.BlobsFailed > 0 {
-		v.stdoutf("Failed to delete %d blob(s)\n", result.BlobsFailed)
+		return fmt.Errorf("%d %w", result.BlobsFailed, errBlobDeleteFailed)
 	}
 
 	return nil

internal/vaultik/prune_count_test.go

@@ -14,10 +14,9 @@ import (
 // the discarded-error bug: getTableCount for a table its query cannot
 // resolve must not silently become 0. A count that could not be read is
 // reported as unknown, which a reader can tell apart from an empty table.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestTableCountForReportSurfacesReadFailure(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	ctx := context.Background()
 
@@ -43,7 +42,7 @@ func TestTableCountForReportSurfacesReadFailure(t *testing.T) {
 	assert.Nil(t, missing, "a failed read is unknown, not a count")
 
 	// The rendered count for a failed read must say unknown, never 0.
-	assert.Equal(t, countUnknown, countText(missing))
+	assert.Equal(t, unknownText, countText(missing))
 	assert.NotEqual(t, "0", countText(missing))
 }
 
@@ -58,7 +57,7 @@ func TestCountTextDistinguishesEmptyFromUnknown(t *testing.T) {
 
 	assert.Equal(t, "0", countText(&zero))
 	assert.Equal(t, "7", countText(&seven))
-	assert.Equal(t, countUnknown, countText(nil))
+	assert.Equal(t, unknownText, countText(nil))
 }
 
 // TestCountDiffUnknownWhenEitherSideUnknown checks that a delta computed

internal/vaultik/prune_delete_failure_test.go

@@ -0,0 +1,98 @@
+package vaultik_test
+
+import (
+	"context"
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage/faultstore"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// refuseBlobDeletes is an OnDelete hook that fails every blob delete.
+func refuseBlobDeletes(key string) bool {
+	return strings.HasPrefix(key, "blobs/")
+}
+
+// TestPruneBlobs_FailedDeleteReturnsError checks that prune returns an
+// error, so the command exits non-zero, when an unreferenced blob cannot
+// be deleted.
+func TestPruneBlobs_FailedDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+
+	// The store holds no manifest, so nothing references this blob.
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	store := faultstore.New(env.store)
+	store.OnDelete = refuseBlobDeletes
+	env.v.Storage = store
+
+	err := env.v.PruneBlobs(&vaultik.PruneOptions{Force: true})
+
+	require.ErrorContains(t, err, "1 blob(s) could not be deleted")
+}
+
+// TestPruneBlobs_JSONFailedDeleteReturnsError checks that prune --json
+// prints its document and then returns an error when an unreferenced
+// blob cannot be deleted.
+func TestPruneBlobs_JSONFailedDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	store := faultstore.New(env.store)
+	store.OnDelete = refuseBlobDeletes
+	env.v.Storage = store
+
+	err := env.v.PruneBlobs(&vaultik.PruneOptions{JSON: true})
+
+	require.Error(t, err, "prune must fail when a blob cannot be deleted")
+
+	var result vaultik.PruneBlobsResult
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &result),
+		"stdout must hold the JSON document, got:\n%s", env.stdout.String())
+	assert.Equal(t, 1, result.BlobsFailed)
+	assert.Equal(t, 0, result.BlobsDeleted)
+}
+
+// TestCreateSnapshotPrune_FailedBlobDeleteReturnsError checks that the
+// prune step of snapshot create --prune fails the run when an
+// unreferenced blob cannot be deleted.
+func TestCreateSnapshotPrune_FailedBlobDeleteReturnsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+	storeDir := filepath.Join(t.TempDir(), "store")
+	v, _, _ := backUpToFileDestination(ctx, t, storeDir)
+
+	// No snapshot references this blob, so the prune step deletes it.
+	orphan := filepath.Join(storeDir, "blobs", testBlobHashA[:2],
+		testBlobHashA[2:4], testBlobHashA)
+	require.NoError(t, os.MkdirAll(filepath.Dir(orphan), 0o750))
+	require.NoError(t, os.WriteFile(orphan, []byte("orphan"), 0o600))
+
+	store := faultstore.New(v.Storage)
+	store.OnDelete = refuseBlobDeletes
+	v.Storage = store
+
+	err := v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		Cron:  true,
+		Prune: true,
+	})
+
+	require.ErrorContains(t, err, "1 blob(s) could not be deleted")
+	assert.FileExists(t, orphan)
+}

internal/vaultik/prune_json_test.go

@@ -0,0 +1,40 @@
+package vaultik_test
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// TestPruneBlobs_JSONDeletesWithoutAsking checks that prune with --json
+// and without --force deletes an unreferenced blob without the
+// confirmation prompt. Stdin is empty, so a prompt would read no answer
+// and cancel, and its text would come before the JSON document.
+func TestPruneBlobs_JSONDeletesWithoutAsking(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+
+	// The store holds no manifest, so nothing references this blob.
+	addBlob(t, env.store.testStorer, testBlobHashA)
+
+	err := env.v.PruneBlobs(&vaultik.PruneOptions{JSON: true})
+	require.NoError(t, err)
+
+	blobKey := "blobs/" + testBlobHashA[:2] + "/" + testBlobHashA[2:4] +
+		"/" + testBlobHashA
+	assert.False(t, env.store.hasKey(blobKey),
+		"the unreferenced blob must be deleted")
+
+	var result vaultik.PruneBlobsResult
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &result),
+		"stdout must hold only the JSON document, got:\n%s",
+		env.stdout.String())
+	assert.Equal(t, 1, result.BlobsDeleted)
+}

internal/vaultik/purge_local_remote_consistency_test.go

@@ -34,6 +34,9 @@ func setupConsistencyTest(
 	repos := database.NewRepositories(db)
 	mockStorage := NewMockStorer()
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, mockStorage.Put(ctx, markerFile, strings.NewReader("")))
+
 	for _, id := range snapshotIDs {
 		parts := strings.Split(id, "_")
 		startedAt, err := time.Parse(time.RFC3339, parts[len(parts)-1])
@@ -42,7 +45,7 @@ func setupConsistencyTest(
 		completedAt := startedAt.Add(5 * time.Minute)
 		snap := &database.Snapshot{
 			ID:             types.SnapshotID(id),
-			Hostname:       testHostname,
+			Hostname:       snapHostname,
 			VaultikVersion: testLabel,
 			StartedAt:      startedAt,
 			CompletedAt:    &completedAt,

internal/vaultik/purge_per_name_test.go

@@ -17,8 +17,10 @@ import (
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
-// Snapshot IDs reused across the purge tests.
+// Snapshot IDs reused across the purge tests, and the hostname they were
+// taken on.
 const (
+	snapHostname = "testhost"
 	snapSystemT0 = "testhost_system_2026-01-01T00:00:00Z"
 	snapHomeT0   = "testhost_home_2026-01-01T00:00:00Z"
 	snapHomeT1   = "testhost_home_2026-01-01T01:00:00Z"
@@ -26,9 +28,12 @@ const (
 )
 
 // setupPurgeTest creates a Vaultik instance with an in-memory database and mock
-// storage pre-populated with the given snapshot IDs. Each snapshot is marked as
+// storage pre-populated with the given snapshot IDs, all taken on hostname.
-// completed. Remote metadata stubs are created so syncWithRemote keeps them.
+// Each snapshot is marked as completed. Remote metadata stubs are created so
-func setupPurgeTest(t *testing.T, snapshotIDs []string) *vaultik.Vaultik {
+// syncWithRemote keeps them.
+func setupPurgeTest(
+	t *testing.T, hostname string, snapshotIDs []string,
+) *vaultik.Vaultik {
 	t.Helper()
 
 	ctx := context.Background()
@@ -39,6 +44,9 @@ func setupPurgeTest(t *testing.T, snapshotIDs []string) *vaultik.Vaultik {
 	repos := database.NewRepositories(db)
 	mockStorage := NewMockStorer()
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, mockStorage.Put(ctx, markerFile, strings.NewReader("")))
+
 	// Insert each snapshot into the DB and create remote metadata stubs.
 	// Use timestamps parsed from snapshot IDs for realistic ordering.
 	for _, id := range snapshotIDs {
@@ -51,7 +59,7 @@ func setupPurgeTest(t *testing.T, snapshotIDs []string) *vaultik.Vaultik {
 		completedAt := startedAt.Add(5 * time.Minute)
 		snap := &database.Snapshot{
 			ID:             types.SnapshotID(id),
-			Hostname:       "testhost",
+			Hostname:       types.Hostname(hostname),
 			VaultikVersion: testLabel,
 			StartedAt:      startedAt,
 			CompletedAt:    &completedAt,
@@ -120,7 +128,7 @@ func TestPurgeKeepLatest_PerName(t *testing.T) {
 		"testhost_system_2026-01-01T04:00:00Z",
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -148,7 +156,7 @@ func TestPurgeKeepLatest_SingleName(t *testing.T) {
 		"testhost_home_2026-01-01T02:00:00Z",
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -176,7 +184,7 @@ func TestPurgeKeepLatest_WithNameFilter(t *testing.T) {
 		"testhost_home_2026-01-01T04:00:00Z",
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -198,7 +206,7 @@ func TestPurgeKeepLatest_NoSnapshots(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
 
-	v := setupPurgeTest(t, nil)
+	v := setupPurgeTest(t, snapHostname, nil)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -216,7 +224,7 @@ func TestPurgeKeepLatest_NameFilterNoMatch(t *testing.T) {
 		"testhost_system_2026-01-01T01:00:00Z",
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -243,7 +251,7 @@ func TestPurgeOlderThan_WithNameFilter(t *testing.T) {
 		snapHomeT0,
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	// Purge only "home" snapshots older than 365 days
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
@@ -262,6 +270,29 @@ func TestPurgeOlderThan_WithNameFilter(t *testing.T) {
 	assert.Contains(t, remaining, snapHomeT0)
 }
 
+func TestPurgeOlderThan_TooLargeDeletesNothing(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	snapshotIDs := []string{snapHomeT0, snapHomeT1}
+
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
+
+	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		OlderThan: "1000y",
+		Force:     true,
+	})
+	require.Error(t, err)
+
+	err = v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		OlderThan: "292y",
+		Force:     true,
+	})
+	require.NoError(t, err)
+
+	assert.ElementsMatch(t, snapshotIDs, listRemainingSnapshots(t, v))
+}
+
 func TestPurgeKeepLatest_ThreeNames(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -277,7 +308,7 @@ func TestPurgeKeepLatest_ThreeNames(t *testing.T) {
 		"testhost_home_2026-01-01T06:00:00Z",
 	}
 
-	v := setupPurgeTest(t, snapshotIDs)
+	v := setupPurgeTest(t, snapHostname, snapshotIDs)
 
 	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		KeepLatest: true,
@@ -291,3 +322,28 @@ func TestPurgeKeepLatest_ThreeNames(t *testing.T) {
 	assert.Contains(t, remaining, "testhost_system_2026-01-01T04:00:00Z")
 	assert.Contains(t, remaining, "testhost_media_2026-01-01T05:00:00Z")
 }
+
+// A hostname may contain underscores, so the snapshot name cannot be found
+// by splitting the ID at them. A purge by name must still select "docs".
+func TestPurgeKeepLatest_HostnameWithUnderscore(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	const (
+		system = "my_host_system_2026-01-01T00:00:00Z"
+		docsT1 = "my_host_docs_2026-01-01T01:00:00Z"
+		docsT2 = "my_host_docs_2026-01-01T02:00:00Z"
+	)
+
+	v := setupPurgeTest(t, "my_host", []string{system, docsT1, docsT2})
+
+	err := v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
+		KeepLatest: true,
+		Force:      true,
+		Names:      []string{"docs"},
+	})
+	require.NoError(t, err)
+
+	assert.ElementsMatch(t, []string{system, docsT2},
+		listRemainingSnapshots(t, v))
+}

internal/vaultik/remote_info_test.go

@@ -0,0 +1,261 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// testBlobHashB is a blob that the manifest written by addRemote does
+// not reference.
+const testBlobHashB = "bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb" +
+	"bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"
+
+// TestRemoteInfo_UnreadableManifestLeavesOrphansUnknown checks that a
+// manifest remote info cannot read makes the orphan figures unknown. A
+// blob referenced only by that snapshot would otherwise be counted as
+// orphaned, and the report would advise running prune.
+func TestRemoteInfo_UnreadableManifestLeavesOrphansUnknown(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+
+	// The readable manifest references blob A only.
+	env.addRemote(t, listRemoteID, time.Date(2026, 3, 2, 0, 0, 0, 0, time.UTC))
+	addBlob(t, env.store.testStorer, testBlobHashA)
+	addBlob(t, env.store.testStorer, testBlobHashB)
+
+	// With every manifest readable, blob B is orphaned.
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	var doc map[string]any
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &doc))
+	assert.InDelta(t, 1, doc["orphaned_blob_count"], 0)
+
+	// A second snapshot whose manifest cannot be decoded. Blob B may be
+	// one of its blobs.
+	unreadableKey := snapshot.RemoteSnapshotKey(listLocalID)
+	require.NoError(t, env.store.Put(context.Background(),
+		"metadata/"+unreadableKey+"/manifest.json.zst",
+		bytes.NewReader([]byte("not a valid manifest"))))
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(false))
+
+	text := env.stdout.String()
+	assert.Contains(t, text, "Orphaned (unreferenced):    unknown "+
+		"(1 manifest(s) could not be read, "+
+		"0 manifest(s) under a non-conforming name skipped)")
+	assert.NotContains(t, text, "vaultik prune")
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	doc = nil
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &doc))
+	assert.Contains(t, doc, "orphaned_blob_count")
+	assert.Nil(t, doc["orphaned_blob_count"])
+	assert.Contains(t, doc, "orphaned_blob_size")
+	assert.Nil(t, doc["orphaned_blob_size"])
+	assert.Equal(t, []any{unreadableKey}, doc["unreadable_manifests"])
+}
+
+// TestRemoteInfo_UnreadableManifestLeavesSnapshotBlobsUnknown checks
+// that the row of a snapshot whose manifest cannot be read gives its
+// blob count and blob size as unknown in the table and as null in
+// --json, not as 0. A directory without a manifest still shows 0: the
+// orphan figures count its blobs as orphaned, so it references none.
+func TestRemoteInfo_UnreadableManifestLeavesSnapshotBlobsUnknown(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	readableKey := env.addRemote(t, listRemoteID,
+		time.Date(2026, 3, 2, 0, 0, 0, 0, time.UTC))
+
+	unreadableKey := snapshot.RemoteSnapshotKey(listLocalID)
+	require.NoError(t, env.store.Put(context.Background(),
+		"metadata/"+unreadableKey+"/manifest.json.zst",
+		bytes.NewReader([]byte("not a valid manifest"))))
+
+	noManifestKey := snapshot.RemoteSnapshotKey("testhost_home_2026-03-03T10:00:00Z")
+	require.NoError(t, env.store.Put(context.Background(),
+		"metadata/"+noManifestKey+"/db.zst.age",
+		bytes.NewReader([]byte("not a valid database"))))
+
+	require.NoError(t, env.v.RemoteInfo(false))
+
+	// The table truncates the remote key, so a row is found by a prefix.
+	wantUnknown := map[string]int{readableKey: 0, unreadableKey: 2, noManifestKey: 0}
+	for key, want := range wantUnknown {
+		var row string
+
+		for line := range strings.SplitSeq(env.stdout.String(), "\n") {
+			if strings.HasPrefix(line, key[:16]) {
+				row = line
+			}
+		}
+
+		require.NotEmpty(t, row, "no table row for %s", key)
+		assert.Equal(t, want, strings.Count(row, "unknown"), "row: %q", row)
+	}
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	var doc struct {
+		Snapshots []map[string]any `json:"snapshots"`
+	}
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &doc))
+	require.Len(t, doc.Snapshots, len(wantUnknown))
+
+	for _, entry := range doc.Snapshots {
+		switch entry["snapshot_id"] {
+		case readableKey:
+			assert.InDelta(t, 1, entry["blob_count"], 0)
+			assert.InDelta(t, fiveMegabytes, entry["blobs_size"], 0)
+		case noManifestKey:
+			assert.InDelta(t, 0, entry["blob_count"], 0)
+			assert.InDelta(t, 0, entry["blobs_size"], 0)
+		default:
+			assert.Equal(t, unreadableKey, entry["snapshot_id"])
+			assert.Contains(t, entry, "blob_count")
+			assert.Nil(t, entry["blob_count"])
+			assert.Contains(t, entry, "blobs_size")
+			assert.Nil(t, entry["blobs_size"])
+		}
+	}
+}
+
+// TestRemoteInfo_SkipsNonConformingMetadataName checks that a directory
+// under metadata/ whose name is not a remote key is left out of the
+// report, and that the orphan figures are unknown when it holds a
+// manifest. The name comes from the destination store; printed raw, its
+// control characters would reach the terminal. Its manifest is not
+// read, so a blob only it references would otherwise be counted as
+// orphaned.
+func TestRemoteInfo_SkipsNonConformingMetadataName(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	env.addRemote(t, listRemoteID, time.Date(2026, 3, 2, 0, 0, 0, 0, time.UTC))
+	addBlob(t, env.store.testStorer, testBlobHashA)
+	addBlob(t, env.store.testStorer, testBlobHashB)
+	require.NoError(t, env.store.Put(context.Background(),
+		"metadata/\x1b[31mred/manifest.json.zst",
+		bytes.NewReader([]byte("not a valid manifest"))))
+
+	require.NoError(t, env.v.RemoteInfo(false))
+
+	text := env.stdout.String()
+	assert.NotContains(t, text, "\x1b")
+	assert.NotContains(t, text, "31mred")
+	assert.Contains(t, text, "Total (1 snapshots)")
+	assert.Contains(t, text, "Orphaned (unreferenced):    unknown "+
+		"(0 manifest(s) could not be read, "+
+		"1 manifest(s) under a non-conforming name skipped)")
+	assert.NotContains(t, text, "vaultik prune")
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	out := env.stdout.String()
+	assert.NotContains(t, out, "31mred")
+
+	var doc map[string]any
+
+	require.NoError(t, json.Unmarshal([]byte(out), &doc))
+	assert.Contains(t, doc, "orphaned_blob_count")
+	assert.Nil(t, doc["orphaned_blob_count"])
+	assert.Contains(t, doc, "orphaned_blob_size")
+	assert.Nil(t, doc["orphaned_blob_size"])
+	assert.InDelta(t, 1, doc["skipped_manifest_count"], 0)
+	assert.NotContains(t, doc, "unreadable_manifests")
+}
+
+// TestRemoteInfo_DoesNotCountNonBlobFileUnderBlobs checks that a file
+// under blobs/ whose name is not a blob hash is not counted as a blob.
+// prune never deletes such a file, so counting it as an orphaned blob
+// would advise a prune that removes nothing.
+func TestRemoteInfo_DoesNotCountNonBlobFileUnderBlobs(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	env.addRemote(t, listRemoteID, time.Date(2026, 3, 2, 0, 0, 0, 0, time.UTC))
+	addBlob(t, env.store.testStorer, testBlobHashA)
+	require.NoError(t, env.store.Put(context.Background(),
+		"blobs/"+testBlobHashA[:2]+"/"+testBlobHashA[2:4]+"/.DS_Store",
+		bytes.NewReader([]byte("junk"))))
+
+	require.NoError(t, env.v.RemoteInfo(false))
+
+	text := env.stdout.String()
+	assert.Contains(t, text, "Total blobs on remote:      1 (")
+	assert.Contains(t, text, "Orphaned (unreferenced):    0 (")
+	assert.NotContains(t, text, "vaultik prune")
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	var doc map[string]any
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &doc))
+	assert.InDelta(t, 1, doc["total_blob_count"], 0)
+	assert.InDelta(t, 0, doc["orphaned_blob_count"], 0)
+	assert.InDelta(t, 0, doc["orphaned_blob_size"], 0)
+}
+
+// TestRemoteInfo_DirectoryWithoutManifestLeavesOrphansKnown checks that
+// a directory under metadata/ holding no manifest.json.zst, such as one
+// left by a backup interrupted before its manifest upload, leaves the
+// orphan figures known. prune does not treat such a directory as a
+// snapshot and deletes the blobs the report lists as orphaned.
+func TestRemoteInfo_DirectoryWithoutManifestLeavesOrphansKnown(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	env := newListEnv(t)
+	env.addRemote(t, listRemoteID, time.Date(2026, 3, 2, 0, 0, 0, 0, time.UTC))
+	addBlob(t, env.store.testStorer, testBlobHashA)
+	addBlob(t, env.store.testStorer, testBlobHashB)
+
+	// One directory under a remote key and one under a non-conforming
+	// name, each holding only a database.
+	names := []string{snapshot.RemoteSnapshotKey(listLocalID), "\x1b[31mred"}
+	for _, name := range names {
+		require.NoError(t, env.store.Put(context.Background(),
+			"metadata/"+name+"/db.zst.age",
+			bytes.NewReader([]byte("not a valid database"))))
+	}
+
+	require.NoError(t, env.v.RemoteInfo(false))
+
+	text := env.stdout.String()
+	assert.NotContains(t, text, "\x1b")
+	assert.Contains(t, text, "Downloading 1 manifest(s)...")
+	assert.Contains(t, text, "Orphaned (unreferenced):    1 (")
+	assert.Contains(t, text, "Run 'vaultik prune' to remove orphaned blobs.")
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.RemoteInfo(true))
+
+	var doc map[string]any
+
+	require.NoError(t, json.Unmarshal(env.stdout.Bytes(), &doc))
+	assert.InDelta(t, 1, doc["orphaned_blob_count"], 0)
+	assert.NotContains(t, doc, "unreadable_manifests")
+	assert.NotContains(t, doc, "skipped_manifest_count")
+}

internal/vaultik/remove_snapshot_test.go

@@ -3,17 +3,21 @@ package vaultik_test
 import (
 	"bytes"
 	"context"
+	"database/sql"
 	"io"
 	"strings"
 	"sync"
 	"testing"
+	"time"
 
 	"github.com/klauspost/compress/zstd"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/database"
 	"sneak.berlin/go/vaultik/internal/log"
 	"sneak.berlin/go/vaultik/internal/snapshot"
 	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/types"
 	"sneak.berlin/go/vaultik/internal/vaultik"
 )
 
@@ -125,6 +129,12 @@ func (s *testStorer) ListStream(
 	return ch
 }
 
+// DeletePartialUploads has nothing to remove: Put stores each object
+// under its key at once.
+func (s *testStorer) DeletePartialUploads(_ context.Context, _ string) error {
+	return nil
+}
+
 func (s *testStorer) Info() storage.Info {
 	return storage.Info{
 		Type:     testLabel,
@@ -208,6 +218,37 @@ func addBlob(t *testing.T, store *testStorer, hash string) {
 	require.NoError(t, err)
 }
 
+// newLocalIndex returns an in-memory local index holding a completed
+// snapshot row for each of ids.
+func newLocalIndex(t *testing.T, ids ...string) *database.Repositories {
+	t.Helper()
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, ":memory:")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+
+	for _, id := range ids {
+		completedAt := time.Now().UTC()
+		snap := &database.Snapshot{
+			ID:          types.SnapshotID(id),
+			Hostname:    snapHostname,
+			StartedAt:   completedAt,
+			CompletedAt: &completedAt,
+		}
+
+		err = repos.WithTx(ctx, func(ctx context.Context, tx *sql.Tx) error {
+			return repos.Snapshots.Create(ctx, tx, snap)
+		})
+		require.NoError(t, err)
+	}
+
+	return repos
+}
+
 // ============================================================================
 // Unit Tests for RemoveSnapshot
 // ============================================================================
@@ -227,6 +268,7 @@ func TestRemoveSnapshot_LocalOnly_PreservesRemote(t *testing.T) {
 	addBlob(t, store, blobA)
 
 	tv := vaultik.NewForTesting(store)
+	tv.Repositories = newLocalIndex(t, "snapshot-001")
 
 	opts := &vaultik.RemoveOptions{Force: true, LocalOnly: true}
 	result, err := tv.RemoveSnapshot("snapshot-001", opts)
@@ -261,7 +303,12 @@ func TestRemoveSnapshot_DefaultRemovesMetadataNotBlobs(t *testing.T) {
 	addBlob(t, store, blobUnique)
 	addBlob(t, store, blobShared)
 
+	// A backup wrote to this destination, so it carries the marker.
+	require.NoError(t, store.Put(context.Background(), markerFile,
+		bytes.NewReader(nil)))
+
 	tv := vaultik.NewForTesting(store)
+	tv.Repositories = newLocalIndex(t, "snapshot-001", "snapshot-002")
 
 	opts := &vaultik.RemoveOptions{Force: true}
 	result, err := tv.RemoveSnapshot("snapshot-001", opts)
@@ -310,6 +357,89 @@ func TestRemoveSnapshot_DryRun(t *testing.T) {
 	assert.Contains(t, tv.Stdout.String(), "[Dry run - no changes made]")
 }
 
+// TestRemoveSnapshot_UnknownIDIsError checks that an ID naming no
+// snapshot in the local index or on the destination store is an error
+// under every option, and that nothing is reported as removed.
+func TestRemoveSnapshot_UnknownIDIsError(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	for name, opts := range map[string]vaultik.RemoveOptions{
+		"force":      {Force: true},
+		"json":       {JSON: true},
+		"dry-run":    {Force: true, DryRun: true},
+		"local-only": {Force: true, LocalOnly: true},
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			store := newTestStorer()
+			addManifest(t, store, "snapshot-001", nil)
+
+			tv := vaultik.NewForTesting(store)
+			tv.Repositories = newLocalIndex(t, "snapshot-001")
+
+			result, err := tv.RemoveSnapshot("snapshot-999", &opts)
+
+			require.ErrorContains(t, err, "snapshot not found")
+			assert.False(t, result.RemoteRemoved)
+			assert.Empty(t, tv.Stdout.String())
+			assert.True(t, store.hasKey(remoteKeyPath("snapshot-001")))
+		})
+	}
+}
+
+// TestRemoveSnapshot_NoMetadataOnDestination checks that removing a
+// snapshot whose metadata is not on the destination store does not
+// report metadata removed from it.
+func TestRemoveSnapshot_NoMetadataOnDestination(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	// A backup wrote to this destination, so it carries the marker.
+	store := newTestStorer()
+	require.NoError(t, store.Put(context.Background(), markerFile,
+		bytes.NewReader(nil)))
+
+	tv := vaultik.NewForTesting(store)
+	tv.Repositories = newLocalIndex(t, "snapshot-001")
+
+	result, err := tv.RemoveSnapshot("snapshot-001",
+		&vaultik.RemoveOptions{Force: true})
+
+	require.NoError(t, err)
+	assert.False(t, result.RemoteRemoved)
+
+	out := tv.Stdout.String()
+	assert.Contains(t, out, "Removed snapshot 'snapshot-001' from local database")
+	assert.NotContains(t, out, "Removed snapshot metadata from remote storage")
+}
+
+// TestRemoveSnapshot_NotInLocalIndex checks that removing a snapshot
+// only the destination store holds removes its metadata there without
+// reporting a removal from the local database.
+func TestRemoveSnapshot_NotInLocalIndex(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	store := newTestStorer()
+	addManifest(t, store, "snapshot-001", nil)
+
+	tv := vaultik.NewForTesting(store)
+	tv.Repositories = newLocalIndex(t)
+
+	result, err := tv.RemoveSnapshot("snapshot-001",
+		&vaultik.RemoveOptions{Force: true})
+
+	require.NoError(t, err)
+	assert.True(t, result.RemoteRemoved)
+	assert.False(t, store.hasKey(remoteKeyPath("snapshot-001")))
+
+	out := tv.Stdout.String()
+	assert.NotContains(t, out, "from local database")
+	assert.Contains(t, out, "Removed snapshot metadata from remote storage")
+}
+
 func TestRemoveAllSnapshots_RequiresForce(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -355,7 +485,8 @@ func TestRemoveAllSnapshots_WithForce(t *testing.T) {
 	out := tv.Stdout.String()
 	assert.Contains(t, out, "Removed 2 snapshot(s)")
 	assert.Contains(t, out, "Removed snapshot metadata from remote storage")
-	assert.Contains(t, out, "vaultik prune")
+	// Remote nuke, the only caller, deletes the blobs next itself.
+	assert.NotContains(t, out, "vaultik prune")
 }
 
 func TestRemoveAllSnapshots_DryRun(t *testing.T) {

internal/vaultik/restore.go

@@ -309,11 +309,22 @@ func (v *Vaultik) restoreAllFiles(
 		return nil, err
 	}
 
-	plan, err := newRestorePlan(v.ctx, repos, files, chunkToBlobMap, blobIDToHash)
+	plan, unplanned, err := newRestorePlan(
+		v.ctx, repos, files, chunkToBlobMap, blobIDToHash)
 	if err != nil {
 		return nil, fmt.Errorf("building restore plan: %w", err)
 	}
 
+	// Ranging over files rather than the map reports them in a stable order.
+	for _, file := range files {
+		if planErr, ok := unplanned[file.ID]; ok {
+			err = v.handleRestoreFileError(plan, opts, result, file, file.ID, planErr)
+			if err != nil {
+				return nil, err
+			}
+		}
+	}
+
 	filesByID, totalBytesExpected := indexRestoreFiles(files)
 
 	v.UI.Beginf("Restoring %s files (%s) to %s.",
@@ -691,6 +702,8 @@ func (v *Vaultik) downloadSnapshotDB(
 		return nil, "", fmt.Errorf("downloading %s: %w", dbKey, err)
 	}
 
+	reader = newCloseOnCancelReader(v.ctx, reader)
+
 	defer func() { _ = reader.Close() }()
 
 	// Decrypt and decompress straight from the storage stream, then stream

internal/vaultik/restore_another_machine_test.go

@@ -42,7 +42,7 @@ func TestRestoreOnAnotherMachine(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")
 

internal/vaultik/restore_interrupt_test.go

@@ -65,7 +65,7 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 	root := t.TempDir()
 
 	dataDir := filepath.Join(root, "source")
-	storeDir := filepath.Join(root, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(root, "restored")
 	dbPath := filepath.Join(root, "index.sqlite")
 
@@ -142,14 +142,14 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 }
 
 // scratchEntries returns the vaultik blob-cache and snapshot-database
-// scratch entries currently present in dir.
+// scratch entries (restore's and deep verify's) currently present in dir.
 func scratchEntries(t *testing.T, dir string) []string {
 	t.Helper()
 
 	var matches []string
 
 	for _, pattern := range []string{
-		"vaultik-blobcache-*", "vaultik-restore-*",
+		"vaultik-blobcache-*", "vaultik-restore-*", "vaultik-verify-*",
 	} {
 		found, err := filepath.Glob(filepath.Join(dir, pattern))
 		require.NoError(t, err)
@@ -164,10 +164,9 @@ func scratchEntries(t *testing.T, dir string) []string {
 // restore while a blob download is in progress. The download fails only
 // because of the cancel, so Restore must return context.Canceled without
 // reporting the file that needs the blob as failed.
-//
-//nolint:paralleltest // installs the global logger via log.Initialize
 func TestRestoreSkipErrorsCancelDuringBlobDownload(t *testing.T) {
 	log.Initialize(log.Config{})
+	t.Parallel()
 
 	fs := afero.NewOsFs()
 	tempDir := t.TempDir()
@@ -207,3 +206,141 @@ func TestRestoreSkipErrorsCancelDuringBlobDownload(t *testing.T) {
 	assert.NotContains(t, out.String(), srcPath,
 		"the cancel was reported as a failed file")
 }
+
+// stallingStorer wraps a Storer so that each download of a key chosen by
+// stalls hands over the first half of the object and then stalls,
+// ignoring the context, as a file:// read on a hung USB stick or network
+// share does. Only closing the returned reader ends the stall.
+type stallingStorer struct {
+	storage.Storer
+
+	stalls  func(key string) bool
+	once    sync.Once
+	stalled chan struct{}
+}
+
+func newStallingStorer(
+	inner storage.Storer, stalls func(key string) bool,
+) *stallingStorer {
+	return &stallingStorer{
+		Storer:  inner,
+		stalls:  stalls,
+		stalled: make(chan struct{}),
+	}
+}
+
+func (s *stallingStorer) Get(
+	ctx context.Context, key string,
+) (io.ReadCloser, error) {
+	rc, err := s.Storer.Get(ctx, key)
+	if err != nil || !s.stalls(key) {
+		return rc, err
+	}
+
+	object, err := io.ReadAll(rc)
+	_ = rc.Close()
+
+	if err != nil {
+		return nil, err
+	}
+
+	// The writer is never closed, so once the first half has been read
+	// the next Read blocks until the reader is closed.
+	pr, pw := io.Pipe()
+
+	go func() {
+		_, _ = pw.Write(object[:len(object)/2])
+
+		s.once.Do(func() { close(s.stalled) })
+	}()
+
+	return pr, nil
+}
+
+// TestInterruptDuringStalledReadCleansTempDir stops a restore and a deep
+// verify while a read of a blob or of the snapshot database is stalled
+// on a source that ignores the context. Stopping must end that read, so
+// the operation unwinds and removes its decrypted scratch files well
+// within the shutdown timeout (issues #325 and #356).
+//
+// Not parallel: it points TMPDIR at a private directory (via t.Setenv)
+// so it can assert on exactly the scratch files the operation created.
+func TestInterruptDuringStalledReadCleansTempDir(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	stalledReads := map[string]func(key string) bool{
+		"blob": func(key string) bool {
+			return strings.HasPrefix(key, "blobs/")
+		},
+		"snapshot database": func(key string) bool {
+			return strings.HasSuffix(key, "/db.zst.age")
+		},
+	}
+
+	operations := map[string]func(v *Vaultik, snapshotID, targetDir string) error{
+		"restore": func(v *Vaultik, snapshotID, targetDir string) error {
+			return v.Restore(&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  targetDir,
+			})
+		},
+		"deep verify": func(v *Vaultik, snapshotID, _ string) error {
+			return v.RunDeepVerify(snapshotID, &VerifyOptions{Deep: true})
+		},
+	}
+
+	for readName, stalls := range stalledReads {
+		for opName, op := range operations {
+			t.Run(readName+" read during "+opName, func(t *testing.T) {
+				fs := afero.NewOsFs()
+				root := t.TempDir()
+
+				// Random content, so the blob stays large enough to stall
+				// part-way through its compressed, encrypted body.
+				cfg, storer, snapshotID, _ := backupOneFile(context.Background(),
+					t, fs, root, "a.bin", randomBytes(t, 256*1024), 0o644)
+
+				scratch := filepath.Join(root, "scratch")
+				require.NoError(t, fs.MkdirAll(scratch, 0o755))
+				t.Setenv("TMPDIR", scratch)
+
+				stalling := newStallingStorer(storer, stalls)
+
+				ctx, cancel := context.WithCancel(context.Background())
+				defer cancel()
+
+				v := newRestoreVaultik(ctx, cfg, stalling, fs)
+				v.cancel = cancel
+
+				var opErr error
+
+				stop := v.StartOperation(func() {
+					opErr = op(v, snapshotID, filepath.Join(root, "restored"))
+				})
+
+				select {
+				case <-stalling.stalled:
+				case <-time.After(30 * time.Second):
+					t.Fatal("the operation never reached the stalled read")
+				}
+
+				// The snapshot database's temp directory is created just
+				// after its read starts, so it can appear after the stall.
+				require.Eventually(t, func() bool {
+					return len(scratchEntries(t, scratch)) > 0
+				}, 10*time.Second, 10*time.Millisecond,
+					"expected decrypted scratch files to exist mid-read")
+
+				stopCtx, stopCancel := context.WithTimeout(
+					context.Background(), 10*time.Second)
+				defer stopCancel()
+
+				require.True(t, stop(stopCtx),
+					"stop timed out; the stalled read was not interrupted")
+				require.ErrorIs(t, opErr, context.Canceled)
+				require.Empty(t, scratchEntries(t, scratch),
+					"decrypted scratch files remained after the interrupt")
+			})
+		}
+	}
+}

internal/vaultik/restore_locality_test.go

@@ -197,7 +197,7 @@ func TestRestoreLocalityAndReadAt(t *testing.T) {
 	tempDir := t.TempDir()
 
 	dataDir := filepath.Join(tempDir, "source")
-	storeDir := filepath.Join(tempDir, "remote")
+	storeDir := t.TempDir()
 	restoreDir := filepath.Join(tempDir, "restored")
 	dbPath := filepath.Join(tempDir, "index.sqlite")