Return errors instead of exiting on rollback, rows close and blob temp file cleanup (closes #345)

WithReadTx ended the process when its rollback failed, including after
Ctrl-C, when database/sql has already rolled the transaction back and
Rollback returns sql.ErrTxDone. It now handles the rollback as WithTx
does. The query functions ended the process when closing their rows
failed; they now ignore the deferred close, because database/sql reports
a close failure through rows.Err(), which each returns, once the rows
are read to the end. The unused database.Fatalf is removed.
cleanupBlobTempFile returns its close or remove failure and the upload
step returns it, so such a failure now fails the backup.

No test makes a rows close fail: SQLite fails it only with the error the
last read already returned.

Model: opus-5-5

README.md

@@ -286,7 +286,9 @@ 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.
 
 **`snapshot list`**: Show every snapshot known to this host — the union
 of the local index and the backup destination store — with timestamps
@@ -349,9 +351,11 @@ 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,
-chunks, blobs the snapshot was the last referrer for) runs
-automatically. If the destination store is unreachable, the local-DB
+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; with
+`--local-only`, the snapshot must be in the local index. Local row
+cleanup (files, chunks, blobs the snapshot was the last referrer for)
+runs automatically. If the destination store is unreachable, 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
@@ -359,7 +363,8 @@ 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
@@ -383,7 +388,8 @@ 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.
 * `--force`: Skip confirmation prompt
-* `--json`: Output stats as JSON
+* `--json`: Output stats as JSON. Also skips the confirmation prompt, as
+  `--force` does.
 
 **`info`**: Display system configuration, storage settings, encryption
 recipients, and local database statistics.
@@ -396,7 +402,10 @@ key is skipped with a warning and is not printed. If a listed
 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`.
+`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
@@ -409,7 +418,8 @@ destination store.
 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
 
 ---
@@ -423,11 +433,24 @@ 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.
 
 **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.

TODO.md

@@ -22,6 +22,130 @@ 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

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

@@ -225,16 +225,20 @@ var errInterrupted = errors.New("interrupted before the command finished")
 // op's cleanup (removing decrypted scratch files) runs before the
 // process exits; the wait is bounded by shutdownTimeout. report is
 // called with a failure so the caller can show it to the user before
-// it becomes errReported. An interrupted op is not reported, whatever
-// it returned; RunOperation returns errInterrupted instead.
+// it becomes errReported.
+//
+// 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
-		failed      bool
-		interrupted bool
+		mu       sync.Mutex
+		finished bool // op returned before any interrupt cancelled it
+		failed   bool // op finished with an error
 	)
 
 	opts.Invokes = append(opts.Invokes,
@@ -247,21 +251,19 @@ func RunOperation(
 						err := op(v)
 
 						// Only stop, called from OnStop below, cancels the
-						// Vaultik context; before op has returned, that
-						// happens only on an interrupt. Check the context,
-						// not err: an interrupted op need not return
+						// 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).
-						switch {
-						case v.Context().Err() != nil:
-							mu.Lock()
-							interrupted = true
-							mu.Unlock()
-						case err != nil:
-							report(err)
+						if v.Context().Err() == nil {
+							if err != nil {
+								report(err)
+							}
 
 							mu.Lock()
-							failed = true
+							finished = true
+							failed = err != nil
 							mu.Unlock()
 						}
 
@@ -282,12 +284,6 @@ func RunOperation(
 					if !stop(ctx) {
 						log.Warn("Shutdown timed out before the operation " +
 							"finished; decrypted temporary files may remain")
-
-						// op has not returned, so the app is stopping on
-						// an interrupt.
-						mu.Lock()
-						interrupted = true
-						mu.Unlock()
 					}
 
 					return nil
@@ -300,16 +296,17 @@ func RunOperation(
 		return err
 	}
 
-	// RunWithApp returns only after OnStop has waited for the goroutine
-	// to return, whether the shutdown came from op finishing or from an
-	// interrupt, so the goroutine's write to failed or interrupted is in
-	// place by the time we read it. If OnStop timed out, the goroutine
-	// may still be running, and OnStop has set interrupted itself.
+	// RunWithApp returns only after the app was asked to stop, either by
+	// 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()
 
 	switch {
-	case interrupted:
+	case !finished:
 		return errInterrupted
 	case failed:
 		return errReported

internal/cli/config.go

@@ -251,6 +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())

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_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_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

@@ -57,7 +57,8 @@ referenced.`,
 	}
 
 	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

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
@@ -71,6 +77,11 @@ 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.
@@ -97,6 +108,13 @@ on Linux), then /etc/vaultik/config.yml.`,
 			}, 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)
 			})
@@ -255,8 +273,12 @@ 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; run 'vaultik snapshot remove <snapshot-id>' again
@@ -280,7 +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

@@ -61,6 +61,9 @@ var (
 	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")
@@ -461,11 +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

@@ -337,6 +337,35 @@ func TestLoadS3PartSize(t *testing.T) {
 	})
 }
 
+// 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

internal/database/blob_chunks.go

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

internal/database/chunk_files.go

@@ -61,12 +61,7 @@ func (r *ChunkFileRepository) GetByChunkHash(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -87,12 +82,7 @@ func (r *ChunkFileRepository) GetByFilePath(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanChunkFiles(rows)
 }
@@ -112,12 +102,7 @@ func (r *ChunkFileRepository) GetByFileID(
 		return nil, fmt.Errorf("querying chunk files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	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() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var chunks []*Chunk
 
@@ -152,12 +147,7 @@ func (r *ChunkRepository) ListUnpacked(
 		return nil, fmt.Errorf("querying unpacked chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	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() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	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() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -88,12 +83,7 @@ func (r *FileChunkRepository) GetByFileID(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanFileChunks(rows)
 }
@@ -117,12 +107,7 @@ func (r *FileChunkRepository) GetByPathTx(
 		return nil, fmt.Errorf("querying file chunks: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	fileChunks, err := r.scanFileChunks(rows)
 	LogSQL("GetByPathTx", "Complete", path, "count", len(fileChunks))

internal/database/files.go

@@ -175,12 +175,7 @@ func (r *FileRepository) ListModifiedSince(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -256,12 +251,7 @@ func (r *FileRepository) ListUnderPath(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 
@@ -304,12 +294,7 @@ func (r *FileRepository) ListIDsWithChunksNotInUploadedBlobs(
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var ids []types.FileID
 
@@ -340,12 +325,7 @@ func (r *FileRepository) ListAll(ctx context.Context) ([]*File, error) {
 		return nil, fmt.Errorf("querying files: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var files []*File
 

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()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = 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()
-			if rollbackErr != nil {
-				Fatalf("failed to rollback transaction: %v", rollbackErr)
-			}
+			// The panic is the failure to report, so a rollback error
+			// is dropped.
+			_ = 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/snapshots.go

@@ -239,12 +239,7 @@ func (r *SnapshotRepository) GetOnlySnapshot(ctx context.Context) (*Snapshot, er
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	snapshots, err := r.scanSnapshotRows(rows)
 	if err != nil {
@@ -279,12 +274,7 @@ func (r *SnapshotRepository) ListRecent(
 		return nil, fmt.Errorf("querying snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -499,12 +489,7 @@ func (r *SnapshotRepository) GetBlobHashes(
 		return nil, fmt.Errorf("querying blob hashes: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	var blobs []string
 
@@ -653,12 +638,7 @@ func (r *SnapshotRepository) GetIncompleteSnapshots(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	return r.scanSnapshotRows(rows)
 }
@@ -681,12 +661,7 @@ func (r *SnapshotRepository) GetIncompleteByHostname(
 		return nil, fmt.Errorf("querying incomplete snapshots: %w", err)
 	}
 
-	defer func() {
-		err := rows.Close()
-		if err != nil {
-			Fatalf("failed to close rows: %v", err)
-		}
-	}()
+	defer func() { _ = rows.Close() }()
 
 	// Same column set as every other multi-row snapshot query, so the
 	// shared scanner applies — including its timestamp normalization.

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 < unit {
+	if b < 0 {
 		return fmt.Sprintf("%d B", b)
 	}
 
-	div, exp := int64(unit), 0
-	for n := b / unit; n >= unit; n /= unit {
-		div *= unit
-		exp++
-	}
-
-	return fmt.Sprintf("%.1f %cB", float64(b)/float64(div), "KMGTPE"[exp])
+	return humanize.Bytes(uint64(b))
 }

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/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)
-	FilesProcessed   atomic.Int64 // Files actually processed in phase 2
-	FilesSkipped     atomic.Int64 // Files skipped due to no changes
-	BytesScanned     atomic.Int64 // Bytes from new/changed files only
-	BytesSkipped     atomic.Int64 // Bytes from unchanged files
-	BytesProcessed   atomic.Int64 // Actual bytes processed (for ETA calculation)
-	ChunksCreated    atomic.Int64
-	BlobsCreated     atomic.Int64
-	BlobsUploaded    atomic.Int64
-	BytesUploaded    atomic.Int64
-	CurrentFile      atomic.Value // stores string
-	TotalSize        atomic.Int64 // Total size to process (set after scan phase)
-	TotalFiles       atomic.Int64 // Total files to process in phase 2
-	ProcessStartTime atomic.Value // stores time.Time when processing starts
-	StartTime        time.Time
-	mu               sync.RWMutex
-	lastDetailTime   time.Time
+	FilesScanned   atomic.Int64 // Total files seen during scan (includes skipped)
+	FilesProcessed atomic.Int64 // Files actually processed in phase 2
+	FilesSkipped   atomic.Int64 // Files skipped due to no changes
+	BytesScanned   atomic.Int64 // Bytes from new/changed files only
+	BytesSkipped   atomic.Int64 // Bytes from unchanged files
+	BytesProcessed atomic.Int64 // Actual bytes processed (for ETA calculation)
+	ChunksCreated  atomic.Int64
+	BlobsCreated   atomic.Int64
+	BlobsUploaded  atomic.Int64
+	BytesUploaded  atomic.Int64
+	CurrentFile    atomic.Value // stores string
+	TotalSize      atomic.Int64 // Size to process in the paths scanned so far
+	TotalFiles     atomic.Int64 // Files to process in the paths scanned so far
+	StartTime      time.Time
+	mu             sync.RWMutex
+	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)
-func (pr *ProgressReporter) SetTotalSize(size int64) {
-	pr.stats.TotalSize.Store(size)
-	pr.stats.ProcessStartTime.Store(time.Now().UTC())
+// AddTotalSize adds the size one path of the snapshot has to process to
+// the total, once that path's scan phase is done, and starts measuring
+// the processing rate again. The processed counts run across every path,
+// 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 {
-		processStart, ok := pr.stats.ProcessStartTime.Load().(time.Time)
-		if ok && !processStart.IsZero() {
-			processElapsed := time.Since(processStart)
-
-			rate := float64(bytesProcessed) / processElapsed.Seconds()
-			if rate > 0 {
-				remainingBytes := totalSize - bytesProcessed
-				remainingSeconds := float64(remainingBytes) / rate
-				eta := time.Duration(remainingSeconds * float64(time.Second))
-				etaStr = " | ETA: " + formatDuration(eta)
-			}
-		}
+	processRate := pr.stats.processRate()
+	if totalSize > 0 && processRate > 0 {
+		remainingBytes := totalSize - bytesProcessed
+		remainingSeconds := float64(remainingBytes) / processRate
+		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 {
-		processStart, ok := pr.stats.ProcessStartTime.Load().(time.Time)
-		if ok && !processStart.IsZero() {
-			processElapsed := time.Since(processStart)
-
-			processRate := float64(bytesProcessed) / processElapsed.Seconds()
-			if processRate > 0 {
-				remainingBytes := totalSize - bytesProcessed
-				remainingSeconds := float64(remainingBytes) / processRate
-				eta := time.Duration(remainingSeconds * float64(time.Second))
-				percentComplete := float64(bytesProcessed) / float64(totalSize) * percentScale
-				log.Info("Overall progress",
-					"percent", fmt.Sprintf("%.1f%%", percentComplete),
-					"processed", humanize.Bytes(safeUint64(bytesProcessed)),
-					"total", humanize.Bytes(safeUint64(totalSize)),
-					"rate", humanize.Bytes(uint64(processRate))+"/s",
-					"eta", formatDuration(eta))
-			}
-		}
+	processRate := pr.stats.processRate()
+	if totalSize > 0 && processRate > 0 {
+		remainingBytes := totalSize - bytesProcessed
+		remainingSeconds := float64(remainingBytes) / processRate
+		eta := time.Duration(remainingSeconds * float64(time.Second))
+		percentComplete := float64(bytesProcessed) / float64(totalSize) * percentScale
+		log.Info("Overall progress",
+			"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

@@ -133,10 +133,13 @@ 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.
 type ScanResult struct {
 	FilesScanned   int
 	FilesSkipped   int
+	FilesFailed    int
 	FilesDeleted   int
 	BytesScanned   int64
 	BytesSkipped   int64
@@ -407,8 +410,8 @@ func (s *Scanner) summarizeScanPhase(
 	}
 
 	if s.progress != nil {
-		s.progress.SetTotalSize(totalSizeToProcess)
-		s.progress.GetStats().TotalFiles.Store(int64(len(filesToProcess)))
+		s.progress.AddTotalSize(totalSizeToProcess)
+		s.progress.GetStats().TotalFiles.Add(int64(len(filesToProcess)))
 	}
 
 	log.Info("Phase 1 complete",
@@ -889,9 +892,10 @@ func (s *Scanner) scanPhase(
 		}
 
 		// Handle symlinks and directories
-		if handled := s.recordSpecialEntry(
-			filePath, info, existingFiles, collector, result); handled {
-			return nil
+		handled, err := s.recordSpecialEntry(
+			filePath, info, existingFiles, collector, result)
+		if handled {
+			return err
 		}
 
 		// Skip other non-regular files (devices, sockets, etc.)
@@ -933,22 +937,25 @@ 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)
-		if file != nil {
-			existingFiles[filePath] = struct{}{}
-			collector.addToProcess(filePath, info, file)
-			s.updateScanEntryStats(result, true, info)
+		file, err := s.buildSymlinkEntry(filePath, info)
+		if err != nil {
+			return true, s.handleWalkError(filePath, err)
 		}
 
-		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 +965,10 @@ 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 +1118,12 @@ 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(path string, info os.FileInfo) *database.File {
+func (s *Scanner) buildSymlinkEntry(
+	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
+		return nil, err
 	}
 
 	var uid, gid uint32
@@ -1136,7 +1142,7 @@ 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.
@@ -1289,6 +1295,11 @@ 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 {
@@ -1306,6 +1317,13 @@ 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)
@@ -1342,13 +1360,32 @@ func (s *Scanner) processPhase(
 	return s.finalizeProcessPhase(ctx, result)
 }
 
-// processFileWithErrorHandling wraps processFileStreaming with error recovery for
-// deleted files and skip-errors mode. Returns (skipped, error).
+// processFileWithErrorHandling records a directory or symlink, or wraps
+// 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
@@ -1362,7 +1399,7 @@ func (s *Scanner) processFileWithErrorHandling(
 			log.Warn("File was deleted during backup, skipping",
 				"path", fileToProcess.Path)
 
-			result.FilesSkipped++
+			countFailedFile(fileToProcess, result)
 
 			return true, nil
 		}
@@ -1373,7 +1410,7 @@ 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
 		}
@@ -1384,6 +1421,13 @@ 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(
@@ -1509,19 +1553,23 @@ func (s *Scanner) handleBlobReady(
 	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)
 	if err != nil {
-		s.cleanupBlobTempFile(blobWithReader)
+		cleanupErr := s.cleanupBlobTempFile(blobWithReader)
 
-		return err
+		return errors.Join(err, cleanupErr)
 	}
 
-	s.cleanupBlobTempFile(blobWithReader)
+	err = s.cleanupBlobTempFile(blobWithReader)
+	if err != nil {
+		return err
+	}
 
 	// Chunks from this blob are now committed to DB - remove from pending set
 	s.removePendingChunkHashes(blobWithReader.InsertedChunkHashes)
@@ -1715,20 +1763,22 @@ 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
@@ -1752,16 +1802,11 @@ 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))
@@ -1979,6 +2024,14 @@ 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

@@ -115,20 +115,37 @@ func ShortHostname(hostname string) string {
 
 // 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) {
 	shortHostname := ShortHostname(hostname)
 
-	// 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)
-	} else {
-		snapshotID = fmt.Sprintf("%s_%s", shortHostname, timestamp)
+
+	for {
+		// Build snapshot ID with optional name
+		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{
@@ -875,7 +892,7 @@ 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

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
-// by passing its own constructor, so the contract is defined once.
+// file://, TestS3Storer against s3://, TestRcloneStorer against rclone's
+// 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

@@ -171,6 +171,11 @@ 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

@@ -50,9 +50,10 @@ 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
-// interrupted write is never listed or trusted as a blob; it is otherwise
-// harmless and is overwritten when the same key is written again.
+// blob. The rclone backend's upload does the same on remotes with a
+// server-side move. List and ListStream skip these files, so a leftover from
+// 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 +242,42 @@ 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{

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/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
-	_, 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
+	return r.upload(ctx, key, bytes.NewReader(buf))
 }
 
 // PutWithProgress stores data with progress reporting.
@@ -89,13 +83,7 @@ func (r *RcloneStorer) PutWithProgress(
 		callback: progress,
 	}
 
-	// Upload the object
-	_, 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
+	return r.upload(ctx, key, pr)
 }
 
 // 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
-// runs entirely in-process.
+// local filesystem directly without any configured remote, so they run
+// 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/storer.go

@@ -71,6 +71,12 @@ 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")
-	ErrEmptyFilePath     = errors.New("file URL path is empty")
+	ErrEmptyStorageURL  = errors.New("storage URL 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
-		if path == "" {
-			return nil, ErrEmptyFilePath
-		}
-
-		return &URL{
-			Scheme: schemeFile,
-			Prefix: path,
-		}, nil
+		return parseFileURL(after)
 	}
 
 	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/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/destination_validation_test.go

@@ -257,6 +257,10 @@ 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/helpers.go

@@ -3,6 +3,7 @@ package vaultik
 import (
 	"errors"
 	"fmt"
+	"math"
 	"regexp"
 	"strconv"
 	"strings"
@@ -47,6 +48,10 @@ const (
 	year  = 365 * day
 )
 
+// maxDuration is the largest value a time.Duration can hold, about 292
+// years.
+const maxDuration = time.Duration(math.MaxInt64)
+
 // A snapshot ID split on "_" has at least a hostname and a trailing
 // timestamp.
 const minSnapshotIDParts = 2
@@ -84,22 +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) {
@@ -145,10 +134,10 @@ func parseSnapshotName(snapshotID, hostname string) string {
 // 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
-// units the input must be whole numbers each followed directly by a unit
-// (2w3d), so 1.0y and 30 days are errors; decimals work only in Go units
-// (1.5h).
+// an unknown unit, a negative value, and a total longer than maxDuration are
+// all rejected. Outside the Go units the input must be whole numbers each
+// followed directly by a unit (2w3d), so 1.0y and 30 days are errors;
+// decimals work only in Go units (1.5h).
 func parseDuration(s string) (time.Duration, error) {
 	if strings.HasPrefix(strings.TrimSpace(s), "-") {
 		return 0, errNegativeDuration
@@ -181,19 +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

@@ -111,6 +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 +177,11 @@ 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
@@ -380,6 +379,10 @@ func (v *Vaultik) collectReferencedBlobsFromManifests(
 	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
 		}
 
@@ -395,7 +398,7 @@ func (v *Vaultik) collectReferencedBlobsFromManifests(
 			continue
 		}
 
-		info.BlobCount = manifest.BlobCount
+		blobCount := manifest.BlobCount
 
 		var blobsSize int64
 
@@ -404,7 +407,8 @@ func (v *Vaultik) collectReferencedBlobsFromManifests(
 			blobsSize += blob.CompressedSize
 		}
 
-		info.BlobsSize = blobsSize
+		info.BlobCount = &blobCount
+		info.BlobsSize = &blobsSize
 	}
 
 	return referencedBlobs, unreadable
@@ -446,23 +450,16 @@ func (v *Vaultik) scanRemoteBlobStorage(
 		v.stdoutf("Scanning blobs...\n")
 	}
 
-	blobCh := v.Storage.ListStream(v.ctx, "blobs/")
-	allBlobs := make(map[string]int64)
+	// Use prune's listing, so a file under blobs/ that prune would never
+	// delete is not counted as a blob, and so not as an orphaned one.
+	allBlobs, err := v.listAllRemoteBlobs()
+	if err != nil {
+		return err
+	}
 
-	for obj := range blobCh {
-		if obj.Err != nil {
-			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
+	for _, size := range allBlobs {
 		result.TotalBlobCount++
-		result.TotalBlobSize += obj.Size
+		result.TotalBlobSize += size
 	}
 
 	// A blob named only by a manifest that could not be read, or by one
@@ -516,13 +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)),
-				ubytes(info.BlobsSize),
+				blobCount,
+				blobsSize,
 			)
 		}
 

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",

internal/vaultik/nuke_remote_test.go

@@ -0,0 +1,64 @@
+package vaultik_test
+
+import (
+	"context"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/snapshot"
+)
+
+// 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, _ := 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)
+}

internal/vaultik/prune.go

@@ -23,9 +23,14 @@ var errNukeRequiresForce = errors.New(
 // 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
-// the next backup starts from scratch.
+// storage, along with any object an upload cut off part-way left under a
+// temporary `.partial` name. 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.
@@ -48,6 +53,15 @@ func (v *Vaultik) NukeRemote(force bool) error {
 		return fmt.Errorf("pruning blobs: %w", err)
 	}
 
+	// 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)
+		}
+	}
+
 	v.UI.Completef("Backup destination store is now empty.")
 
 	return nil

internal/vaultik/prune_count_test.go

@@ -42,7 +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))
 }
 
@@ -57,7 +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_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_per_name_test.go

@@ -267,6 +267,29 @@ 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()

internal/vaultik/remote_info_test.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"strings"
 	"testing"
 	"time"
 
@@ -69,6 +70,74 @@ func TestRemoteInfo_UnreadableManifestLeavesOrphansUnknown(t *testing.T) {
 	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
@@ -116,6 +185,39 @@ func TestRemoteInfo_SkipsNonConformingMetadataName(t *testing.T) {
 	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

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)
@@ -262,6 +304,7 @@ func TestRemoveSnapshot_DefaultRemovesMetadataNotBlobs(t *testing.T) {
 	addBlob(t, store, blobShared)
 
 	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 +353,84 @@ 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()
+
+	tv := vaultik.NewForTesting(newTestStorer())
+	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()

internal/vaultik/restore_interrupt_test.go

@@ -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)
@@ -206,3 +206,119 @@ func TestRestoreSkipErrorsCancelDuringBlobDownload(t *testing.T) {
 	assert.NotContains(t, out.String(), srcPath,
 		"the cancel was reported as a failed file")
 }
+
+// stallingBlobStorer wraps a Storer so that every blob download hands
+// over the first half of the blob and then stalls, ignoring the context,
+// as a file:// blob on a hung USB stick or network share does. Only
+// closing the returned reader ends the stall.
+type stallingBlobStorer struct {
+	storage.Storer
+
+	once    sync.Once
+	stalled chan struct{}
+}
+
+func newStallingBlobStorer(inner storage.Storer) *stallingBlobStorer {
+	return &stallingBlobStorer{Storer: inner, stalled: make(chan struct{})}
+}
+
+func (s *stallingBlobStorer) Get(
+	ctx context.Context, key string,
+) (io.ReadCloser, error) {
+	rc, err := s.Storer.Get(ctx, key)
+	if err != nil || !strings.HasPrefix(key, "blobs/") {
+		return rc, err
+	}
+
+	blob, 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(blob[:len(blob)/2])
+
+		s.once.Do(func() { close(s.stalled) })
+	}()
+
+	return pr, nil
+}
+
+// TestInterruptDuringStalledBlobReadCleansTempDir stops a restore and a
+// deep verify while a blob read 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
+// (issue #325).
+//
+// 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 TestInterruptDuringStalledBlobReadCleansTempDir(t *testing.T) {
+	log.Initialize(log.Config{})
+
+	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 name, op := range operations {
+		t.Run(name, 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 := newStallingBlobStorer(storer)
+
+			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 a blob read")
+			}
+
+			require.NotEmpty(t, scratchEntries(t, scratch),
+				"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 blob 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/snapshot.go

@@ -23,12 +23,19 @@ var (
 	errNoSnapshotsInConfig  = errors.New("no snapshots configured")
 	errSnapshotVerifyFailed = errors.New("verification failed")
 	errRemoveAllNeedsForce  = errors.New("--all requires --force")
+	errSnapshotNotFound     = errors.New("snapshot not found")
 	errInvalidTableName     = errors.New("invalid table name")
 	errNoAgeRecipients      = errors.New(
 		"creating a snapshot needs at least one public key in " +
 			"age_recipients (generate a keypair with: age-keygen)")
+	errKeepNewerThanNeedsPrune = errors.New("--keep-newer-than requires --prune")
 )
 
+// ErrPostBackupPruneFailed wraps a failure of the post-backup prune,
+// which runs only after every snapshot was created and uploaded, so the
+// caller can report it apart from a failed backup.
+var ErrPostBackupPruneFailed = errors.New("post-backup prune failed")
+
 // listRecentLimit caps how many snapshot rows are fetched from the
 // local index for listing and purge operations.
 const listRecentLimit = 10000
@@ -44,6 +51,26 @@ type SnapshotCreateOptions struct {
 	Snapshots     []string // Optional list of snapshot names to process (empty = all)
 }
 
+// Validate fails when KeepNewerThan is set without Prune or is not a
+// valid duration. Only the post-backup prune reads KeepNewerThan, after
+// the backup, so the caller checks it before starting one.
+func (opts *SnapshotCreateOptions) Validate() error {
+	if opts.KeepNewerThan == "" {
+		return nil
+	}
+
+	if !opts.Prune {
+		return errKeepNewerThanNeedsPrune
+	}
+
+	_, err := parseDuration(opts.KeepNewerThan)
+	if err != nil {
+		return fmt.Errorf("--keep-newer-than: %w", err)
+	}
+
+	return nil
+}
+
 // CreateSnapshot executes the snapshot creation operation
 func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 	// config.Load accepts an empty list, since listing, verifying and
@@ -122,7 +149,7 @@ func (v *Vaultik) CreateSnapshot(opts *SnapshotCreateOptions) error {
 	if opts.Prune {
 		err = v.runPostBackupPrune(snapshotNames, opts.KeepNewerThan)
 		if err != nil {
-			return fmt.Errorf("post-backup prune: %w", err)
+			return fmt.Errorf("%w: %w", ErrPostBackupPruneFailed, err)
 		}
 	}
 
@@ -185,6 +212,7 @@ type snapshotStats struct {
 	totalBlobs         int
 	totalBytesSkipped  int64
 	totalFilesSkipped  int
+	totalFilesFailed   int
 	totalFilesDeleted  int
 	totalBytesDeleted  int64
 	totalBytesUploaded int64
@@ -315,6 +343,7 @@ func (v *Vaultik) scanAllDirectories(
 		stats.totalChunks += result.ChunksCreated
 		stats.totalBlobs += result.BlobsCreated
 		stats.totalFilesSkipped += result.FilesSkipped
+		stats.totalFilesFailed += result.FilesFailed
 		stats.totalBytesSkipped += result.BytesSkipped
 		stats.totalFilesDeleted += result.FilesDeleted
 		stats.totalBytesDeleted += result.BytesDeleted
@@ -326,6 +355,7 @@ func (v *Vaultik) scanAllDirectories(
 			"path", dir,
 			"files", result.FilesScanned,
 			"files_skipped", result.FilesSkipped,
+			"files_failed", result.FilesFailed,
 			"bytes", result.BytesScanned,
 			"bytes_skipped", result.BytesSkipped,
 			"chunks", result.ChunksCreated,
@@ -359,9 +389,11 @@ func (v *Vaultik) finalizeSnapshotMetadata(
 		return fmt.Errorf("getting snapshot blob sizes: %w", err)
 	}
 
+	// file_count and total_size leave out the files that could not be
+	// stored; stats.totalBytes already does.
 	extStats := snapshot.ExtendedBackupStats{
 		BackupStats: snapshot.BackupStats{
-			FilesScanned:  stats.totalFiles,
+			FilesScanned:  stats.totalFiles - stats.totalFilesFailed,
 			TotalSize:     stats.totalBytes + stats.totalBytesSkipped,
 			ChunksCreated: stats.totalChunks,
 			BlobsCreated:  stats.totalBlobs,
@@ -409,7 +441,8 @@ func (v *Vaultik) printSnapshotSummary(
 	snapshotID string, startTime time.Time, stats *snapshotStats,
 ) {
 	snapshotDuration := time.Since(startTime)
-	totalFilesChanged := stats.totalFiles - stats.totalFilesSkipped
+	totalFilesChanged := stats.totalFiles - stats.totalFilesSkipped -
+		stats.totalFilesFailed
 	totalBytesAll := stats.totalBytes + stats.totalBytesSkipped
 
 	var compressionRatio float64
@@ -426,6 +459,10 @@ func (v *Vaultik) printSnapshotSummary(
 		v.UI.Count(stats.totalFiles),
 		v.UI.Count(totalFilesChanged),
 		v.UI.Count(stats.totalFilesSkipped))
+	if stats.totalFilesFailed > 0 {
+		filesMsg += fmt.Sprintf(", %s failed", v.UI.Count(stats.totalFilesFailed))
+	}
+
 	if stats.totalFilesDeleted > 0 {
 		filesMsg += fmt.Sprintf(", %s deleted", v.UI.Count(stats.totalFilesDeleted))
 	}
@@ -596,7 +633,7 @@ func (v *Vaultik) confirmAndExecutePurge(
 			v.stdoutf("  %s (%s, %s)\n",
 				snap.ID,
 				snap.Timestamp.Format("2006-01-02 15:04:05"),
-				formatBytes(snap.CompressedSize))
+				ubytes(snap.CompressedSize))
 		}
 	}
 
@@ -633,7 +670,7 @@ func (v *Vaultik) confirmAndExecutePurge(
 				"snapshot_id", snapshotID, "error", err)
 		}
 
-		err = v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID))
+		_, err = v.deleteRemoteSnapshotByKey(snapshot.RemoteSnapshotKey(snapshotID))
 		if err != nil {
 			return fmt.Errorf("deleting snapshot %s from remote: %w", snapshotID, err)
 		}
@@ -1143,6 +1180,11 @@ func (v *Vaultik) RemoveSnapshot(
 		return result, err
 	}
 
+	inLocalIndex, err := v.findSnapshotToRemove(snapshotID, opts)
+	if err != nil {
+		return result, err
+	}
+
 	if opts.DryRun {
 		return v.removeSnapshotDryRun(result, snapshotID, opts)
 	}
@@ -1151,11 +1193,13 @@ func (v *Vaultik) RemoveSnapshot(
 		return result, nil
 	}
 
-	log.Info("Removing snapshot from local database", "snapshot_id", snapshotID)
+	if inLocalIndex {
+		log.Info("Removing snapshot from local database", "snapshot_id", snapshotID)
 
-	err = v.deleteSnapshotFromLocalDB(snapshotID)
-	if err != nil {
-		return result, fmt.Errorf("removing from local database: %w", err)
+		err = v.deleteSnapshotFromLocalDB(snapshotID)
+		if err != nil {
+			return result, fmt.Errorf("removing from local database: %w", err)
+		}
 	}
 
 	if !opts.LocalOnly {
@@ -1174,9 +1218,11 @@ func (v *Vaultik) RemoveSnapshot(
 		return result, v.outputRemoveJSON(result)
 	}
 
-	v.stdoutf("Removed snapshot '%s' from local database\n", snapshotID)
+	if inLocalIndex {
+		v.stdoutf("Removed snapshot '%s' from local database\n", snapshotID)
+	}
 
-	if !opts.LocalOnly && result.RemoteRemoved {
+	if result.RemoteRemoved {
 		v.printlnStdout("Removed snapshot metadata from remote storage")
 		v.stdoutf("\nNote: The removed snapshot's blobs remain on the remote. "+
 			"Run '%s' to delete any blobs no longer referenced by any "+
@@ -1186,6 +1232,45 @@ func (v *Vaultik) RemoveSnapshot(
 	return result, nil
 }
 
+// findSnapshotToRemove reports whether the local index has a row for
+// snapshotID, and returns errSnapshotNotFound when there is nothing to
+// remove: no row and, unless LocalOnly is set, no metadata on the
+// destination store either. The destination store is consulted only
+// when the local index has no row, so a snapshot in the local index can
+// still be removed from it while the destination store is unreachable.
+func (v *Vaultik) findSnapshotToRemove(
+	snapshotID string, opts *RemoveOptions,
+) (bool, error) {
+	if v.Repositories != nil {
+		snap, err := v.Repositories.Snapshots.GetByID(v.ctx, snapshotID)
+		if err != nil {
+			return false, fmt.Errorf("looking up snapshot in local index: %w", err)
+		}
+
+		if snap != nil {
+			return true, nil
+		}
+	}
+
+	if opts.LocalOnly {
+		return false, fmt.Errorf("%w in the local index: %s",
+			errSnapshotNotFound, snapshotID)
+	}
+
+	objects, err := v.remoteSnapshotObjects(snapshot.RemoteSnapshotKey(snapshotID))
+	if err != nil {
+		return false, fmt.Errorf("looking up snapshot on destination store: %w", err)
+	}
+
+	if len(objects) == 0 {
+		return false, fmt.Errorf(
+			"%w in the local index or on the destination store: %s",
+			errSnapshotNotFound, snapshotID)
+	}
+
+	return false, nil
+}
+
 // removeSnapshotDryRun reports what RemoveSnapshot would do without
 // making changes.
 func (v *Vaultik) removeSnapshotDryRun(
@@ -1245,14 +1330,14 @@ func (v *Vaultik) confirmRemoveSnapshot(snapshotID string, opts *RemoveOptions)
 // didn't finish and to run `vaultik snapshot remove` for the snapshot
 // again once the destination store is reachable (`vaultik prune` never
 // removes snapshot metadata). Returns true when the remote removal
-// succeeded.
+// succeeded and deleted something.
 func (v *Vaultik) removeSnapshotRemote(snapshotID string, opts *RemoveOptions) bool {
 	log.Info("Removing snapshot metadata from remote storage",
 		"snapshot_id", snapshotID)
 
 	remoteKey := snapshot.RemoteSnapshotKey(snapshotID)
 
-	err := v.deleteRemoteSnapshotByKey(remoteKey)
+	deleted, err := v.deleteRemoteSnapshotByKey(remoteKey)
 	if err != nil {
 		log.Warn("Could not remove snapshot metadata from remote storage; "+
 			"run '"+snapshotRemoveCommandHint+"' with the snapshot's ID "+
@@ -1270,7 +1355,7 @@ func (v *Vaultik) removeSnapshotRemote(snapshotID string, opts *RemoveOptions) b
 		return false
 	}
 
-	return true
+	return deleted > 0
 }
 
 // RemoveAllSnapshots removes every snapshot known to the local
@@ -1562,7 +1647,7 @@ func (v *Vaultik) removeAllLocalSnapshots(
 		}
 
 		if !opts.LocalOnly {
-			err = v.deleteRemoteSnapshotByKey(
+			_, err = v.deleteRemoteSnapshotByKey(
 				snapshot.RemoteSnapshotKey(snapshotID))
 			if err != nil {
 				log.Warn("Failed to remove snapshot metadata from remote",
@@ -1588,7 +1673,7 @@ func (v *Vaultik) removeAllOrphanRemotes(
 	for _, key := range orphanRemoteKeys {
 		log.Info("Removing orphan remote snapshot", "remote_key", key)
 
-		err := v.deleteRemoteSnapshotByKey(key)
+		_, err := v.deleteRemoteSnapshotByKey(key)
 		if err != nil {
 			log.Warn("Failed to remove orphan from remote",
 				"remote_key", key, "error", err)
@@ -1635,34 +1720,44 @@ func (v *Vaultik) deleteSnapshotFromLocalDB(snapshotID string) error {
 }
 
 // deleteRemoteSnapshotByKey removes everything under
-// metadata/<remoteKey>/ on the destination store. The argument is a
-// remote key (double-SHA256 derivation), not a human snapshot ID;
-// callers that have a human ID must hash via snapshot.RemoteSnapshotKey
-// first.
-func (v *Vaultik) deleteRemoteSnapshotByKey(remoteKey string) error {
-	prefix := fmt.Sprintf("metadata/%s/", remoteKey)
-	objectCh := v.Storage.ListStream(v.ctx, prefix)
-
-	var objectsToDelete []string
-
-	for object := range objectCh {
-		if object.Err != nil {
-			return fmt.Errorf("listing objects: %w", object.Err)
-		}
-
-		objectsToDelete = append(objectsToDelete, object.Key)
+// metadata/<remoteKey>/ on the destination store and returns how many
+// objects it deleted. The argument is a remote key (double-SHA256
+// derivation), not a human snapshot ID; callers that have a human ID
+// must hash via snapshot.RemoteSnapshotKey first.
+func (v *Vaultik) deleteRemoteSnapshotByKey(remoteKey string) (int, error) {
+	objectsToDelete, err := v.remoteSnapshotObjects(remoteKey)
+	if err != nil {
+		return 0, err
 	}
 
-	for _, key := range objectsToDelete {
-		err := v.Storage.Delete(v.ctx, key)
+	for i, key := range objectsToDelete {
+		err = v.Storage.Delete(v.ctx, key)
 		if err != nil {
-			return fmt.Errorf("removing %s: %w", key, err)
+			return i, fmt.Errorf("removing %s: %w", key, err)
 		}
 
 		log.Debug("Deleted remote object", "key", key)
 	}
 
-	return nil
+	return len(objectsToDelete), nil
+}
+
+// remoteSnapshotObjects lists the keys of every object under
+// metadata/<remoteKey>/ on the destination store.
+func (v *Vaultik) remoteSnapshotObjects(remoteKey string) ([]string, error) {
+	prefix := fmt.Sprintf("metadata/%s/", remoteKey)
+
+	var keys []string
+
+	for object := range v.Storage.ListStream(v.ctx, prefix) {
+		if object.Err != nil {
+			return nil, fmt.Errorf("listing objects: %w", object.Err)
+		}
+
+		keys = append(keys, object.Key)
+	}
+
+	return keys, nil
 }
 
 // outputRemoveJSON outputs the removal result as JSON
@@ -1756,9 +1851,9 @@ func (v *Vaultik) PruneDatabase() (*PruneResult, error) {
 	return result, nil
 }
 
-// countUnknown is what a count reads as when its query could not be run,
-// distinct from "0", which means the table really was empty.
-const countUnknown = "unknown"
+// unknownText is what a count or size reads as when it could not be
+// determined, distinct from "0", which is a real zero.
+const unknownText = "unknown"
 
 // tableCountForReport returns the row count of a table for the prune
 // summary, or nil if the count could not be read. A read failure is
@@ -1795,7 +1890,7 @@ func countDiff(before, after *int64) *int64 {
 // one that could not be queried.
 func countText(count *int64) string {
 	if count == nil {
-		return countUnknown
+		return unknownText
 	}
 
 	return strconv.FormatInt(*count, 10)

internal/vaultik/snapshot_failed_file_test.go

@@ -0,0 +1,123 @@
+package vaultik_test
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/config"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/ui"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// openFailFs is the real filesystem, except that opening path fails
+// with err. Phase 1 of a backup only lstats a file, so it still counts
+// path; phase 2 is the first to open it.
+type openFailFs struct {
+	afero.OsFs
+
+	path string
+	err  error
+}
+
+//nolint:ireturn // afero.Fs.Open is defined to return the interface.
+func (f *openFailFs) Open(name string) (afero.File, error) {
+	if name == f.path {
+		return nil, &os.PathError{Op: "open", Path: name, Err: f.err}
+	}
+
+	return f.OsFs.Open(name)
+}
+
+// A file that phase 2 cannot open is reported as failed, not as
+// unchanged, and neither the summary's data total nor the snapshots row
+// counts it. See https://git.eeqj.de/sneak/vaultik/issues/280.
+func TestSnapshotSummaryCountsFileNotStoredAsFailed(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	tests := []struct {
+		name       string
+		openErr    error
+		skipErrors bool
+	}{
+		// What a normal user gets opening a file with mode 000.
+		{name: "unopenable under skip-errors",
+			openErr: os.ErrPermission, skipErrors: true},
+		// What opening a file removed after phase 1 gives.
+		{name: "removed between the phases",
+			openErr: os.ErrNotExist, skipErrors: false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			ctx := context.Background()
+
+			// The scan walks the source path with symlinks resolved, so
+			// failedPath must be spelled the same way to match.
+			tempDir, err := filepath.EvalSymlinks(t.TempDir())
+			require.NoError(t, err)
+
+			srcDir := filepath.Join(tempDir, "src")
+			failedPath := filepath.Join(srcDir, "failed.txt")
+			storedContent := []byte("this file is backed up")
+			storedSize := int64(len(storedContent))
+
+			fs := &openFailFs{path: failedPath, err: tt.openErr}
+			require.NoError(t, fs.MkdirAll(srcDir, 0o755))
+			require.NoError(t, afero.WriteFile(fs,
+				filepath.Join(srcDir, "stored.txt"), storedContent, 0o644))
+			require.NoError(t, afero.WriteFile(fs,
+				failedPath, []byte("this file cannot be opened"), 0o644))
+
+			cfg := faultTestConfig()
+			cfg.IndexPath = filepath.Join(tempDir, "index.sqlite")
+			cfg.Snapshots = map[string]config.SnapshotConfig{
+				"src": {Paths: []string{srcDir}},
+			}
+
+			store, err := storage.NewFileStorer(filepath.Join(tempDir, "remote"))
+			require.NoError(t, err)
+
+			db, err := database.New(ctx, cfg.IndexPath)
+			require.NoError(t, err)
+			t.Cleanup(func() { _ = db.Close() })
+
+			repos := database.NewRepositories(db)
+			out := &bytes.Buffer{}
+			v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+			v.UI = ui.NewWithColor(out, false)
+
+			require.NoError(t, v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+				SkipErrors: tt.skipErrors,
+				Snapshots:  []string{"src"},
+			}))
+
+			summary := out.String()
+			assert.Contains(t, summary,
+				"Files: 2 examined, 1 backed up, 0 unchanged, 1 failed.")
+			assert.Contains(t, summary,
+				fmt.Sprintf("Data: %s total (%s backed up).",
+					v.UI.Size(storedSize), v.UI.Size(storedSize)))
+
+			snap, err := repos.Snapshots.GetByID(ctx,
+				localSnapshotID(ctx, t, repos, "src"))
+			require.NoError(t, err)
+			require.NotNil(t, snap)
+
+			assert.Equal(t, int64(1), snap.FileCount)
+			assert.Equal(t, storedSize, snap.TotalSize)
+		})
+	}
+}

internal/vaultik/snapshot_list.go

@@ -553,8 +553,8 @@ func (v *Vaultik) printSnapshotTable(snapshots []SnapshotInfo) error {
 
 		if snap.LocallyTracked {
 			id = snap.ID.String()
-			uncompressed = formatBytes(snap.UncompressedSize)
-			newChunks = formatBytes(snap.NewChunkSize)
+			uncompressed = ubytes(snap.UncompressedSize)
+			newChunks = ubytes(snap.NewChunkSize)
 		} else {
 			id = formatRemoteOnlyID(snap.RemoteKey)
 			uncompressed = remoteOnlyCell
@@ -564,7 +564,7 @@ func (v *Vaultik) printSnapshotTable(snapshots []SnapshotInfo) error {
 		_, err = fmt.Fprintf(w, "%s\t%s\t%s\t%s\t%s\n",
 			id,
 			snap.Timestamp.Format("2006-01-02 15:04:05"),
-			formatBytes(snap.CompressedSize),
+			ubytes(snap.CompressedSize),
 			uncompressed,
 			newChunks)
 		if err != nil {

internal/vaultik/snapshot_list_test.go

@@ -212,8 +212,8 @@ const (
 	listConfiguredName = "home"
 	listLocalID        = "testhost_home_2026-03-01T10:00:00Z"
 	listRemoteID       = "otherhost_media_2026-03-02T11:22:33Z"
-	// fiveMegabytes formats as "5.0 MB" through formatBytes.
-	fiveMegabytes = 5 * 1024 * 1024
+	// fiveMegabytes is printed as "5.0 MB" by every command.
+	fiveMegabytes = 5_000_000
 )
 
 // TestListSnapshots_RemoteWithoutSecretKey is the regression guard for
@@ -306,6 +306,48 @@ func TestListSnapshots_RemoteOnlyRowRendering(t *testing.T) {
 		"are not in the local index")
 }
 
+// TestSnapshotListAndPurgePrintSizesLikeRemoteInfo checks that the
+// `snapshot list` table and the purge listing print a byte count the way
+// `remote info` does: in powers of 1000, as the MB label says.
+func TestSnapshotListAndPurgePrintSizesLikeRemoteInfo(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	const wantSize = "5.0 MB" // fiveMegabytes
+
+	env := newListEnv(t)
+
+	// A remote-only snapshot: `snapshot list` and `remote info` both take
+	// its size from the manifest.
+	remoteKey := env.addRemote(t, listRemoteID,
+		time.Date(2026, 3, 2, 11, 22, 33, 0, time.UTC))
+
+	// A locally tracked snapshot of the same size for the purge to list.
+	// The purge prints the snapshot's blob size, the last argument here.
+	start := time.Date(2026, 3, 1, 10, 0, 0, 0, time.UTC)
+	env.addLocal(t, listLocalID, start)
+	env.addRemote(t, listLocalID, start)
+	require.NoError(t, env.v.Repositories.Snapshots.UpdateCounts(
+		context.Background(), nil, listLocalID, 0, 0, 0, 0, fiveMegabytes))
+
+	require.NoError(t, env.v.RemoteInfo(false))
+	assert.Contains(t, findTableRow(t, env.stdout.String(), remoteKey[:16]),
+		wantSize, "remote info")
+
+	env.stdout.Reset()
+	require.NoError(t, env.v.ListSnapshots(false))
+	assert.Contains(t, findTableRow(t, env.stdout.String(),
+		"<remote only:"+remoteKey[:12]+">"), wantSize, "snapshot list")
+
+	// Without --force and with nothing on stdin the purge lists what it
+	// would delete, reads no confirmation, and deletes nothing.
+	env.stdout.Reset()
+	require.NoError(t, env.v.PurgeSnapshotsWithOptions(
+		&vaultik.SnapshotPurgeOptions{OlderThan: "1d"}))
+	assert.Contains(t, findTableRow(t, env.stdout.String(), listLocalID),
+		wantSize, "snapshot purge")
+}
+
 // findTableRow returns the single output line containing needle.
 func findTableRow(t *testing.T, out, needle string) string {
 	t.Helper()

internal/vaultik/snapshot_skip_errors_test.go

@@ -0,0 +1,75 @@
+package vaultik_test
+
+import (
+	"context"
+	"fmt"
+	"path/filepath"
+	"testing"
+
+	"github.com/spf13/afero"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/vaultik/internal/config"
+	"sneak.berlin/go/vaultik/internal/database"
+	"sneak.berlin/go/vaultik/internal/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/vaultik"
+)
+
+// --skip-errors skips only a file that cannot be opened or read. A local
+// index error while a directory is recorded stops the backup, and the
+// snapshot is not recorded as complete. See
+// https://git.eeqj.de/sneak/vaultik/issues/284.
+func TestSkipErrorsBackupStopsOnIndexErrorRecordingDirectory(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+
+	// The scan walks the source path with symlinks resolved, so dirPath
+	// must be spelled the same way to match the row the scan inserts.
+	tempDir, err := filepath.EvalSymlinks(t.TempDir())
+	require.NoError(t, err)
+
+	srcDir := filepath.Join(tempDir, "src")
+	dirPath := filepath.Join(srcDir, "dir")
+
+	fs := afero.NewOsFs()
+	require.NoError(t, fs.MkdirAll(dirPath, 0o755))
+	require.NoError(t, afero.WriteFile(fs,
+		filepath.Join(dirPath, "file.txt"), []byte("file content"), 0o644))
+
+	cfg := faultTestConfig()
+	cfg.IndexPath = filepath.Join(tempDir, "index.sqlite")
+	cfg.Snapshots = map[string]config.SnapshotConfig{
+		"tree": {Paths: []string{srcDir}},
+	}
+
+	store, err := storage.NewFileStorer(filepath.Join(tempDir, "remote"))
+	require.NoError(t, err)
+
+	db, err := database.New(ctx, cfg.IndexPath)
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	// The local index refuses the directory's files row.
+	_, err = db.Conn().ExecContext(ctx, fmt.Sprintf(`
+		CREATE TRIGGER refuse_directory BEFORE INSERT ON files
+		WHEN NEW.path = '%s'
+		BEGIN SELECT RAISE(ABORT, 'simulated index error'); END`, dirPath))
+	require.NoError(t, err)
+
+	repos := database.NewRepositories(db)
+	v := newBackupVaultik(ctx, cfg, store, repos, db, fs)
+
+	err = v.CreateSnapshot(&vaultik.SnapshotCreateOptions{
+		SkipErrors: true,
+		Snapshots:  []string{"tree"},
+	})
+	require.ErrorContains(t, err, "simulated index error")
+
+	snapshots, err := repos.Snapshots.ListRecent(ctx, listRecentTestLimit)
+	require.NoError(t, err)
+	require.Len(t, snapshots, 1)
+	assert.Nil(t, snapshots[0].CompletedAt)
+}

internal/vaultik/storage_bind.go

@@ -9,7 +9,7 @@ import (
 )
 
 // errStorageBindingMismatch is returned when the local index database is
-// bound to a different destination than the configured storage_url.
+// bound to a different destination than the configured one.
 var errStorageBindingMismatch = errors.New(
 	"local index is bound to a different backup destination")
 
@@ -24,8 +24,9 @@ var errStorageBindingMismatch = errors.New(
 //
 // Behaviour:
 //   - On first use (empty stored value), record the configured
-//     storage_url and log the binding.
-//   - When the stored value matches the configured storage_url, do
+//     destination (storage_url, or the s3.* endpoint, bucket and
+//     prefix) and log the binding.
+//   - When the stored value matches the configured destination, do
 //     nothing and return nil.
 //   - When the two differ, refuse with an error that tells the user
 //     how to recover (revert the config, or run `vaultik database
@@ -46,12 +47,10 @@ func (v *Vaultik) EnsureStorageBinding() error {
 
 	configured := v.Config.StorageURL
 	if configured == "" {
-		// Some legacy configs still use the split s3.* keys instead of
-		// storage_url. Falling back to a synthetic URL for those would
-		// hide the fact that the binding is loose. Instead, treat
-		// unset as "nothing to bind against" — the check is
-		// necessarily best-effort for those configs.
-		return nil
+		// The s3.* form: bind on its endpoint, bucket and prefix,
+		// written the way storage_url would name the same destination.
+		s3 := v.Config.S3
+		configured = "s3://" + s3.Bucket + "/" + s3.Prefix + "?endpoint=" + s3.Endpoint
 	}
 
 	stored, err := v.Repositories.LocalMeta.Get(v.ctx, database.LocalMetaKeyStorageURL)
@@ -96,7 +95,8 @@ func buildBindingMismatchMessage(stored, configured string) string {
 		"future snapshots referencing blobs that don't exist at the new destination.\n" +
 		"\n" +
 		"To proceed, either:\n" +
-		"  - revert storage_url in your config to the bound destination, or\n" +
+		"  - revert storage_url (or the s3.* endpoint, bucket and prefix) in your\n" +
+		"    config to the bound destination, or\n" +
 		"  - run 'vaultik database delete' to discard the local index and rebuild it\n" +
 		"    from a fresh full backup against the new destination"
 }

internal/vaultik/storage_bind_test.go

@@ -76,3 +76,51 @@ func TestEnsureStorageBinding_MismatchRefuses(t *testing.T) {
 	assert.Contains(t, err.Error(), "file:///mnt/backups/new")
 	assert.Contains(t, err.Error(), "vaultik database delete")
 }
+
+// A config with no storage_url names its destination with s3.endpoint,
+// s3.bucket and s3.prefix; a change to any of them must be refused just
+// as a changed storage_url is.
+func TestEnsureStorageBinding_S3FieldsChangeRefuses(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	bound := config.S3Config{
+		Endpoint: "https://s3.example.com",
+		Bucket:   "bucket-a",
+		Prefix:   "backups",
+	}
+
+	tests := []struct {
+		name    string
+		changed config.S3Config
+	}{
+		{"bucket", config.S3Config{
+			Endpoint: bound.Endpoint, Bucket: "bucket-b", Prefix: bound.Prefix,
+		}},
+		{"prefix", config.S3Config{
+			Endpoint: bound.Endpoint, Bucket: bound.Bucket, Prefix: "other",
+		}},
+		{"endpoint", config.S3Config{
+			Endpoint: "https://s3.other.com", Bucket: bound.Bucket, Prefix: bound.Prefix,
+		}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+
+			tv, _ := buildBindTestVaultik(t, "")
+
+			tv.Config.S3 = bound
+			require.NoError(t, tv.EnsureStorageBinding(), "first use binds")
+			require.NoError(t, tv.EnsureStorageBinding(), "same destination passes")
+
+			tv.Config.S3 = tt.changed
+			err := tv.EnsureStorageBinding()
+			require.Error(t, err)
+			assert.Contains(t, err.Error(),
+				"local index is bound to a different backup destination")
+			assert.Contains(t, err.Error(), "vaultik database delete")
+		})
+	}
+}