Name each prefix once in restore, verify and purge errors (closes #327)

A missing blob during restore, a failed shallow verify and an invalid
purge duration each printed their prefix twice. FetchBlob now returns
the storage error as is, since restore and deep verify name the blob
themselves; purge returns the duration parser's error unchanged; and a
failed shallow verify wraps "problems found in storage" instead of
"verification failed", which the command already prints. Deep verify
words a storage error other than not found as a blob it could not
check, and its log line no longer says S3.

Already fixed for #314: the
deep verify size message naming S3 and other errors called missing;
tests now check both.
Not tested: the log line.

Model: opus-5-5

go.mod

@@ -22,6 +22,7 @@ require (
 	github.com/spf13/cobra v1.10.1
 	github.com/spf13/pflag v1.0.10
 	github.com/stretchr/testify v1.11.1
+	go.uber.org/dig v1.19.0
 	go.uber.org/fx v1.24.0
 	golang.org/x/sync v0.18.0
 	golang.org/x/sys v0.38.0
@@ -256,7 +257,6 @@ require (
 	go.opentelemetry.io/otel/metric v1.38.0 // indirect
 	go.opentelemetry.io/otel/trace v1.38.0 // indirect
 	go.shabbyrobe.org/gocovmerge v0.0.0-20230507111327-fa4f82cfbf4d // indirect
-	go.uber.org/dig v1.19.0 // indirect
 	go.uber.org/multierr v1.11.0 // indirect
 	go.uber.org/zap v1.27.0 // indirect
 	go.yaml.in/yaml/v2 v2.4.3 // indirect

internal/cli/app.go

@@ -8,12 +8,12 @@ import (
 	"errors"
 	"fmt"
 	"path/filepath"
-	"strings"
 	"sync"
 	"time"
 
 	"github.com/adrg/xdg"
 	"github.com/spf13/cobra"
+	"go.uber.org/dig"
 	"go.uber.org/fx"
 	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/database"
@@ -146,16 +146,12 @@ func (e *startupError) Error() string {
 //	could not build arguments for function "Y" (file:line): failed to build U:
 //	received non-nil error from function "Z" (file:line): <real error>
 //
-// Users care about the real error, not the DI plumbing. We strip everything
+// Users care about the real error, not the DI plumbing. dig.RootCause,
-// up through the last "): " (which is always the close-paren of an fx
+// from the library fx is built on, unwraps those layers and returns the
-// function-location annotation followed by the wrapped error).
+// real error whole. Cutting the text at "): " instead would also cut a
+// real error that contains it, such as the blob_size_limit one.
 func cleanStartupError(err error) error {
-	msg := err.Error()
+	return &startupError{msg: dig.RootCause(err).Error()}
-	if idx := strings.LastIndex(msg, "): "); idx >= 0 {
-		msg = msg[idx+3:]
-	}
-
-	return &startupError{msg: msg}
 }
 
 // RunApp starts the fx application, blocks until it is asked to stop, and

internal/cli/app_test.go

@@ -2,58 +2,65 @@ package cli //nolint:testpackage // needs access to unexported cleanStartupError
 
 import (
 	"errors"
+	"os"
 	"path/filepath"
 	"testing"
 
+	"go.uber.org/fx"
+	"sneak.berlin/go/vaultik/internal/config"
 	"sneak.berlin/go/vaultik/internal/pidlock"
 )
 
+// TestCleanStartupError starts a real fx app whose config has a
+// blob_size_limit below the minimum. fx puts annotations naming the
+// functions it was building in front of the config error; only those
+// may be removed, although the config error itself contains "): ".
 func TestCleanStartupError(t *testing.T) {
 	t.Parallel()
 
-	tests := []struct {
+	path := filepath.Join(t.TempDir(), "config.yml")
-		name string
+
-		in   string
+	err := os.WriteFile(path, []byte("snapshots:\n"+
-		want string
+		"  test:\n"+
-	}{
+		"    paths: [/tmp/vaultik-test-source]\n"+
-		{
+		"storage_url: file:///tmp/vaultik-test-storage\n"+
-			name: "real fx error chain",
+		"chunk_size: 10MB\n"+
-			in: `could not build arguments for function ` +
+		"blob_size_limit: 30MB\n"), 0o600)
-				`"sneak.berlin/go/vaultik/internal/cli".newSnapshotCreateCommand.func1.1 ` +
+	if err != nil {
-				`(/Users/user/dev/vaultik/internal/cli/snapshot.go:71): ` +
+		t.Fatalf("write config: %v", err)
-				`failed to build *vaultik.Vaultik: ` +
-				`could not build arguments for function ` +
-				`"sneak.berlin/go/vaultik/internal/vaultik".New ` +
-				`(/Users/user/dev/vaultik/internal/vaultik/vaultik.go:59): ` +
-				`failed to build storage.Storer: ` +
-				`received non-nil error from function ` +
-				`"sneak.berlin/go/vaultik/internal/storage".NewStorer ` +
-				`(/Users/user/dev/vaultik/internal/storage/module.go:23): ` +
-				`creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-			want: `creating base path: mkdir /Volumes/BACKUPS: permission denied`,
-		},
-		{
-			name: "no fx wrapping",
-			in:   "plain error",
-			want: "plain error",
-		},
-		{
-			name: "single fx wrapping",
-			in:   `received non-nil error from function "foo" (file.go:1): underlying problem`,
-			want: "underlying problem",
-		},
 	}
 
-	for _, tt := range tests {
+	app := fx.New(
-		t.Run(tt.name, func(t *testing.T) {
+		fx.NopLogger,
-			t.Parallel()
+		fx.Supply(config.Path(path)),
+		config.Module,
+		fx.Invoke(func(*config.Config) {}),
+	)
 
-			//nolint:err113 // test constructs errors from table input
+	err = app.Start(t.Context())
-			got := cleanStartupError(errors.New(tt.in)).Error()
+	if err == nil {
-			if got != tt.want {
+		t.Fatal("app started; want a blob_size_limit error")
-				t.Errorf("got %q, want %q", got, tt.want)
+	}
-			}
+
-		})
+	want := "failed to load config: invalid config: blob_size_limit must " +
+		"be at least the largest chunk the chunker can emit (chunk_size " +
+		"times the FastCDC size spread): need at least 40000000 bytes"
+
+	got := cleanStartupError(err).Error()
+	if got != want {
+		t.Errorf("got %q, want %q", got, want)
+	}
+}
+
+// TestCleanStartupErrorNotFromFx checks that an error fx did not wrap is
+// shown unchanged.
+func TestCleanStartupErrorNotFromFx(t *testing.T) {
+	t.Parallel()
+
+	//nolint:err113 // a one-off error standing in for a startup failure
+	got := cleanStartupError(errors.New("plain error")).Error()
+	if got != "plain error" {
+		t.Errorf("got %q, want %q", got, "plain error")
 	}
 }
 

internal/vaultik/blob_fetch.go

@@ -104,7 +104,8 @@ 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.
+// blocked inside it. A storage error is returned as is: restore and deep
+// verify name the blob in their own errors.
 func (v *Vaultik) FetchBlob(
 	ctx context.Context, blobHash string,
 ) (io.ReadCloser, error) {
@@ -119,7 +120,7 @@ func (v *Vaultik) FetchBlob(
 
 	rc, err := v.Storage.Get(ctx, blobPath)
 	if err != nil {
-		return nil, fmt.Errorf("downloading blob %s: %w", shortHash(blobHash), err)
+		return nil, err
 	}
 
 	return newCloseOnCancelReader(ctx, rc), nil

internal/vaultik/blobcache.go

@@ -39,9 +39,9 @@ type blobDiskCacheEntry struct {
 // internally and will fall back to LRU eviction if curBytes exceeds
 // maxBytes. Restore passes math.MaxInt64 as maxBytes and drives
 // eviction itself via Delete() through restoreSweeper, which deletes
-// each blob the moment every file that references its chunks has been
+// each blob once no file still to be restored in this run needs it.
-// written. LRU never fires under that configuration; it is kept as a
+// LRU never fires under that configuration; it is kept as a safety net
-// safety net for callers that don't manage eviction themselves.
+// for callers that don't manage eviction themselves.
 //
 // Get/ReadAt/peak-Len counters are debugging instrumentation used by
 // tests to assert that the restore code path uses ReadAt rather than

internal/vaultik/deep_verify_test.go

@@ -6,6 +6,7 @@ import (
 	"crypto/rand"
 	"io"
 	"os"
+	"path"
 	"path/filepath"
 	"testing"
 
@@ -142,7 +143,8 @@ func TestDeepVerifyJSONCountsBlobs(t *testing.T) {
 
 // TestDeepVerifyJSONCountsWrongSizeBlob checks that under --deep --json a
 // blob of the wrong size is counted in mismatched and fails the run, and
-// that the check goes on past it to count a missing blob after it.
+// that the check goes on past it to count a missing blob after it. The
+// store is on disk, so the message must not call it S3.
 func TestDeepVerifyJSONCountsWrongSizeBlob(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -168,11 +170,14 @@ func TestDeepVerifyJSONCountsWrongSizeBlob(t *testing.T) {
 	require.Equal(t, 1, got.Mismatched)
 	require.Equal(t, 1, got.Missing)
 	require.Zero(t, got.Verified)
+	require.Contains(t, got.ErrorMessage, "with the wrong size")
+	require.NotContains(t, got.ErrorMessage, "S3")
 }
 
 // TestDeepVerifyStopsOnBlobThatCannotBeChecked checks that a blob that is
 // present in storage but cannot be checked is not counted as missing:
-// deep verify stops and fails with the storage error.
+// deep verify stops and fails, saying it could not check the blob, with
+// the storage error.
 func TestDeepVerifyStopsOnBlobThatCannotBeChecked(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -187,6 +192,8 @@ func TestDeepVerifyStopsOnBlobThatCannotBeChecked(t *testing.T) {
 	require.Error(t, err)
 	require.Equal(t, "failed", got.Status)
 	require.Zero(t, got.Missing)
+	require.Contains(t, got.ErrorMessage,
+		"could not check blob "+path.Base(keys[0]))
 	require.Contains(t, got.ErrorMessage, errStatDenied.Error())
 }
 

internal/vaultik/prune_count_test.go

@@ -1,13 +1,17 @@
 package vaultik //nolint:testpackage // exercises unexported count helpers
 
 import (
+	"bytes"
 	"context"
 	"testing"
+	"time"
 
 	"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/ui"
 )
 
 // TestTableCountForReportSurfacesReadFailure is the regression guard for
@@ -76,3 +80,59 @@ func TestCountDiffUnknownWhenEitherSideUnknown(t *testing.T) {
 	assert.Nil(t, countDiff(&before, nil), "unknown after yields unknown delta")
 	assert.Nil(t, countDiff(nil, nil))
 }
+
+// TestPruneDatabaseSummaryLabelsTotals removes the only incomplete
+// snapshot from an index that also holds a complete one. The figure in
+// brackets is every snapshot left in the local index, so the line must
+// say so rather than read as incomplete snapshots that remain.
+func TestPruneDatabaseSummaryLabelsTotals(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	ctx := context.Background()
+
+	db, err := database.New(ctx, ":memory:")
+	require.NoError(t, err)
+	t.Cleanup(func() { _ = db.Close() })
+
+	repos := database.NewRepositories(db)
+
+	startedAt := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	completedAt := startedAt.Add(time.Minute)
+
+	require.NoError(t, repos.Snapshots.Create(ctx, nil, &database.Snapshot{
+		ID:          "host_complete_2026-01-02T03:04:05Z",
+		Hostname:    "host",
+		StartedAt:   startedAt,
+		CompletedAt: &completedAt,
+	}))
+	require.NoError(t, repos.Snapshots.Create(ctx, nil, &database.Snapshot{
+		ID:        "host_incomplete_2026-01-02T03:04:05Z",
+		Hostname:  "host",
+		StartedAt: startedAt,
+	}))
+
+	var out bytes.Buffer
+
+	v := &Vaultik{
+		DB:           db,
+		Repositories: repos,
+		SnapshotManager: snapshot.NewSnapshotManager(
+			snapshot.SnapshotManagerParams{Repos: repos}),
+		UI: ui.NewWithColor(&out, false),
+	}
+	v.SetContext(ctx)
+
+	result, err := v.PruneDatabase()
+	require.NoError(t, err)
+	require.Equal(t, int64(1), result.SnapshotsDeleted)
+
+	assert.Contains(t, out.String(), "Incomplete snapshots: 1 removed "+
+		"(total snapshots left in the local index: 1).")
+	assert.Contains(t, out.String(), "Orphaned files: 0 removed "+
+		"(total files left in the local index: 0).")
+	assert.Contains(t, out.String(), "Orphaned chunks: 0 removed "+
+		"(total chunks left in the local index: 0).")
+	assert.Contains(t, out.String(), "Orphaned blobs: 0 removed "+
+		"(total blobs left in the local index: 0).")
+}

internal/vaultik/purge_per_name_test.go

@@ -281,7 +281,8 @@ func TestPurgeOlderThan_TooLargeDeletesNothing(t *testing.T) {
 		OlderThan: "1000y",
 		Force:     true,
 	})
-	require.Error(t, err)
+	require.EqualError(t, err,
+		`invalid duration: "1000y" is longer than about 292 years`)
 
 	err = v.PurgeSnapshotsWithOptions(&vaultik.SnapshotPurgeOptions{
 		OlderThan: "292y",

internal/vaultik/restore.go

@@ -323,12 +323,6 @@ func (v *Vaultik) restoreAllFiles(
 		_ = blobCache.Close()
 	}()
 
-	// Per-restore sweep state: every blob_size_limit/100 bytes written,
-	// scan the cache and delete any blob whose remaining file references
-	// are all already restored.
-	sweeper := newRestoreSweeper(v.ctx, repos, blobCache,
-		v.Config.BlobSizeLimit.Int64()/sweepIntervalDivisor)
-
 	blobByHash, blobIDToHash, err := v.buildBlobIndexes(repos)
 	if err != nil {
 		return nil, err
@@ -340,6 +334,12 @@ func (v *Vaultik) restoreAllFiles(
 		return nil, fmt.Errorf("building restore plan: %w", err)
 	}
 
+	// Per-restore sweep state: every blob_size_limit/100 bytes written,
+	// scan the cache and delete any blob that no pending file in the
+	// plan references.
+	sweeper := newRestoreSweeper(plan, blobCache,
+		v.Config.BlobSizeLimit.Int64()/sweepIntervalDivisor)
+
 	// Ranging over files rather than the map reports them in a stable order.
 	for _, file := range files {
 		if planErr, ok := unplanned[file.ID]; ok {
@@ -427,10 +427,8 @@ func (v *Vaultik) runRestoreLoop(
 			continue
 		}
 
-		// Record the file as restored so the sweeper can free blobs
+		// Drop the file from the plan's indexes so future picks ignore
-		// once all referencing files are done, and drop it from the
+		// it and the sweeper can free blobs no pending file needs.
-		// plan's indexes so future picks ignore it.
-		session.sweeper.fileRestored(fileID.String())
 		plan.finishFile(fileID)
 
 		processed++

internal/vaultik/restore_interrupt_test.go

@@ -74,7 +74,8 @@ func TestRestoreCleansTempDirOnInterrupt(t *testing.T) {
 	buildLocalityFixture(t, fs, dataDir)
 
 	cfg, storer, snapshotID := setupLocalityBackup(
-		context.Background(), t, fs, dataDir, storeDir, dbPath)
+		context.Background(), t, fs, dataDir, storeDir, dbPath,
+		localitySourceBytes)
 
 	// Point the "" temp paths (the blob cache directory and the
 	// snapshot-database directory) at a private directory so the test can

internal/vaultik/restore_locality_test.go

@@ -63,6 +63,10 @@ type localityCopy struct {
 	data []byte
 }
 
+// localitySourceBytes is the size of each source file of the locality
+// fixture.
+const localitySourceBytes = 1024 * 1024
+
 // buildLocalityFixture writes the adversarial source layout described
 // in TestRestoreLocalityAndReadAt: 15 sources of 1 MiB each (which the
 // backup packs into 3 blobs of 5 chunks) plus 9 interleaved-name copies
@@ -73,7 +77,6 @@ func buildLocalityFixture(
 	t.Helper()
 
 	const (
-		srcBytes   = 1024 * 1024
 		srcCount   = 15
 		blobsCount = 3
 		perBlob    = srcCount / blobsCount
@@ -83,7 +86,7 @@ func buildLocalityFixture(
 	for i := range srcCount {
 		s := &localitySource{
 			path: fmt.Sprintf("src-%03d.bin", i+1),
-			data: randomBytes(t, srcBytes),
+			data: randomBytes(t, localitySourceBytes),
 		}
 		sources[i] = s
 		require.NoError(t, afero.WriteFile(
@@ -115,19 +118,20 @@ func buildLocalityFixture(
 
 // setupLocalityBackup runs the backup half of the locality test: a
 // snapshot of dataDir into a file storer with sizes tuned so 15 one-chunk
-// files pack into 3 blobs. The index database is closed before returning
+// files of srcBytes pack into 3 blobs. The index database is closed
-// so the restore half runs from remote bytes only.
+// before returning so the restore half runs from remote bytes only.
 func setupLocalityBackup(
 	ctx context.Context, t *testing.T, fs afero.Fs,
-	dataDir, storeDir, dbPath string,
+	dataDir, storeDir, dbPath string, srcBytes int64,
 ) (*config.Config, *storage.FileStorer, string) {
 	t.Helper()
 
-	// chunkSize avg = 4 MiB makes minSize = 1 MiB, so a 1 MiB file
+	// An average chunk size of four times srcBytes makes the minimum
-	// becomes one chunk. maxBlobSize = 5 MiB packs exactly 5 chunks
+	// chunk srcBytes, so each source file becomes one chunk. A blob limit
-	// per blob, yielding 3 blobs from 15 source files.
+	// of five times srcBytes packs exactly 5 chunks per blob, yielding 3
-	chunkSize := int64(4 * 1024 * 1024)
+	// blobs from 15 source files.
-	maxBlobSize := int64(5 * 1024 * 1024)
+	chunkSize := 4 * srcBytes
+	maxBlobSize := 5 * srcBytes
 
 	storer, err := storage.NewFileStorer(storeDir)
 	require.NoError(t, err)
@@ -208,7 +212,7 @@ func TestRestoreLocalityAndReadAt(t *testing.T) {
 	ctx := context.Background()
 
 	cfg, storer, snapshotID := setupLocalityBackup(
-		ctx, t, fs, dataDir, storeDir, dbPath)
+		ctx, t, fs, dataDir, storeDir, dbPath, localitySourceBytes)
 
 	// Wrap the storer so we can count downloads per blob key.
 	counter := newCountingStorer(storer)

internal/vaultik/restore_plan.go

@@ -31,7 +31,8 @@ var (
 //     with no further downloads.
 //   - blobFiles: for each blob, the set of pending files referencing
 //     it. Used to short-circuit "when this blob lands, which files
-//     become ready" without a global scan.
+//     become ready" without a global scan, and by the sweeper to tell
+//     whether a cached blob is still needed.
 type restorePlan struct {
 	fileBlobs map[types.FileID]map[string]struct{}
 	blobFiles map[string]map[types.FileID]struct{}
@@ -237,6 +238,12 @@ func (p *restorePlan) filesReferencingBlob(blobHash string) []types.FileID {
 	return out
 }
 
+// pendingFileUsesBlob reports whether any pending file references the
+// named blob, whether or not the blob is cached.
+func (p *restorePlan) pendingFileUsesBlob(blobHash string) bool {
+	return len(p.blobFiles[blobHash]) > 0
+}
+
 // hasPending reports whether any unfinished files remain.
 func (p *restorePlan) hasPending() bool {
 	return len(p.fileBlobs) > 0

internal/vaultik/restore_plan_test.go

@@ -63,15 +63,6 @@ func TestRunRestoreLoopFailsOnAbandonedFiles(t *testing.T) {
 
 	t.Cleanup(func() { _ = cache.Close() })
 
-	v := &Vaultik{ctx: ctx}
-	session := &restoreSession{
-		v:       v,
-		ctx:     ctx,
-		repos:   repos,
-		sweeper: newRestoreSweeper(ctx, repos, cache, 1),
-		result:  &RestoreResult{},
-	}
-
 	// A file that is still pending but whose uncached-blob set is empty
 	// and which was never queued as ready: the loop can neither restore
 	// nor download it. This is the abandonment the guard must catch.
@@ -83,6 +74,15 @@ func TestRunRestoreLoopFailsOnAbandonedFiles(t *testing.T) {
 		cached:    map[string]struct{}{},
 	}
 
+	v := &Vaultik{ctx: ctx}
+	session := &restoreSession{
+		v:       v,
+		ctx:     ctx,
+		repos:   repos,
+		sweeper: newRestoreSweeper(plan, cache, 1),
+		result:  &RestoreResult{},
+	}
+
 	err = v.runRestoreLoop(session, plan, map[types.FileID]*database.File{}, 0)
 	require.ErrorIs(t, err, errRestoreIncomplete)
 }

internal/vaultik/restore_skip_errors_test.go

@@ -231,7 +231,8 @@ func (f trailingByteFile) Close() error {
 }
 
 // TestRestoreMissingBlobAbortsWithoutSkipErrors checks that a deleted blob
-// still ends the restore with an error when SkipErrors is not set.
+// still ends the restore with an error when SkipErrors is not set, and
+// that the error names the blob once.
 func TestRestoreMissingBlobAbortsWithoutSkipErrors(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -247,6 +248,8 @@ func TestRestoreMissingBlobAbortsWithoutSkipErrors(t *testing.T) {
 	})
 
 	require.ErrorIs(t, err, storage.ErrNotFound)
+	assert.Equal(t, 1, strings.Count(err.Error(), "downloading blob"),
+		err.Error())
 }
 
 // TestRestoreChunkInNoBlobAbortsWithoutSkipErrors checks that a chunk held

internal/vaultik/restore_sweeper.go

@@ -1,46 +1,30 @@
 package vaultik
 
-import (
+// restoreSweeper frees cached blobs once no file still to be restored in
-	"context"
+// this run needs them. It works as follows:
-	"fmt"
-
-	"sneak.berlin/go/vaultik/internal/database"
-	"sneak.berlin/go/vaultik/internal/log"
-)
-
-// restoreSweeper frees cached blobs once all files that reference any of
-// their chunks have been restored. It works as follows:
 //
-//  1. Callers add a file's ID to an in-memory restored set via
+//  1. After each chunk is restored, chunkRestored accumulates a running
-//     fileRestored once the file is fully written to disk.
-//  2. After each chunk is restored, chunkRestored accumulates a running
 //     byte count.
-//  3. When the accumulator crosses a threshold (one hundredth of the
+//  2. When the accumulator crosses a threshold (one hundredth of the
 //     configured blob size — so a sweep runs about a hundred times per
 //     blob's worth of restored bytes), the sweeper iterates every key in
-//     the cache. For each cached blob it asks the DB which files
+//     the cache and deletes each blob that the restore plan no longer
-//     reference any chunk in that blob, then compares that list against
+//     needs.
-//     the in-memory restored set. If any referencing file is missing
-//     from the set the blob is kept; otherwise the cache entry is
-//     deleted.
 //
-// All DB reads happen against the snapshot's temporary metadata DB,
+// The plan drops a file once it is restored or has failed, and holds
-// which is local, indexed, and not under contention — the queries are
+// only the files this run restores, so files outside the requested paths
-// cheap and run at most once per blob per sweep interval.
+// never keep a blob.
 type restoreSweeper struct {
-	ctx        context.Context //nolint:containedctx // ctx bound at construction by design
+	plan       *restorePlan
-	repos      *database.Repositories
 	cache      *blobDiskCache
 	threshold  int64
 	bytesAccum int64
-	restored   map[string]struct{}
 }
 
 // newRestoreSweeper returns a sweeper that triggers eviction every
 // `threshold` bytes restored. Callers should pass blob_size_limit/100.
 func newRestoreSweeper(
-	ctx context.Context,
+	plan *restorePlan,
-	repos *database.Repositories,
 	cache *blobDiskCache,
 	threshold int64,
 ) *restoreSweeper {
@@ -49,21 +33,12 @@ func newRestoreSweeper(
 	}
 
 	return &restoreSweeper{
-		ctx:       ctx,
+		plan:      plan,
-		repos:     repos,
 		cache:     cache,
 		threshold: threshold,
-		restored:  make(map[string]struct{}),
 	}
 }
 
-// fileRestored records a file as fully restored. After this call, any
-// blob whose only remaining references come from files in the restored
-// set may be evicted on the next sweep.
-func (s *restoreSweeper) fileRestored(fileID string) {
-	s.restored[fileID] = struct{}{}
-}
-
 // chunkRestored accounts n bytes against the sweep threshold and runs a
 // sweep if the threshold has been crossed since the last sweep.
 func (s *restoreSweeper) chunkRestored(n int64) {
@@ -76,60 +51,11 @@ func (s *restoreSweeper) chunkRestored(n int64) {
 	s.sweep()
 }
 
-// sweep deletes any cached blob whose chunks are no longer referenced
+// sweep deletes every cached blob that no pending file references.
-// by an unrestored file. Per-blob DB failures are logged and the blob
-// is kept — we'd rather hold a blob longer than risk a re-download.
 func (s *restoreSweeper) sweep() {
 	for _, blobHash := range s.cache.Keys() {
-		needed, err := s.blobStillNeeded(blobHash)
+		if !s.plan.pendingFileUsesBlob(blobHash) {
-		if err != nil {
-			log.Debug("sweeper referencing-files query failed",
-				"blob_hash", blobHash[:16], "error", err)
-
-			continue
-		}
-
-		if !needed {
 			s.cache.Delete(blobHash)
 		}
 	}
 }
-
-// blobStillNeeded returns true if any file that references a chunk in
-// this blob has not yet been restored. On any error the function
-// returns true — keeping the blob is always the safe answer because we
-// can't prove we're done with it.
-func (s *restoreSweeper) blobStillNeeded(blobHash string) (bool, error) {
-	rows, err := s.repos.DB().Conn().QueryContext(s.ctx, `
-		SELECT DISTINCT fc.file_id
-		FROM file_chunks fc
-		JOIN blob_chunks bc ON bc.chunk_hash = fc.chunk_hash
-		JOIN blobs b ON b.id = bc.blob_id
-		WHERE b.blob_hash = ?
-	`, blobHash)
-	if err != nil {
-		return true, fmt.Errorf("querying referencing files: %w", err)
-	}
-
-	defer func() { _ = rows.Close() }()
-
-	for rows.Next() {
-		var fileID string
-
-		err := rows.Scan(&fileID)
-		if err != nil {
-			return true, fmt.Errorf("scanning file_id: %w", err)
-		}
-
-		if _, ok := s.restored[fileID]; !ok {
-			return true, nil
-		}
-	}
-
-	err = rows.Err()
-	if err != nil {
-		return true, err
-	}
-
-	return false, nil
-}

internal/vaultik/restore_sweeper_test.go

@@ -0,0 +1,163 @@
+package vaultik //nolint:testpackage // reads the restore blob cache's peak size
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"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/log"
+	"sneak.berlin/go/vaultik/internal/storage"
+	"sneak.berlin/go/vaultik/internal/ui"
+)
+
+// TestRestoreFreesBlobsNoPendingFileNeeds asserts that the restore cache
+// frees a blob once no file still to be restored in this run needs it.
+// Every original file has a copy with the same content, so every blob
+// holds chunks of copies: a restore that leaves the copies out of its
+// paths, or fails on them, must not keep the blobs for them.
+func TestRestoreFreesBlobsNoPendingFileNeeds(t *testing.T) {
+	log.Initialize(log.Config{})
+	t.Parallel()
+
+	// Small files keep the backup and both restores quick.
+	const fileBytes = 16 * 1024
+
+	fs := afero.NewOsFs()
+	tempDir := t.TempDir()
+
+	dataDir := filepath.Join(tempDir, "source")
+	originalsDir := filepath.Join(dataDir, "originals")
+	copiesDir := filepath.Join(dataDir, "copies")
+	storeDir := filepath.Join(tempDir, "remote")
+	dbPath := filepath.Join(tempDir, "index.sqlite")
+
+	originals := writeFilesWithCopies(t, fs, originalsDir, copiesDir, fileBytes)
+
+	ctx := context.Background()
+
+	cfg, storer, snapshotID := setupLocalityBackup(
+		ctx, t, fs, dataDir, storeDir, dbPath, fileBytes)
+
+	t.Run("restore by path", func(t *testing.T) {
+		t.Parallel()
+
+		restoreDir := filepath.Join(tempDir, "by-path")
+
+		peak, err := restorePeakCachedBlobs(ctx, t, fs, cfg, storer,
+			&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  restoreDir,
+				Paths:      []string{originalsDir},
+			})
+		require.NoError(t, err)
+
+		requireRestoredFiles(t, fs, restoreDir, originals)
+		assert.LessOrEqualf(t, peak, 1,
+			"peak cached blobs was %d; blobs were kept for unselected files",
+			peak)
+	})
+
+	t.Run("restore with failed files", func(t *testing.T) {
+		t.Parallel()
+
+		restoreDir := filepath.Join(tempDir, "with-failures")
+
+		// A regular file where the copies directory belongs makes every
+		// copy fail to restore.
+		blocker := filepath.Join(restoreDir, copiesDir)
+		require.NoError(t, fs.MkdirAll(filepath.Dir(blocker), 0o755))
+		require.NoError(t, afero.WriteFile(fs, blocker, nil, 0o644))
+
+		peak, err := restorePeakCachedBlobs(ctx, t, fs, cfg, storer,
+			&RestoreOptions{
+				SnapshotID: snapshotID,
+				TargetDir:  restoreDir,
+				SkipErrors: true,
+			})
+		require.ErrorIs(t, err, errFilesFailedRestore)
+
+		requireRestoredFiles(t, fs, restoreDir, originals)
+		assert.LessOrEqualf(t, peak, 1,
+			"peak cached blobs was %d; blobs were kept for failed files", peak)
+	})
+}
+
+// writeFilesWithCopies writes 15 random files of fileBytes into
+// originalsDir and a copy of each into copiesDir. With
+// setupLocalityBackup given the same size, each file is one chunk and the
+// 15 chunks fill 3 blobs. Returns the originals' content by path.
+func writeFilesWithCopies(
+	t *testing.T, fs afero.Fs, originalsDir, copiesDir string, fileBytes int,
+) map[string][]byte {
+	t.Helper()
+
+	const fileCount = 15
+
+	require.NoError(t, fs.MkdirAll(originalsDir, 0o755))
+	require.NoError(t, fs.MkdirAll(copiesDir, 0o755))
+
+	originals := make(map[string][]byte, fileCount)
+
+	for i := range fileCount {
+		name := fmt.Sprintf("file-%02d.bin", i)
+		data := randomBytes(t, fileBytes)
+		original := filepath.Join(originalsDir, name)
+		originals[original] = data
+
+		require.NoError(t, afero.WriteFile(fs, original, data, 0o644))
+		require.NoError(t, afero.WriteFile(
+			fs, filepath.Join(copiesDir, name), data, 0o644))
+	}
+
+	return originals
+}
+
+// requireRestoredFiles requires every file in want to have been restored
+// under restoreDir with its content intact.
+func requireRestoredFiles(
+	t *testing.T, fs afero.Fs, restoreDir string, want map[string][]byte,
+) {
+	t.Helper()
+
+	for path, data := range want {
+		got, err := afero.ReadFile(fs, filepath.Join(restoreDir, path))
+		require.NoErrorf(t, err, "file missing after restore: %s", path)
+		require.Equalf(t, data, got, "byte mismatch for %s", path)
+	}
+}
+
+// restorePeakCachedBlobs runs a restore and returns the most blobs its
+// cache held at once, together with the restore's error.
+func restorePeakCachedBlobs(
+	ctx context.Context, t *testing.T, fs afero.Fs,
+	cfg *config.Config, storer storage.Storer, opts *RestoreOptions,
+) (int, error) {
+	t.Helper()
+
+	var cache *blobDiskCache
+
+	v := &Vaultik{
+		Config:  cfg,
+		Storage: storer,
+		Fs:      fs,
+		Stdout:  io.Discard,
+		Stderr:  io.Discard,
+		UI:      ui.NewWithColor(io.Discard, false),
+		restoreCacheObserver: func(c *blobDiskCache) {
+			cache = c
+		},
+	}
+	v.SetContext(ctx)
+
+	err := v.Restore(opts)
+
+	require.NotNil(t, cache, "restoreCacheObserver must fire during restore")
+
+	return cache.PeakLen(), err
+}

internal/vaultik/shallow_verify_test.go

@@ -27,7 +27,7 @@ import (
 // then grows one stored blob so its size no longer matches the manifest.
 // Shallow verify must then fail, count the grown blob as a size mismatch,
 // and drop it from the verified count rather than continuing to report it
-// as checked.
+// as checked, and its error must not repeat the cli layer's prefix.
 func TestShallowVerifyDetectsWrongBlobSize(t *testing.T) {
 	log.Initialize(log.Config{})
 	t.Parallel()
@@ -96,6 +96,12 @@ func TestShallowVerifyDetectsWrongBlobSize(t *testing.T) {
 	require.Equal(t, 1, bad.Mismatched)
 	require.Equal(t, healthy.BlobCount-1, bad.Verified,
 		"the wrong-sized blob must not be counted as verified")
+
+	// The cli layer prints this error after its own "Verification failed:".
+	err = newVerifier(io.Discard).VerifySnapshotWithOptions(
+		snapshotID, &vaultik.VerifyOptions{})
+	require.EqualError(t, err,
+		"problems found in storage: 1 blobs have the wrong size")
 }
 
 // TestShallowVerifyDetectsMissingDatabase backs up a real snapshot,

internal/vaultik/snapshot.go

@@ -23,6 +23,7 @@ var (
 	errSnapshotNotInConfig  = errors.New("snapshot not found in config")
 	errNoSnapshotsInConfig  = errors.New("no snapshots configured")
 	errSnapshotVerifyFailed = errors.New("verification failed")
+	errProblemsInStorage    = errors.New("problems found in storage")
 	errRemoveAllNeedsForce  = errors.New("--all requires --force")
 	errSnapshotNotFound     = errors.New("snapshot not found")
 	errSnapshotRemoveFailed = errors.New("snapshot(s) could not be removed")
@@ -612,7 +613,7 @@ func selectSnapshotsToPurge(
 	case opts.OlderThan != "":
 		duration, err := parseDuration(opts.OlderThan)
 		if err != nil {
-			return nil, fmt.Errorf("invalid duration: %w", err)
+			return nil, err
 		}
 
 		cutoff := time.Now().UTC().Add(-duration)
@@ -941,8 +942,10 @@ func (v *Vaultik) formatVerifyResult(
 		v.printVerifySummary(result, failure)
 	}
 
+	// Not errSnapshotVerifyFailed: the cli layer prints this after its own
+	// "Verification failed:".
 	if failure != "" {
-		return fmt.Errorf("%w: %s", errSnapshotVerifyFailed, failure)
+		return fmt.Errorf("%w: %s", errProblemsInStorage, failure)
 	}
 
 	return nil
@@ -1902,13 +1905,17 @@ func (v *Vaultik) PruneDatabase() (*PruneResult, error) {
 	snapshotsRemain := countDiff(snapshotCountBefore, &result.SnapshotsDeleted)
 
 	v.UI.Completef("Pruned local index database.")
-	v.UI.Detailf("Incomplete snapshots: %s removed (%s remain).",
+	v.UI.Detailf("Incomplete snapshots: %s removed "+
+		"(total snapshots left in the local index: %s).",
 		countText(&result.SnapshotsDeleted), countText(snapshotsRemain))
-	v.UI.Detailf("Orphaned files: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned files: %s removed "+
+		"(total files left in the local index: %s).",
 		countText(result.FilesDeleted), countText(fileCountAfter))
-	v.UI.Detailf("Orphaned chunks: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned chunks: %s removed "+
+		"(total chunks left in the local index: %s).",
 		countText(result.ChunksDeleted), countText(chunkCountAfter))
-	v.UI.Detailf("Orphaned blobs: %s removed (%s remain).",
+	v.UI.Detailf("Orphaned blobs: %s removed "+
+		"(total blobs left in the local index: %s).",
 		countText(result.BlobsDeleted), countText(blobCountAfter))
 
 	return result, nil

internal/vaultik/verify.go

@@ -680,7 +680,7 @@ func (v *Vaultik) verifyManifestAgainstDatabase(
 func (v *Vaultik) verifyBlobExistenceFromDB(
 	blobs []snapshot.BlobInfo, result *VerifyResult,
 ) error {
-	log.Info("Verifying blob existence in S3", "blob_count", len(blobs))
+	log.Info("Verifying blob existence in storage", "blob_count", len(blobs))
 
 	for i, blob := range blobs {
 		// The hash is read from the snapshot database, which is not
@@ -700,7 +700,7 @@ func (v *Vaultik) verifyBlobExistenceFromDB(
 			result.Missing++
 			result.MissingSize += blob.CompressedSize
 		case err != nil:
-			return fmt.Errorf("checking blob %s in storage: %w",
+			return fmt.Errorf("could not check blob %s in storage: %w",
 				blob.Hash, err)
 		case stat.Size != blob.CompressedSize:
 			log.Warn("Blob has the wrong size in storage",