Limit and retry the module download in the Docker build (closes #351)

Each `go mod download` in the Dockerfile now runs under a 180-second
time limit and is tried up to three times. Without the limit, a stalled
connection to the module proxy held the step, and every build on the
host sharing it, until the kernel dropped the connection. When all
three attempts fail, the step prints "module download failed" and
exits 1, so the build stops there and does not look like a test or
lint failure. The loop is repeated in each of the three stages because
a Dockerfile cannot share shell code between stages. A test checks
every download step for the limit, the retries, the message and the
final `exit 1`.

Judgement call: 180 seconds and three attempts, against a normal
download of 13 to 24 seconds.
Deviation: REPO_POLICIES.md's model Go Dockerfile still shows a bare
`RUN go mod download`.

Model: opus-5-5

internal/database/repositories.go

@@ -3,7 +3,6 @@ package database
 import (
 	"context"
 	"database/sql"
-	"errors"
 	"fmt"
 )
 
@@ -61,24 +60,22 @@ func (r *Repositories) WithTx(ctx context.Context, fn TxFunc) error {
 
 	defer func() {
 		if p := recover(); p != nil {
-			// The panic is the failure to report, so a rollback error
-			// is dropped.
-			_ = tx.Rollback()
+			rollbackErr := tx.Rollback()
+			if rollbackErr != nil {
+				Fatalf("failed to rollback transaction: %v", rollbackErr)
+			}
 
 			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

@@ -4,7 +4,6 @@ import (
 	"context"
 	"database/sql"
 	"errors"
-	"strings"
 	"testing"
 	"time"
 
@@ -236,125 +235,6 @@ 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()
 

internal/log/tty_handler.go

@@ -10,8 +10,6 @@ import (
 	"sync"
 	"time"
 	"unicode"
-
-	"github.com/dustin/go-humanize"
 )
 
 // groupSeparator joins an open group path to an attribute key. This
@@ -306,13 +304,18 @@ func formatDuration(d time.Duration) string {
 	}
 }
 
-// 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.
+// formatBytes formats bytes in a human-readable way
 func formatBytes(b int64) string {
-	if b < 0 {
+	const unit = 1024
+	if b < unit {
 		return fmt.Sprintf("%d B", b)
 	}
 
-	return humanize.Bytes(uint64(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])
 }

internal/log/tty_handler_test.go

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

internal/vaultik/blob_fetch.go

@@ -103,8 +103,6 @@ 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) {
@@ -122,47 +120,5 @@ func (v *Vaultik) FetchBlob(
 		return nil, fmt.Errorf("downloading blob %s: %w", shortHash(blobHash), err)
 	}
 
-	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()
+	return rc, nil
 }

internal/vaultik/helpers.go

@@ -89,6 +89,22 @@ 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) {

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 (restore's and deep verify's) currently present in dir.
+// scratch entries 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-verify-*",
+		"vaultik-blobcache-*", "vaultik-restore-*",
 	} {
 		found, err := filepath.Glob(filepath.Join(dir, pattern))
 		require.NoError(t, err)
@@ -206,119 +206,3 @@ 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

@@ -633,7 +633,7 @@ func (v *Vaultik) confirmAndExecutePurge(
 			v.stdoutf("  %s (%s, %s)\n",
 				snap.ID,
 				snap.Timestamp.Format("2006-01-02 15:04:05"),
-				ubytes(snap.CompressedSize))
+				formatBytes(snap.CompressedSize))
 		}
 	}
 

internal/vaultik/snapshot_list.go

@@ -553,8 +553,8 @@ func (v *Vaultik) printSnapshotTable(snapshots []SnapshotInfo) error {
 
 		if snap.LocallyTracked {
 			id = snap.ID.String()
-			uncompressed = ubytes(snap.UncompressedSize)
-			newChunks = ubytes(snap.NewChunkSize)
+			uncompressed = formatBytes(snap.UncompressedSize)
+			newChunks = formatBytes(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"),
-			ubytes(snap.CompressedSize),
+			formatBytes(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 is printed as "5.0 MB" by every command.
-	fiveMegabytes = 5_000_000
+	// fiveMegabytes formats as "5.0 MB" through formatBytes.
+	fiveMegabytes = 5 * 1024 * 1024
 )
 
 // TestListSnapshots_RemoteWithoutSecretKey is the regression guard for
@@ -306,48 +306,6 @@ 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()