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

Dockerfile

@@ -5,7 +5,16 @@
 FROM golangci/golangci-lint:v2.14.0@sha256:ad862ba6b3798cbe0fd9fd7408d498fd74fbd2623a92406b2fd3898faf0bf98f AS lint
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Without the time limit a stalled connection to the module proxy holds
+# this step, and every build on the host sharing it, until the kernel
+# drops the connection, which can take an hour. The download normally
+# takes under half a minute.
+RUN for attempt in 1 2 3; do \
+        timeout 180 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 180s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 # `golangci-lint run` silently ignores an unknown top-level key in
 # .golangci.yml, such as a misspelt `linters:`; `config verify` fails on it.
@@ -18,7 +27,13 @@ RUN golangci-lint run --config .golangci.yml ./...
 FROM golang:1.26.1@sha256:cd78d88e00afadbedd272f977d375a6247455f3a4b1178f8ae8bbcb201743a8a AS test
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 180 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 180s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 RUN go test -timeout 90s -race -cover ./... || \
     { echo "--- Rerunning with -v for details ---"; \
@@ -36,7 +51,13 @@ RUN apk add --no-cache git
 RUN git config --system --add safe.directory /src
 WORKDIR /src
 COPY go.mod go.sum ./
-RUN go mod download
+# Limited and retried as in the lint phase above.
+RUN for attempt in 1 2 3; do \
+        timeout 180 go mod download && exit 0; \
+        echo "go mod download attempt $attempt failed or ran over 180s" >&2; \
+    done; \
+    echo "module download failed: giving up after 3 attempts" >&2; \
+    exit 1
 COPY . .
 
 # The VERSION build arg when one is given, otherwise

cmd/vaultik/dockermoddownload_test.go

@@ -0,0 +1,45 @@
+package main_test
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// TestDockerfileBoundsEveryModuleDownload fails unless every
+// `go mod download` in the Dockerfile runs under a time limit, is tried
+// a fixed number of times, and then fails the build with a message that
+// names the module download (issue #351). Like the version tests, this
+// parses the Dockerfile; that a stalled download really is cut off and
+// retried was checked by hand.
+func TestDockerfileBoundsEveryModuleDownload(t *testing.T) {
+	t.Parallel()
+
+	var downloads []string
+
+	for _, instruction := range instructions(t, productDockerfile) {
+		if strings.Contains(instruction, "go mod download") {
+			downloads = append(downloads, instruction)
+		}
+	}
+
+	require.NotEmpty(t, downloads, "%s must download modules",
+		productDockerfile)
+
+	for _, download := range downloads {
+		assert.Contains(t, download, "timeout 180 go mod download",
+			"%s: a module download must have a time limit: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "for attempt in 1 2 3",
+			"%s: a module download must be retried: %s",
+			productDockerfile, download)
+		assert.Contains(t, download, "module download failed",
+			"%s: a module download that never succeeds must say so: %s",
+			productDockerfile, download)
+		assert.True(t, strings.HasSuffix(download, "exit 1"),
+			"%s: a module download that never succeeds must fail the"+
+				" build: %s", productDockerfile, download)
+	}
+}

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
 	}
 

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()
 

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/vaultik/helpers.go

@@ -89,22 +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) {

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"),
-				formatBytes(snap.CompressedSize))
+				ubytes(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 = 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()