libimage: do not fail disk usage when an image is removed mid-walk

common/libimage/corrupted_test.go

@@ -91,3 +91,29 @@ func TestCorruptedLayers(t *testing.T) {
 	_, err = image.Inspect(ctx, nil)
 	require.NoError(t, err, "inspecting repaired image should work")
 }
+
+func TestDiskUsageRemovedImage(t *testing.T) {
+	// An image can be removed by a parallel process between DiskUsage's
+	// listing and the per-image walk. The stale handle then fails with
+	// "image not known", which is not a corruption. DiskUsage skips it;
+	// that requires the wrapped error to keep its type.
+	runtime := testNewRuntime(t)
+	ctx := context.Background()
+	pullOptions := &PullOptions{}
+	pullOptions.Writer = os.Stdout
+
+	pulledImages, err := runtime.Pull(ctx, "quay.io/libpod/alpine_nginx:latest", config.PullPolicyAlways, pullOptions)
+	require.NoError(t, err)
+	require.Len(t, pulledImages, 1)
+	staleImage := pulledImages[0]
+
+	_, errs := runtime.store.DeleteImage(staleImage.ID(), true)
+	require.Nil(t, errs, "deleting the image from the store")
+
+	err = staleImage.isCorrupted(ctx, "")
+	require.Error(t, err, "stale handle reports an error")
+	require.ErrorIs(t, err, storage.ErrImageUnknown, "the wrapped error keeps its type")
+
+	_, _, err = runtime.DiskUsage(ctx)
+	require.NoError(t, err, "disk usage does not fail because of a removed image")
+}

common/libimage/disk_usage.go

@@ -4,6 +4,7 @@ package libimage
 
 import (
 	"context"
+	"errors"
 	"time"
 
 	"github.com/sirupsen/logrus"
@@ -73,6 +74,12 @@ func (r *Runtime) DiskUsage(ctx context.Context) ([]ImageDiskUsage, int64, error
 	for _, image := range images {
 		usages, err := diskUsageForImage(ctx, image, layerMap, layerCount, &totalSize)
 		if err != nil {
+			// The image can be removed by a parallel process between the
+			// listing above and reading it here. That is not a corruption,
+			// skip it instead of failing the whole disk usage.
+			if errors.Is(err, storage.ErrImageUnknown) {
+				continue
+			}
 			return nil, -1, err
 		}
 		allUsages = append(allUsages, usages...)

common/libimage/image.go

@@ -102,7 +102,7 @@ func (i *Image) isCorrupted(ctx context.Context, name string) error {
 		if name == "" {
 			name = i.ID()[:12]
 		}
-		return fmt.Errorf("Image %s exists in local storage but may be corrupted (remove the image to resolve the issue): %v", name, err)
+		return fmt.Errorf("Image %s exists in local storage but may be corrupted (remove the image to resolve the issue): %w", name, err)
 	}
 	return img.Close()
 }