make layer-tree lookup errors non-fatal

Internally, Podman constructs a tree of layers in containers/storage to
quickly compute relations among layers and hence images.  To compute the
tree, we intersect all local layers with all local images.  So far,
lookup errors have been fatal which has turned out to be a mistake since
it seems fairly easy to cause storage corruptions, for instance, when
killing builds.  In that case, a (partial) image may list a layer which
does not exist (anymore).  Since the errors were fatal, there was no
easy way to clean up and many commands were erroring out.

To improve usability, turn the fatal errors into warnings that guide the
user into resolving the issue.  In this case, a `podman system reset`
may be the approriate way for now.

[NO TESTS NEEDED] because I have no reliable way to force it.

[1] https://github.com/containers/podman/issues/8148#issuecomment-778253474

Signed-off-by: Valentin Rothberg <rothberg@redhat.com>

1 files changed, 13 insertions, 2 deletions

diff --git a/libpod/image/layer_tree.go b/libpod/image/layer_tree.go
index 18101575e..dde39dba1 100644
--- a/libpod/image/layer_tree.go
+++ b/libpod/image/layer_tree.go
@@ -5,6 +5,7 @@ import (
 
 	ociv1 "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/pkg/errors"
+	"github.com/sirupsen/logrus"
 )
 
 // layerTree is an internal representation of local layers.
@@ -84,7 +85,12 @@ func (ir *Runtime) layerTree() (*layerTree, error) {
 		}
 		node, exists := tree.nodes[topLayer]
 		if !exists {
-			return nil, errors.Errorf("top layer %s of image %s not found in layer tree", img.TopLayer(), img.ID())
+			// Note: erroring out in this case has turned out having been a
+			// mistake. Users may not be able to recover, so we're now
+			// throwing a warning to guide them to resolve the issue and
+			// turn the errors non-fatal.
+			logrus.Warnf("Top layer %s of image %s not found in layer tree. The storage may be corrupted, consider running `podman system reset`.", topLayer, img.ID())
+			continue
 		}
 		node.images = append(node.images, img)
 	}
@@ -107,7 +113,12 @@ func (t *layerTree) children(ctx context.Context, parent *Image, all bool) ([]st
 
 	parentNode, exists := t.nodes[parent.TopLayer()]
 	if !exists {
-		return nil, errors.Errorf("layer not found in layer tree: %q", parent.TopLayer())
+		// Note: erroring out in this case has turned out having been a
+		// mistake. Users may not be able to recover, so we're now
+		// throwing a warning to guide them to resolve the issue and
+		// turn the errors non-fatal.
+		logrus.Warnf("Layer %s not found in layer. The storage may be corrupted, consider running `podman system reset`.", parent.TopLayer())
+		return children, nil
 	}
 
 	parentID := parent.ID()