fix: skip processing ztocs/indexes that are found in the artifacts store but not in the content store

Signed-off-by: Swapnanil-Gupta <swpnlg@amazon.com>

Author: Swapnanil-Gupta

cmd/soci/commands/ztoc/list.go

@@ -18,14 +18,17 @@ package ztoc
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"os"
 	"text/tabwriter"
 
 	"github.com/awslabs/soci-snapshotter/cmd/soci/commands/internal"
 	"github.com/awslabs/soci-snapshotter/soci"
+	"github.com/containerd/containerd/errdefs"
 	"github.com/containerd/containerd/images"
+	"github.com/containerd/log"
 	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 	"github.com/urfave/cli/v3"
 )
@@ -96,7 +99,11 @@ var listCommand = &cli.Command{
 			for _, indexInfo := range indexInfos {
 				readerAt, err := cs.ReaderAt(ctx, indexInfo.Descriptor)
 				if err != nil {
-					return fmt.Errorf("failed to read a soci index with digest %s from the content store (try running \"soci rebuild-db\" first): %w", indexInfo.Descriptor.Digest, err)
+					if errors.Is(err, errdefs.ErrNotFound) {
+						log.G(ctx).Debug("a soci index was found in the artifact store but not in the content store, skipping")
+						continue
+					}
+					return fmt.Errorf("failed to read a soci index with digest %s from the content store: %w", indexInfo.Descriptor.Digest, err)
 				}
 				var sociIndex soci.Index
 				err = soci.DecodeIndex(io.NewSectionReader(readerAt, 0, indexInfo.Descriptor.Size), &sociIndex)

integration/index_test.go

@@ -79,7 +79,7 @@ func prepareCustomSociIndices(t *testing.T, sh *shell.Shell, images []testImageI
 			}
 		}
 		img.imgInfo = dockerhub(imgName, withPlatform(platform))
-		img.sociIndexDigest = buildIndex(sh, img.imgInfo, withIndexBuildConfig(indexBuildConfig), withMinLayerSize(0), withRunRebuildDbBeforeCreate())
+		img.sociIndexDigest = buildIndex(sh, img.imgInfo, withIndexBuildConfig(indexBuildConfig), withMinLayerSize(0))
 		ztocDigests, err := getZtocDigestsForImage(sh, img.imgInfo)
 		if err != nil {
 			t.Fatalf("could not get ztoc digests: %v", err)
@@ -345,7 +345,7 @@ func TestSociIndexRemoveAndRebuildWithSharedLayers(t *testing.T) {
 		// the following 2 images share layers
 		img1 := dockerhub(nginxAlpineImage)
 		img2 := dockerhub(nginxAlpineImage2)
-		imgDigest := buildIndex(sh, img1, withMinLayerSize(0), withRunRebuildDbBeforeCreate())
+		imgDigest := buildIndex(sh, img1, withMinLayerSize(0))
 		if imgDigest == "" {
 			t.Fatal("failed to get soci index for nginx alpine test image")
 		}

integration/run_test.go

@@ -128,8 +128,6 @@ func TestRunMultipleContainers(t *testing.T) {
 				buildIndexOpts := []indexBuildOption{withMinLayerSize(0)}
 				if tt.forceRecreateZtocs {
 					buildIndexOpts = append(buildIndexOpts, withForceRecreateZtocs(true))
-				} else {
-					buildIndexOpts = append(buildIndexOpts, withRunRebuildDbBeforeCreate())
 				}
 				indexDigest := buildIndex(sh, regConfig.mirror(container.containerImage), buildIndexOpts...)
 				sh.X("soci", "push", "--user", regConfig.mirror(container.containerImage).creds, regConfig.mirror(container.containerImage).ref)

integration/util_soci_test.go

@@ -92,19 +92,6 @@ func withForceRecreateZtocs(forceRecreateZtocs bool) indexBuildOption {
 	}
 }
 
-// withRebuildDbBeforeCreate syncs the artifact store with the content store before calling "soci create"
-// We do this because the artifact store could be out of sync with the content store when 'soci create' is called.
-// This is problematic in cases where we create soci indexes for some images, delete those indexes and immediately recreate
-// them (like in TestSociIndexRemove) - as there could be ztoc entries in the artifact store which are not present in the
-// content store, causing 'soci create/convert' without --force flag to throw an error.
-//
-// We can run this by default and probably remove this option in the future when the race condition with rebuild-db is solved.
-func withRunRebuildDbBeforeCreate() indexBuildOption {
-	return func(ibc *indexBuildConfig) {
-		ibc.runRebuildDbBeforeCreate = true
-	}
-}
-
 // withSpanSize overrides the default span size to use when creating an index with `buildIndex`
 func withSpanSize(spanSize int64) indexBuildOption {
 	return func(ibc *indexBuildConfig) {

soci/soci_index.go

@@ -632,21 +632,26 @@ func (b *IndexBuilder) buildSociLayer(ctx context.Context, desc ocispec.Descript
 
 	existingZtoc := b.getExistingZtocForLayer(desc)
 	if existingZtoc != nil {
+		var shouldOverwrite bool
 		reader, err := b.blobStore.Fetch(ctx, *existingZtoc)
 		if err != nil {
 			if errors.Is(err, errdefs.ErrNotFound) {
-				return nil, fmt.Errorf("a ztoc entry was found in the artifact store but not in the content store (try running \"soci rebuild-db\" first or try again with \"--force\" flag): %w", err)
+				log.G(ctx).Debug("a ztoc entry was found in the artifact store but not in the content store, overwriting existing ztoc")
+				shouldOverwrite = true
+			} else {
+				return nil, fmt.Errorf("failed to fetch existing ztoc: %w", err)
 			}
-			return nil, fmt.Errorf("failed to fetch existing ztoc: %w", err)
-		}
-		toc, err := ztoc.Unmarshal(reader)
-		if err != nil {
-			return nil, fmt.Errorf("cannot unmarshal existing ztoc: %w", err)
 		}
+		if !shouldOverwrite {
+			toc, err := ztoc.Unmarshal(reader)
+			if err != nil {
+				return nil, fmt.Errorf("cannot unmarshal existing ztoc: %w", err)
+			}
 
-		fmt.Printf("layer %s -> ztoc %s (already exists)\n", desc.Digest, existingZtoc.Digest)
-		b.addSociLayerAnnotations(&desc, existingZtoc, toc)
-		return existingZtoc, err
+			fmt.Printf("layer %s -> ztoc %s (already exists)\n", desc.Digest, existingZtoc.Digest)
+			b.addSociLayerAnnotations(&desc, existingZtoc, toc)
+			return existingZtoc, err
+		}
 	}
 
 	ra, err := b.contentStore.ReaderAt(ctx, desc)