Fix OCPBUGS-17546: pod catalogsource generated by oc-mirror will cras…

…hloopBackOff randomly (#699)

Co-authored-by: skhoury <skhoury@redhat.com>

pkg/cli/mirror/catalog_images.go

@@ -225,7 +225,7 @@ func (o *MirrorOptions) processCatalogRefs(ctx context.Context, catalogsByImage
 			return fmt.Errorf("unable to determine location of cache for image %s. Cache generation failed: %v", ctlgRef, err)
 		}
 
-		configLayerToAdd, err := builder.LayerFromPath("/configs", filepath.Join(artifactDir, config.IndexDir, "index.json"))
+		configLayerToAdd, err := builder.LayerFromPathWithUidGid("/configs", filepath.Join(artifactDir, config.IndexDir), 0, 0)
 		if err != nil {
 			return fmt.Errorf("error creating add layer: %v", err)
 		}
@@ -240,20 +240,6 @@ func (o *MirrorOptions) processCatalogRefs(ctx context.Context, catalogsByImage
 		layersToDelete = append(layersToDelete, deletedConfigLayer)
 
 		if withCacheRegeneration {
-			// read the location of the cache
-			dat, err := os.ReadFile(filepath.Join(artifactDir, config.OPMCacheLocationPlaceholder))
-			if err != nil {
-				return fmt.Errorf("unable to determine location of cache for image %s. Cache generation failed: %v", ctlgRef, err)
-			}
-			cacheLocation := string(dat)
-			cacheLocationElmts := strings.Split(strings.TrimPrefix(cacheLocation, string(os.PathSeparator)), string(os.PathSeparator))
-
-			// white out layer /tmp
-			deletedCacheLayer, err := deleteLayer("/.wh." + cacheLocationElmts[0])
-			if err != nil {
-				return fmt.Errorf("error creating deleted cache layer: %v", err)
-			}
-			layersToDelete = append(layersToDelete, deletedCacheLayer)
 
 			opmCmdPath := filepath.Join(artifactDir, config.OpmBinDir, "opm")
 			_, err = os.Stat(opmCmdPath)
@@ -272,7 +258,9 @@ func (o *MirrorOptions) processCatalogRefs(ctx context.Context, catalogsByImage
 			if err := cmd.Run(); err != nil {
 				return fmt.Errorf("error regenerating the cache for %v: %v", ctlgRef, err)
 			}
-			cacheLayerToAdd, err := builder.LayerFromPathWithUidGid("/tmp/cache", filepath.Join(artifactDir, config.TmpDir), cacheFolderUID, cacheFolderGID)
+			// Fix OCPBUGS-17546:
+			// Add the cache under /cache in a new layer (instead of white-out /tmp/cache, which resulted in crashLoopBackoff only on some clusters)
+			cacheLayerToAdd, err := builder.LayerFromPathWithUidGid("/cache", filepath.Join(artifactDir, config.TmpDir), cacheFolderUID, cacheFolderGID)
 			if err != nil {
 				return fmt.Errorf("error creating add layer: %v", err)
 			}
@@ -298,10 +286,9 @@ func (o *MirrorOptions) processCatalogRefs(ctx context.Context, catalogsByImage
 			}
 			cfg.Config.Labels = labels
 			// Although it was prefered to keep the entrypoint and command as it was
-			// we couldnt guarantie that the cache-dir was /tmp/cache, and therefore
-			// we had to specify the command for the newly built catalog
+			// we couldnt reuse /tmp/cache as the cache directory (OCPBUGS-17546)
 			if withCacheRegeneration {
-				cfg.Config.Cmd = []string{"serve", "/configs", "--cache-dir=/tmp/cache"}
+				cfg.Config.Cmd = []string{"serve", "/configs", "--cache-dir=/cache"}
 			} else { // this means that no cache was found in the original catalog (old catalog with opm < 1.25)
 				cfg.Config.Cmd = []string{"serve", "/configs"}
 			}

pkg/image/builder/image_builder.go

@@ -9,6 +9,7 @@ import (
 	"os"
 	"path/filepath"
 	"strings"
+	"time"
 
 	"github.com/go-logr/logr"
 	"github.com/google/go-containerregistry/pkg/name"
@@ -361,7 +362,7 @@ func LayerFromPathWithUidGid(targetPath, path string, uid int, gid int) (v1.Laye
 		if !info.IsDir() {
 			hdr.Size = info.Size()
 		}
-
+		hdr.ChangeTime = time.Now()
 		if info.Mode().IsDir() {
 			hdr.Typeflag = tar.TypeDir
 		} else if info.Mode().IsRegular() {

test/e2e/lib/check.sh

@@ -26,9 +26,9 @@ function check_bundles() {
   mv index.json $index_dir
 
   # extract the cache from the tar file
-  local cache_dir="${extraction_dir}/tmp/cache"
-  tar xvf $extraction_dir/temp.tar /tmp
-  mv tmp "${extraction_dir}"
+  local cache_dir="${extraction_dir}/cache"
+  tar xvf $extraction_dir/temp.tar /cache
+  mv cache "${extraction_dir}"
 
   # extract the opm binary from the tar file
   local opm_path="${extraction_dir}/opm"