WIP: bug fix: OCPBUGS-97746: verify and sync catalog.jsonl file is available before…

internal/catalogd/controllers/core/clustercatalog_controller.go

@@ -270,6 +270,12 @@ func (r *ClusterCatalogReconciler) reconcile(ctx context.Context, catalog *ocv1.
 	}
 	baseURL := r.Storage.BaseURL(catalog.Name)
 
+	if err := r.Storage.VerifyAndSync(catalog.Name); err != nil {
+		verifyErr := fmt.Errorf("error verifying catalog content before serving: %w", err)
+		updateStatusProgressing(&catalog.Status, catalog.GetGeneration(), verifyErr)
+		return ctrl.Result{}, verifyErr
+	}
+
 	updateStatusProgressing(&catalog.Status, catalog.GetGeneration(), nil)
 	updateStatusServing(&catalog.Status, canonicalRef, unpackTime, baseURL, catalog.GetGeneration())
 

internal/catalogd/controllers/core/clustercatalog_controller_test.go

@@ -31,9 +31,11 @@ func newMockStore(ctrl *gomock.Controller, shouldError bool) *mockstorage.MockIn
 	if shouldError {
 		m.EXPECT().Store(gomock.Any(), gomock.Any(), gomock.Any()).Return(errors.New("mockstore store error")).AnyTimes()
 		m.EXPECT().Delete(gomock.Any()).Return(errors.New("mockstore delete error")).AnyTimes()
+		m.EXPECT().VerifyAndSync(gomock.Any()).Return(errors.New("mockstore verify error")).AnyTimes()
 	} else {
 		m.EXPECT().Store(gomock.Any(), gomock.Any(), gomock.Any()).Return(nil).AnyTimes()
 		m.EXPECT().Delete(gomock.Any()).Return(nil).AnyTimes()
+		m.EXPECT().VerifyAndSync(gomock.Any()).Return(nil).AnyTimes()
 	}
 	m.EXPECT().BaseURL(gomock.Any()).Return("URL").AnyTimes()
 	m.EXPECT().ContentExists(gomock.Any()).Return(true).AnyTimes()

internal/catalogd/storage/localdir.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"io"
 	"io/fs"
 	"net/http"
 	"net/url"
@@ -230,6 +231,38 @@ func (s *LocalDirV1) ContentExists(catalog string) bool {
 	return true
 }
 
+// VerifyAndSync verifies that catalog.jsonl exists at RootDir/<catalog>/catalog.jsonl,
+// calls fsync to ensure the file is durably written to disk, and confirms the file is
+// readable by attempting a small read. It should be called after Store() succeeds and
+// before marking the catalog as Serving.
+func (s *LocalDirV1) VerifyAndSync(catalog string) error {
+	s.m.RLock()
+	defer s.m.RUnlock()
+
+	path := catalogFilePath(s.catalogDir(catalog))
+
+	if _, err := os.Stat(path); err != nil {
+		return fmt.Errorf("catalog.jsonl not found at %q: %w", path, err)
+	}
+
+	f, err := os.Open(path)
+	if err != nil {
+		return fmt.Errorf("catalog.jsonl not readable at %q: %w", path, err)
+	}
+	defer f.Close()
+
+	if err := f.Sync(); err != nil {
+		return fmt.Errorf("fsync failed for catalog.jsonl at %q: %w", path, err)
+	}
+
+	buf := make([]byte, 1)
+	if _, err := f.Read(buf); err != nil && !errors.Is(err, io.EOF) {
+		return fmt.Errorf("catalog.jsonl read failed at %q: %w", path, err)
+	}
+
+	return nil
+}
+
 func (s *LocalDirV1) catalogDir(catalog string) string {
 	return filepath.Join(s.RootDir, catalog)
 }

internal/catalogd/storage/storage.go

@@ -15,6 +15,12 @@ type Instance interface {
 	Delete(catalog string) error
 	ContentExists(catalog string) bool
 
+	// VerifyAndSync confirms that catalog.jsonl exists on disk for the given
+	// catalog, flushes it to stable storage via fsync, and verifies the file
+	// is readable. It must be called after Store and before marking the
+	// catalog as Serving.
+	VerifyAndSync(catalog string) error
+
 	BaseURL(catalog string) string
 	StorageServerHandler() http.Handler
 }