Restore retained snapshot base after standby snapshot failure

Author: cursoragent

lib/instances/standby.go

@@ -105,7 +105,7 @@ func (m *manager) standbyInstance(
 		log.ErrorContext(ctx, "snapshot failed, attempting to resume VM", "instance_id", id, "error", err)
 		hv.Resume(ctx)
 		if promotedExistingBase {
-			if rollbackErr := discardPromotedRetainedSnapshotTarget(snapshotDir); rollbackErr != nil {
+			if rollbackErr := discardPromotedRetainedSnapshotTarget(snapshotDir, retainedBaseDir); rollbackErr != nil {
 				log.WarnContext(ctx, "failed to discard promoted snapshot target after snapshot error", "instance_id", id, "error", rollbackErr)
 			}
 		}
@@ -209,8 +209,8 @@ func prepareRetainedSnapshotTarget(snapshotDir string, retainedBaseDir string) (
 	return false, nil
 }
 
-func discardPromotedRetainedSnapshotTarget(snapshotDir string) error {
-	return os.RemoveAll(snapshotDir)
+func discardPromotedRetainedSnapshotTarget(snapshotDir string, retainedBaseDir string) error {
+	return restoreRetainedSnapshotBase(snapshotDir, retainedBaseDir)
 }
 
 func restoreRetainedSnapshotBase(snapshotDir string, retainedBaseDir string) error {

lib/instances/standby_test.go

@@ -29,10 +29,12 @@ func TestDiscardPromotedRetainedSnapshotTargetAfterSnapshotError(t *testing.T) {
 	// Simulate a partially written snapshot target before the snapshot API returns an error.
 	require.NoError(t, os.WriteFile(filepath.Join(snapshotDir, "partial-marker"), []byte("partial"), 0644))
 
-	require.NoError(t, discardPromotedRetainedSnapshotTarget(snapshotDir))
+	require.NoError(t, discardPromotedRetainedSnapshotTarget(snapshotDir, retainedBaseDir))
 
 	_, err = os.Stat(snapshotDir)
 	assert.True(t, os.IsNotExist(err), "snapshot failures should discard the promoted snapshot target")
 	_, err = os.Stat(retainedBaseDir)
-	assert.True(t, os.IsNotExist(err), "snapshot failures should not restore the promoted base for reuse")
+	require.NoError(t, err, "snapshot failures should restore the promoted base for reuse")
+	_, err = os.Stat(filepath.Join(retainedBaseDir, "base-marker"))
+	assert.NoError(t, err, "base snapshot contents should be restored for retry")
 }