rbd: check local image state during promote operation

internal/rbd/errors.go

@@ -42,4 +42,6 @@ var (
 	ErrMissingImageNameInVolID = errors.New("rbd image name information can not be empty in volID")
 	// ErrDecodeClusterIDFromMonsInVolID is returned when mons hash decoding on migration volID.
 	ErrDecodeClusterIDFromMonsInVolID = errors.New("failed to get clusterID from monitors hash in volID")
+	// ErrUnHealthyMirroredImage is returned when mirrored image is not healthy.
+	ErrUnHealthyMirroredImage = errors.New("mirrored image is not healthy")
 )

internal/rbd/replicationcontrollerserver.go

@@ -536,6 +536,11 @@ func (rs *ReplicationServer) PromoteVolume(ctx context.Context,
 		}
 	}
 
+	err = checkHealthyPrimary(ctx, rbdVol)
+	if err != nil {
+		return nil, status.Error(codes.Internal, err.Error())
+	}
+
 	var mode librbd.ImageMirrorMode
 	mode, err = getMirroringMode(ctx, req.GetParameters())
 	if err != nil {
@@ -565,6 +570,35 @@ func (rs *ReplicationServer) PromoteVolume(ctx context.Context,
 	return &replication.PromoteVolumeResponse{}, nil
 }
 
+// checkHealthyPrimary checks if the image is a healhty primary or not.
+// healthy primary image will be in up+stopped state, for states other
+// than this it returns an error message.
+func checkHealthyPrimary(ctx context.Context, rbdVol *rbdVolume) error {
+	mirrorStatus, err := rbdVol.getImageMirroringStatus()
+	if err != nil {
+		return err
+	}
+	localStatus, err := mirrorStatus.LocalStatus()
+	if err != nil {
+		// LocalStatus can fail if the local site status is not found in
+		// mirroring status. Log complete sites status to debug why getting
+		// local status failed
+		log.ErrorLog(ctx, "mirroring status is %+v", mirrorStatus)
+
+		return fmt.Errorf("failed to get local status: %w", err)
+	}
+
+	if !localStatus.Up && localStatus.State != librbd.MirrorImageStatusStateStopped {
+		return fmt.Errorf("%s %w. State is up=%t, state=%q",
+			rbdVol,
+			ErrUnHealthyMirroredImage,
+			localStatus.Up,
+			localStatus.State)
+	}
+
+	return nil
+}
+
 // DemoteVolume extracts the RBD volume information from the
 // volumeID, If the image is present, mirroring is enabled and the
 // image is in promoted state it will demote the volume as secondary.