Detect inconsistent writes to datasets properties (#633)

* emit warning when trying to export dataset properties while overwriting properties which are newer than the ones most recently seen

* update changelog

Author: philippotto

webknossos/Changelog.md

@@ -17,6 +17,7 @@ For upgrade instructions, please check the respective *Breaking Changes* section
 ### Added
 
 ### Changed
+- Inconsistent writes to datasets properties (e.g., caused due to multiprocessing) are detected automatically. The warning can be escalated to an exception with `warnings.filterwarnings("error", module="webknossos", message=r"\[WARNING\]")`. [#633](https://github.com/scalableminds/webknossos-libs/pull/633)
 
 ### Fixed
 

webknossos/tests/conftest.py

@@ -75,6 +75,13 @@ def error_on_deprecations() -> Generator:
         yield
 
 
+@pytest.fixture(autouse=True, scope="function")
+def error_on_warnings() -> Generator:
+    with warnings.catch_warnings():
+        warnings.filterwarnings("error", module="webknossos", message=r"\[WARNING\]")
+        yield
+
+
 ### VCR.py / pytest-recording CONFIG
 
 

webknossos/tests/test_dataset.py

@@ -1977,3 +1977,24 @@ def test_pickle_view(tmp_path: Path) -> None:
 
     # Make sure that the attributes of the MagView (not View) still exist
     assert pickled_mag1.layer is not None
+
+
+def test_warn_outdated_properties(tmp_path: Path) -> None:
+
+    ds1 = Dataset(tmp_path / "ds", scale=(1, 1, 1))
+    ds2 = Dataset.open(tmp_path / "ds")
+
+    # Change ds1 and undo it again
+    ds1.add_layer("color", COLOR_CATEGORY).add_mag(1)
+    ds1.delete_layer("color")
+
+    # Changing ds2 should work fine, since the properties on disk
+    # haven't changed.
+    ds2.add_layer("segmentation", SEGMENTATION_CATEGORY, largest_segment_id=1).add_mag(
+        1
+    )
+
+    with pytest.raises(UserWarning):
+        # Changing ds1 should raise a warning, since ds1
+        # does not know about the change in ds2
+        ds1.add_layer("color", COLOR_CATEGORY)

webknossos/webknossos/dataset/dataset.py

@@ -148,30 +148,24 @@ def __init__(
                 )
 
         self.path = dataset_path
-        with open(
-            self.path / PROPERTIES_FILE_NAME, encoding="utf-8"
-        ) as datasource_properties:
-            data = json.load(datasource_properties)
-
-            self._properties: DatasetProperties = dataset_converter.structure(
-                data, DatasetProperties
+        self._properties: DatasetProperties = self._load_properties()
+        self._last_read_properties = copy.deepcopy(self._properties)
+
+        self._layers: Dict[str, Layer] = {}
+        # construct self.layer
+        for layer_properties in self._properties.data_layers:
+            num_channels = _extract_num_channels(
+                layer_properties.num_channels,
+                Path(dataset_path),
+                layer_properties.name,
+                layer_properties.wkw_resolutions[0].resolution
+                if len(layer_properties.wkw_resolutions) > 0
+                else None,
             )
+            layer_properties.num_channels = num_channels
 
-            self._layers: Dict[str, Layer] = {}
-            # construct self.layer
-            for layer_properties in self._properties.data_layers:
-                num_channels = _extract_num_channels(
-                    layer_properties.num_channels,
-                    Path(dataset_path),
-                    layer_properties.name,
-                    layer_properties.wkw_resolutions[0].resolution
-                    if len(layer_properties.wkw_resolutions) > 0
-                    else None,
-                )
-                layer_properties.num_channels = num_channels
-
-                layer = self._initialize_layer_from_properties(layer_properties)
-                self._layers[layer_properties.name] = layer
+            layer = self._initialize_layer_from_properties(layer_properties)
+            self._layers[layer_properties.name] = layer
 
         if dataset_existed_already:
             if scale is None:
@@ -825,14 +819,31 @@ def get_or_create(
     def __repr__(self) -> str:
         return repr("Dataset(%s)" % self.path)
 
+    def _load_properties(self) -> DatasetProperties:
+        with open(
+            self.path / PROPERTIES_FILE_NAME, encoding="utf-8"
+        ) as datasource_properties:
+            data = json.load(datasource_properties)
+        return dataset_converter.structure(data, DatasetProperties)
+
     def _export_as_json(self) -> None:
+
+        properties_on_disk = self._load_properties()
+
+        if properties_on_disk != self._last_read_properties:
+            warnings.warn(
+                "[WARNING] While exporting the dataset's properties, properties were found on disk which are newer than the ones that were seen last time. The properties will be overwritten. This is likely happening because multiple processes changed the metadata of this dataset."
+            )
+
         with open(self.path / PROPERTIES_FILE_NAME, "w", encoding="utf-8") as outfile:
             json.dump(
                 dataset_converter.unstructure(self._properties),
                 outfile,
                 indent=4,
             )
 
+            self._last_read_properties = copy.deepcopy(self._properties)
+
     def _initialize_layer_from_properties(self, properties: LayerProperties) -> Layer:
         if properties.category == COLOR_CATEGORY:
             return Layer(self, properties)