Fix massive DSO-API performance issue, due to deepcopy()

DRF performs a deepcopy() of the serializer._declared_fields before it
will handle the request. This is analogous to how Django forms work, and
allows updating the fields with request state and 'parent' references.

Since the deepcopy() also touched dataset definitions of Amsterdam Schema,
it basically took 14sec to copy all those dicts as well. This can be
avoided, bringing to performance of DSO-API back to reasonable levels.

src/dso_api/dynamic_api/serializers/factories.py

@@ -534,11 +534,9 @@ def _build_serializer_embedded_field(
 
     embedded_field = EmbeddedFieldClass(
         serializer_class=cast(type[base.DynamicSerializer], serializer_class),
-        # serializer_class=serializer_class,
+        field_schema=field_schema,
         source=model_field.name,
     )
-    # Attach the field schema so access rules can be applied here.
-    embedded_field.field_schema = field_schema
 
     # The field name is still generated from the model_field, in case this is a reverse field.
     serializer_part.add_embedded_field(toCamelCase(model_field.name), embedded_field)

src/dso_api/dynamic_api/serializers/fields.py

@@ -63,8 +63,18 @@ def __init__(self, table_schema: DatasetTableSchema, *args, **kwargs):
         # Init adds temporal definitions at construction, removing runtime model lookups.
         # It also allows the PK optimization to be used.
         super().__init__(*args, **kwargs)
+
+        # Link the table to perform temporal query filtering
         self.table_schema = table_schema
 
+    def __deepcopy__(self, memo):
+        # Fix a massive performance hit when DRF performs a deepcopy() of all fields
+        result = self.__class__.__new__(self.__class__)
+        memo[id(result)] = self
+        result.__dict__.update({k: v for k, v in self.__dict__.items() if k != "table_schema"})
+        result.table_schema = self.table_schema
+        return result
+
     def use_pk_only_optimization(self):
         return True  # only need to have an "id" here.
 

src/rest_framework_dso/fields.py

@@ -15,6 +15,7 @@
 from rest_framework.exceptions import ValidationError
 from rest_framework_gis.fields import GeometryField
 from schematools.contrib.django.models import LooseRelationField
+from schematools.types import DatasetFieldSchema
 
 from rest_framework_dso.utils import group_dotted_names, unlazy_object
 
@@ -289,14 +290,24 @@ def __init__(
         self,
         serializer_class: type[serializers.Serializer],
         *,
+        field_schema: DatasetFieldSchema | None = None,
         source=None,
     ):
         self.serializer_class = serializer_class
         self.source = source
 
         self.field_name = None
+        self.field_schema = field_schema
         self.parent_serializer_class = None
 
+    def __deepcopy__(self, memo):
+        # Fix a massive performance hit when DRF performs a deepcopy() of all fields
+        result = self.__class__.__new__(self.__class__)
+        memo[id(result)] = self
+        result.__dict__.update({k: v for k, v in self.__dict__.items() if k != "field_schema"})
+        result.field_schema = self.field_schema
+        return result
+
     def __repr__(self):
         try:
             parent_serializer = self.parent_serializer_class.__name__