Fix lots of potential stale reads in Element.add_parent and .remove_child

Part 1/2 of #881 (closed), resolves ARKINDEX-BACKEND-SD

This fixes six possible stale reads in both add_parent and remove_child, so this affects at least CreateElement, CreateElementParent, DestroyElementParent and the move_element RQ task. There still are more elsewhere :D

See https://gitlab.com/arkindex/backend/-/issues/881#note_714675726 for details

arkindex/documents/models.py

@@ -221,7 +221,7 @@ class Element(IndexableModel):
         delete_element(self.id)
         Element.objects.filter(id=self.id)._raw_delete(using='default')
 
-    @transaction.atomic(using="default")
+    @transaction.atomic
     def add_parent(self, parent, skip_children=False):
         '''
         Add an element as ancestor
@@ -230,12 +230,12 @@ class Element(IndexableModel):
 
         # Load existing paths on current element, and the parent paths
         # in a single query
-        paths = ElementPath.objects.filter(element_id__in=(self.id, parent.id)).values('element_id', 'path')
+        paths = ElementPath.objects.using('default').filter(element_id__in=(self.id, parent.id)).values('element_id', 'path')
         existing_paths = [p['path'] for p in paths if p['element_id'] == self.id]
         parent_paths = [p['path'] for p in paths if p['element_id'] == parent.id]
 
         # Get the next order for this type in the parent
-        order = parent.get_next_order(self.type)
+        order = parent.get_next_order(self.type_id)
 
         # Build all parent paths needed
         if not parent_paths:
@@ -262,10 +262,10 @@ class Element(IndexableModel):
                 order,
             ))
 
-        # Update children's paths in a non optimized way as this is pretty rare
+        # Update children's paths in a non optimized way
         if skip_children is False:
             to_delete = []
-            children = ElementPath.objects.filter(path__contains=[self.id])
+            children = ElementPath.objects.using('default').filter(path__contains=[self.id])
             for child in children:
                 base_path = child.path
 
@@ -299,7 +299,7 @@ class Element(IndexableModel):
         Uses the primary database to avoid stale reads and duplicate orderings,
         and path__overlap to let Postgres use the GIN index before filtering by last item.
         """
-        assert isinstance(type, ElementType)
+        assert isinstance(type, (ElementType, uuid.UUID))
         return ElementPath.objects \
                           .using('default') \
                           .filter(path__overlap=[self.id], path__last=self.id, element__type=type) \
@@ -313,16 +313,16 @@ class Element(IndexableModel):
         assert isinstance(child, Element)
 
         # Check child is a direct child of this element
-        if not ElementPath.objects.filter(element=child, path__last=self.id).exists():
+        if not ElementPath.objects.using('default').filter(element=child, path__last=self.id).exists():
             return
 
         new_epaths = {}
 
         # If child doesn't have any other parent, updates are needed
-        if not ElementPath.objects.filter(element=child).exclude(path__last=self.id).exists():
+        if not ElementPath.objects.using('default').filter(element=child).exclude(path__last=self.id).exists():
 
             # Get all the child's children's paths
-            children_epaths = list(ElementPath.objects.filter(
+            children_epaths = list(ElementPath.objects.using('default').filter(
                 element__in=Element.objects.get_descending(child.id),
                 path__contains=[self.id]
             ))