Adding a new model backend

Flask-Admin makes a few assumptions about the database models that it works with. If you want to implement your own database backend, and still have Flask-Admin’s model views work as expected, then you should take note of the following:

1. Each model must have one field which acts as a primary key to uniquely identify instances of that model. However, there are no restriction on the data type or the field name of the primary key field.
2. Models must make their data accessible as python properties.

If that is the case, then you can implement your own database backend by extending the BaseModelView class, and implementing the set of scaffolding methods listed below.

Extending BaseModelView

Start off by defining a new class, which derives from from BaseModelView:

class MyDbModel(BaseModelView):
    pass

This class inherits BaseModelView’s __init__ method, which accepts a model class as first argument. The model class is stored as the attribute self.model so that other methods may access it.

Now, implement the following scaffolding methods for the new class:

1. get_pk_value()

   This method returns a primary key value from the model instance. In the SQLAlchemy backend, it gets the primary key from the model using scaffold_pk(), caches it and then returns the value from the model whenever requested.

   For example:

   class MyDbModel(BaseModelView):
       def get_pk_value(self, model):
           return self.model.id
   

2. scaffold_list_columns()

   Returns a list of columns to be displayed in a list view. For example:

   class MyDbModel(BaseModelView):
       def scaffold_list_columns(self):
           columns = []
   
           for p in dir(self.model):
               attr = getattr(self.model)
               if isinstance(attr, MyDbColumn):
                   columns.append(p)
   
           return columns
   

3. scaffold_sortable_columns()

   Returns a dictionary of sortable columns. The keys in the dictionary should correspond to the model’s field names. The values should be those variables that will be used for sorting.

   For example, in the SQLAlchemy backend it is possible to sort by a foreign key field. So, if there is a field named user, which is a foreign key for the Users table, and the Users table also has a name field, then the key will be user and value will be Users.name.

   If your backend does not support sorting, return None or an empty dictionary.

4. init_search()

   Initialize search functionality. If your backend supports full-text search, do initializations and return True. If your backend does not support full-text search, return False.

   For example, SQLAlchemy backend reads value of the self.searchable_columns and verifies if all fields are of text type, if they’re local to the current model (if not, it will add a join, etc) and caches this information for future use.

5. scaffold_form()

   Generate WTForms form class from the model.

   For example:

   class MyDbModel(BaseModelView):
       def scaffold_form(self):
           class MyForm(Form):
               pass
   
           # Do something
           return MyForm
   

6. get_list()

   This method should return list of model instances with paging, sorting, etc applied.

   For SQLAlchemy backend it looks like:

   1. If search was enabled and provided search value is not empty, generate LIKE statements for each field from self.searchable_columns

   2. If filter values were passed, call apply method with values:

      for flt, value in filters:
          query = self._filters[flt].apply(query, value)
      

   3. Execute query to get total number of rows in the database (count)

   4. If sort_column was passed, will do something like (with some extra FK logic which is omitted in this example):

      if sort_desc:
          query = query.order_by(desc(sort_field))
      else:
          query = query.order_by(sort_field)
      

   5. Apply paging

   6. Return count, list as a tuple

7. get_one()

   Return a model instance by its primary key.

8. create_model()

   Create a new instance of the model from the Form object.

9. update_model()

   Update the model instance with data from the form.

10. delete_model()

    Delete the specified model instance from the data store.

11. is_valid_filter()

    Verify whether the given object is a valid filter.

12. scaffold_filters()

    Return a list of filter objects for one model field.

    This method will be called once for each entry in the self.column_filters setting.

    If your backend does not know how to generate filters for the provided field, it should return None.

    For example:

    class MyDbModel(BaseModelView):
        def scaffold_filters(self, name):
            attr = getattr(self.model, name)
    
            if isinstance(attr, MyDbTextField):
                return [MyEqualFilter(name, name)]
    

Implementing filters

Each model backend should have its own set of filter implementations. It is not possible to use the filters from SQLAlchemy models in a non-SQLAlchemy backend. This also means that different backends might have different set of available filters.

The filter is a class derived from BaseFilter which implements at least two methods:

1. apply()
2. operation()

apply method accepts two parameters: query object and a value from the client. Here you can add filtering logic for the filter type.

Lets take SQLAlchemy model backend as an example:

All SQLAlchemy filters derive from BaseSQLAFilter class.

Each filter implements one simple filter SQL operation (like, not like, greater, etc) and accepts a column as input parameter.

Whenever model view wants to apply a filter to a query object, it will call apply method in a filter class with a query and value. Filter will then apply real filter operation.

For example:

class MyBaseFilter(BaseFilter):
    def __init__(self, column, name, options=None, data_type=None):
        super(MyBaseFilter, self).__init__(name, options, data_type)

        self.column = column

class MyEqualFilter(MyBaseFilter):
    def apply(self, query, value):
        return query.filter(self.column == value)

    def operation(self):
        return gettext('equals')

    # You can validate values. If value is not valid,
    # return `False`, so filter will be ignored.
    def validate(self, value):
        return True

    # You can "clean" values before they will be
    # passed to the your data access layer
    def clean(self, value):
        return value

Feel free ask questions if you have problems adding a new model backend. Also, if you get stuck, try taking a look at the SQLAlchemy model backend and use it as a reference.