A clean approach to app configuration

app config configuration schema python
pip install grift==0.7.1


Build Status Coverage Status Updates Python 3


Classes to define, load, validate, and store values for an application's configuration.


pip install grift

Define Configuration Schema

Create a class that inherits from BaseConfig and add a ConfigProperty for each setting

from grift import BaseConfig, ConfigProperty, DictLoader
from schematics.types import BooleanType, StringType, IntType

class AppConfig(BaseConfig):
    # defaults can be specified, and properties can be optional
    DEBUG = ConfigProperty(property_type=BooleanType(), required=False, default=False)

    # attribute name does not need to match the property key
    ELASTICSEARCH_HOST = ConfigProperty(property_key='ES_HOST', property_type=StringType())
    ELASTICSEARCH_SHARDS = ConfigProperty(property_key='ES_SHARDS', property_type=IntType(), default=5)

    # properties can be excluded from varz
    AWS_SECRET_KEY = ConfigProperty(property_type=StringType(), exclude_from_varz=True)


property_key is the key used to pull a property value from a loader. If unspecified, defaults to the name of the attribute on the Config class (e.g. 'DEBUG' and 'AWS_SECRET_KEY' in the example above).

To skip validation, leave property_type as the default (None). To define new property type classes, subclass schematic.types.BaseType (see .property_types.DictType as an example).

When required is True, the value cannot be None (i.e. the value must exist in one of the loaders or a default value should be specified.) However, if a default is specified (and not None), the requirement is always satisfied -- required is effectively False.

exclude_from_varz indicates whether the property should be left out of BaseConfig's varz dict (defaults to False).

Load and Validate Settings

Initialize the custom configuration class with one or more loaders. Each loader specifies a source for loading configuration settings.

config_dict = {
    'DEBUG': 1,
    'ES_HOST': 'http://localhost:9200',
    'ES_SHARDS': '1',
    'AWS_SECRET_KEY': 'whatever'
loaders = [DictLoader(config_dict)]

app_config = AppConfig(loaders)

The config can be initialized with multiple loaders. Each ConfigProperty's value is assigned from the first loader that contains the ConfigProperty.property_key. For example, with ES_SHARDS = '5' in the environment:

loaders = [EnvLoader(), DictLoader(config_dict)]
app_config2 = AppConfig(loaders)

app_config2.ELASTICSEARCH_SHARDS  # 5 (from env)
app_config2.ELASTICSEARCH_HOST  # u'http://localhost:9200'  (from dict)

When the configuration class is initialized, the sequence for loading a value is as follows, for each ConfigProperty:

  • If property_key is not defined, use the name of the attribute on the class (e.g. DEBUG in AppConfig).
  • Check whether the property_key exists in each of the loaders. Iterate through the loaders in the order provided, stopping at the first loader where the key exists.
  • If the key exists in one of the loaders:
    • Pull the value of the property_key from that loader.
    • If a property_type class is defined for the ConfigProperty, use to_native() to convert the loaded value to the appropriate type and validate() to check any other assumptions (e.g. max string length, connectivity to a network type, etc). An exception may be raised at this stage, if the value cannot be converted or validated.
    • If no property_type was defined, use the value as is.
  • If the key does not exist in any of the loaders:
    • If a default value was specified, use it.
    • If no default value was specified AND the property is required, raise an exception.
    • Otherwise, the value of the property is set to None
Default loaders are available in loaders.DEFAULT_LOADERS. The default is to prefer the EnvLoader, which reads in environment variables. If $SETTINGS_PATH is defined in the env, a second loader is
added to pull in settings from a json file at the specified path.

Property Types

Use schematics.types classes to convert and validate values at load time.

A custom property type can be created by extending schematics.type.BaseType. Implement .to_native() to convert a value type (returning the converted value or raising an exception for incompatible types). Define one or more methods with names that start with validate_ (e.g. .validate_length()) to add validation steps. Validation methods should raise schematics.exceptions.ValidationError for failed checks.

Access Property Values

>>> app_config.DEBUG
Note that when the class is initialized, attributes that are ConfigProperty instances are set to
the loaded values:
>>> type(app_config.ELASTICSEARCH_SHARDS)

Get public configuration settings

The varz property of BaseConfig classes is a dict with the values for each ConfigProperty attribute. Any ConfigProperty can be excluded from varz by specifying exclude_from_varz=True.

>>> app_config.varz
    'DEBUG': True,
    'ELASTICSEARCH_HOST': 'http://localhost:9200',
    'ES_SHARDS': 1

All ConfigProperty values can be accessed in a dict, using .as_dict():

>>> app_config.as_dict()
{'AWS_SECRET_KEY': 'whatever'
 'DEBUG': 1,
 'ES_HOST': 'http://localhost:9200',
 'ES_SHARDS': '1'}

Maximizing Startup Guarantees

You may want to set up your config class to maximize startup guarantees of having the right configuration set. There are a few property types that attempt to make a basic connection with whatever network resouce is specified. The supported protocols are http, postgres, redis, amqp, and etcd. By default, the validator will back off 5 times before giving up, but that can be overridden with the 'max_tries' kwarg.

For example:

class AppConfig(BaseConfig):
     DATABASE_URL = ConfigProperty(property_type=PostgresType(), default='postgres://...')
     REDIS_URL = ConfigProperty(property_type=RedisType(max_tries=1))
     SHARED_CONFIG =  ConfigProperty(property_type=StringType(), default='A')

class DeploymentConfig(AppConfig):
    DATABASE_URL = ConfigProperty(property_type=PostgresType())

ConfigCls = AppConfig if deploy.env not in [STAGE, PROD] else DeployedConfig
config = ConfigCls(loaders)

An important distinction in this example is that the config schema changes based on the deploy env. For the staging and production environments, DeploymentConfig will fail to initialize if DATABASE_URL isn't set.


Licensed under the Apache 2.0 License. Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Copyright 2017 Kensho Technologies, LLC.