6.4 KiB


Python application configuration

What is Bison?

Bison is a configuration solution for Python applications that aims to be simple and intuitive. It supports:

  • reading from YAML config files
  • reading from environment variables
  • setting explicit values
  • setting defaults
  • configuration validation
  • configuration access/manipulation with dot notation

Instead of implementing custom configuration reading and parsing, you can use Bison to handle it for you.

Bison was inspired by Viper and the lack of good application configuration solutions for Python (at least, in my opinion). Documentation for Bison can be found on ReadtheDocs

Bison uses the following precedence order. Each item in the list takes precedence over the item below it.

  • override (e.g. calling Bison.set())
  • environment
  • config file
  • defaults


Bison can be installed with pip

pip install bison

or with pipenv

pipenv install bison

Using Bison

Creating a configuration Scheme

A configuration scheme is not required by Bison, but having one allows you to set default values for configuration fields as well as do configuration validation. It is pretty easy to create a new Scheme:

scheme = bison.Scheme()

A Scheme is really just a container for configuration options, so without any options, a Scheme is somewhat useless.

Configuration Options

There are currently three types of configuration options:

  • bison.Option
  • bison.DictOption
  • bison.ListOption

Their intended functionality should be mostly obvious from their names. An Option represents a singular value in a configuration. A DictOption represents a dictionary or mapping of values in a configuration. A ListOption represents a list of values in a configuration.

See the documentation for more on how options can be configured.

Any number of options can be added to a Scheme, but as a simple example we can define a Scheme which expects a key "log", and a key "count".

scheme = bison.Scheme(

Configuration Validation

Validation operates based on the constraints set on the options. Above, there are no constraints (other than the need for those keys to exist), so any value for "log" and "count" will pass validation.

An option can be constrained in different ways by using its keyword arguments. For example, to ensure the value for "count" is an integer,

bison.Option('count', field_type=int)

Or, to restrict the values to a set of choices

bison.Option('log', choices=['debug', 'info', 'warn', 'error'])

The documentation goes into more detail about other validation settings.

Setting Defaults

If a default value is not set on an option, it is considered required. In these cases, if the key specified by that value is not present in the parse configuration, it will cause a validation failure.

If a default value is set, then the absence of that field in the configuration will not cause a validation failure.

bison.Option('log', default='info')

Configuring Bison

Once you have a Scheme to use (if you'd like to), it will need to be passed to a Bison object to manage the config building.

scheme = bison.Scheme()
config = bison.Bison(scheme)

There are a few options that can be set on the Bison object to change how it searches for and builds the unified configuration.

For reading configuration files

config.config_name = 'config'  # name of the config file (no extension)
config.add_config_paths(       # paths to look in for the config file
config.config_format = bison.YAML # the config format to use

For reading environment variables

config.env_prefix = "MY_APP"  # the prefix to use for environment variables
config.auto_env = True  #  automatically bind all options to env variables based on their key

Building the unified config

Once the scheme has been set (if using) and Bison has been configured, the only thing left to do is to read in all the config sources and parse them into a unified config. This is done simply with



Below is a complete example for parsing a hypothetical application configuration which is described by the following YAML config.

log: debug
port: 5000
    timeout: 3
  - host:
    port: 5001
  - host:
    port: 5013
  - host:
    port: 5044
import bison

# the scheme for the configuration. this allows us to set defaults
# and validate configuration data
config_scheme = bison.Scheme(
    bison.Option('log', default='info', choices=['debug', 'info', 'warn', 'error']),
    bison.Option('port', field_type=int),
    bison.DictOption('settings', scheme=bison.Scheme(
        bison.DictOption('requests', scheme=bison.Scheme(
            bison.Option('timeout', field_type=int)
    bison.ListOption('backends', member_scheme=bison.Scheme(
        bison.Option('host', field_type=str),
        bison.Option('port', field_type=int)

# create a new Bison instance to store and manage configuration data
config = bison.Bison(scheme=config_scheme)

# set the config file name to 'app' (default is 'config') and set the
# search paths to '.' and '/tmp/app/config'
config.config_name = 'app'
config.add_config_paths('.', '/tmp/app/config')

# set the environment variable prefix and enable auto-env
config.env_prefix = 'MY_APP'
config.auto_env = True

# finally, parse the config sources to build the unified configuration

See the example directory for this example along with demonstrations of how to access configuration data.

Future Work

There is more that can be done to improve Bison and expand its functionality. If you wish to contribute, open a pull request. If you have questions or feature requests, open an issue. Below are some high level ideas for future improvements:

  • Support additional configuration formats (JSON, TOML, ...)
  • Versioned configurations