pkonfig 1.2

Pythonic agile application configuration helpers

PKonfig

P stands for Python.

Prerequisites

• Pythonic configuration management helpers.
• Multiple sources of configs (environment variables, dotenv files, YAML, JSON, TOML, INI) with agile order configuration.
• Configs validation mechanics based on type hints or user defined classes.
• Minimal external dependencies.
• Follow Fail-fast principle.
• Autocomplete in modern IDEs.

Features

• User defined config source order.
• Multilevel configs for environment variables and dotenv config sources.
• Custom aliases for fields or groups of configs.
• Configs type casting
• Config values validation based on type and/or value.
• High performance.
• Extendable API.

Installation

To install basic PKonfig without YAML and TOML support run:

pip install pkonfig

YAML files parsing is handled with PyYaml:

pip install pkonfig[yaml]

TOML files handled with help of Tomli:

pip install pkonfig[toml]

And if both TOML and YAML is needed:

pip install pkonfig[toml,yaml]

For production no .env files are needed but propper environment variables should be set. In case some of required variables missing KeyError exception raised while AppConfig instantiation.

Quickstart

The most basic usage example when environment variables are used for production environment and DotEnv files are used for local development.

Create config module config.py:

from collections import ChainMap
from pkonfig import Config, EmbeddedConfig, Env, DotEnv, LogLevel, Choice


class PG(EmbeddedConfig):
  host = "localhost"
  port = 5432
  user: str
  password: str

  
class AppConfig(Config):
    db = PG()
    log_level = LogLevel("INFO")
    env = Choice(["local", "prod", "test"], default="prod")


storage = ChainMap(DotEnv(".env", missing_ok=True), Env())
config = AppConfig(storage)

For local development create DotEnv file in root app folder .env:

APP_DB_HOST=localhost
APP_DB_USER=postgres
APP_DB_PASSWORD=postgres
APP_ENV=local
APP_LOG_LEVEL=debug

Then elsewhere in app you could run:

from config import config

print(config.env)           # local
print(config.log_level)     # 20
print(config.db.host)       # localhost
print(config.db.port)       # 5432
print(config.db.user)       # postgres
print(config.db.password)   # postgres

Usage

Config sources

PKonfig implements several config sources out of the box. All config sources implement Mapping protocol and default values could be set up during initialization.

Environment variables

The most common way to configure application is environment variables. To parse environment variables and store values in multilevel structure class Env could be used. Common pattern is naming variables with multiple words describing the exact purpose more precise: PG_HOST, PG_PORT and REDIS_HOST, REDIS_PORT could be treated as two groups:

• PG
  • HOST
  • PORT
• REDIS
  • HOST
  • PORT

PKonfig respects this convention so that Env has two optional arguments:

• delimiter string that will be used to split configuration levels taken from keys;
• prefix string that is used to identify keys that are related to the given app and omit everything else.

from os import environ
from pkonfig import Env


environ["APP_OUTER"] = "foo"
environ["APP_INNER_KEY"] = "baz"
environ["NOPE"] = "qwe"

source = Env(delimiter="_", prefix="APP", some_key="some")

print(source["outer"])          # foo
print(source["inner"]["key"])   # baz
print(source["nope"])           # raises KeyError

Env ignores key cases and ignores all keys starting not from prefix. To change this behaviour set prefix to None. In this case you will get all key value pairs:

from os import environ
from pkonfig import Env

environ["NOPE"] = "qwe"

source = Env(prefix=None)

print(source["nope"])   # qwe

DotEnv

In the same manner as environment variables DotEnv files could be used. DotEnv requires file name as a string or a path and also accepts delimiter and prefix optional arguments. missing_ok argument defines whether DotEnv raises exception when given file not found. When file not found and missing_ok is set DotEnv contains empty dictionary.

from pkonfig import DotEnv


config_source = DotEnv("test.env", delimiter="_", prefix="APP", missing_ok=True)

Ini

INI files are quite common and class Ini is build on top of configparser.ConfigParser:

from pkonfig import Ini

storage = Ini("config.ini", missing_ok=False)
print(storage["bitbucket.org"]["User"])  # hg
print(storage["bitbucket.org"]["ServerAliveInterval"])  # 45

In case when config.ini:

[DEFAULT]
ServerAliveInterval = 45

[bitbucket.org]
User = hg

Ini also accepts missing_ok argument to ignore missing file. Most of ConfigParser arguments are also accepted to modify parser behaviour.

Json

Json class uses json.load to read given JSON file and respects missing_ok argument:

from pkonfig import Json


storage = Json("config.json", missing_ok=False)

Yaml

To parse YAML files PyYaml could be used wrapped with Yaml class:

from pkonfig import Yaml

storage = Yaml("config.yaml", missing_ok=False)

Toml

TOML files are parsed with tomli wrapped with Toml helper class:

from pkonfig import Toml


storage = Toml("config.toml", missing_ok=False)

Source order

Any source for BaseConfig should implement Mapper protocol. So it is easy to implement custom or combine existing implementations. Recommended way to combine multiple sources of configs is ChainMap:

from collections import ChainMap
from pkonfig import Env, DotEnv, Yaml


config_source = ChainMap(
    DotEnv("test.env", missing_ok=True),
    Env(),
    Yaml("base_config.yaml")
)

In this example we created ChainMap that looks for key until finds one in the given mappers sequence. The first one source for configs is test.env file that might not exist and could be used for local development only. Environment variables are used as the second one config source. Dotenv file will be preferred source in this example. The last one source is base_config.yaml that should exist or FileNotFoundError exception raised.

You can customize source order in this way or even create your own logic implementing Mapper protocol.

Config

To implement application config class user should inherit from pkonfig.config.Config class and define required fields:

from pkonfig import Config


class AppConfig(Config):
    foo: float
    baz: int


storage = {"foo": "0.33", "baz": 1}
config = AppConfig(storage)

print(config.foo)   # 0.33
print(config.baz)   # 1

To build more granular config structure EmbeddedConfig class is used:

from pkonfig import Config, EmbeddedConfig


class Inner(EmbeddedConfig):
    key: str


class AppConfig(Config):
    inner = Inner()
    foo: float
    baz: int


storage = {
    "foo": "0.33", 
    "baz": 1, 
    "inner": {"key": "value"}
}
config = AppConfig(storage)

print(config.inner.key)   # value

Multilevel Config

Grouping might be useful when there are lots of config parameters. To achieve this EmbeddedConfig class should be inherited:

from pkonfig import DotEnv, Config, EmbeddedConfig


class PgConfig(EmbeddedConfig):
    host: str
    port: int = 5432


class RedisConfig(EmbeddedConfig):
    host: str
    port: int = 6379


class AppConfig(Config):
    pg = PgConfig()
    redis = RedisConfig()


config = AppConfig(
    DotEnv(".env", delimiter="__", prefix="APP")
)

print(config.pg.host)       # db_host
print(config.pg.port)       # 6432
print(config.redis.host)    # redis

.env content:

APP__PG__HOST=db_host
APP__PG__PORT=6432
APP__REDIS__HOST=redis

In this example we customized delimiter with two underscores, default is '_'.

Aliases

All Config fields accept alias argument. When storage class searches for config attribute in its source either attribute name is used or alias when it is set.

config.py:

from pkonfig import DotEnv, EmbeddedConfig, Config, Int, Str


class HostConfig(EmbeddedConfig):
    host: str
    port: int
    user: str
    password = Str(alias="pass")


class AppConfig(Config):
    pg = HostConfig(alias="db")
    foo_baz = Int(alias="my_alias")


config = AppConfig(DotEnv(".env", delimiter="__"))

.env content:

APP__DB__HOST=db_host
APP__DB__PORT=6432
APP__DB__PASS=password
APP__DB__USER=postgres
APP__MY_ALIAS=123

In this example storage will seek in dotenv file parameters named by given alias. Elsewhere in an app:

from config import config


print(config.foo_baz)       # 123
print(config.pg.password)   # password

PKonfig fields

All simple Python data types are implemented in field types: Bool, Int, Float, Str, Byte, ByteArray. All fields with known type converted to descriptors during class creation. Fields in Config classes may be defined in several ways:

Using types:

from pathlib import Path
from pkonfig import Config


class AppConfig(Config):
    foo: str
    baz: int
    flag: bool
    file: Path

Using default values:

from pathlib import Path
from pkonfig import Config


class AppConfig(Config):
    foo = "some"
    baz = 1
    flag = False
    file = Path("some.text")

Given values will be used as default values.

Using PKonfig fields directly

from pkonfig import Config, PathField, Str, Int, Bool


class AppConfig(Config):
    foo = Str()
    baz = Int()
    flag = Bool()
    file = PathField()

Caching

All PKonfig field types are Python descriptors that are responsible for type casting and data validation. In most cases there is no need to do this job every time the value is accessed. To avoid undesirable calculations caching is used. So that type casting and validation is done only once during Config object initialization. In case when configuration may change during application lifecycle user may disable this behaviour:

from pkonfig import Config, Int


class AppConfig(Config):
    attr = Int(no_cache=True)

In given example attr will do type casting and validation every time this attribute is accessed.

Default values

If value is not set in config source user can use default value. None could be used as default value:

from pkonfig import Config, Int, Str


class AppConfig(Config):
    int_attr = Int(None)
    str_attr = Str(None)

config = AppConfig({})
print(config.str_attr)    # None
print(config.int_attr)    # None

When None is default value field is treated as nullable.

Field nullability

To handle type casting and validation fields should not be nullable. In case None is a valid value and should be used without casting and validation option nullable could be set:

from pkonfig import Int, Config


class AppConfig(Config):
    int_attr = Int(nullable=True)

config = AppConfig(dict(int_attr=None))
print(config.int_attr)    # None

In this example None comes from storage and type casting is omitted.

Custom descriptor or property

from pkonfig import Config


class AppConfig(Config):
    flag = True
    baz = "test"
    
    @property
    def value(self):
        return self.flag and self.baz == "test" 


config = AppConfig({})
print(config.value)  # True

Custom field types

User can customize how field validation and casting is done. The recommended way is to implement validate method:

from pkonfig import Config, Int


class OnlyPositive(Int):
    def validate(self, value):
        if value < 0:
            raise ValueError("Only positive values accepted")


class AppConfig(Config):
    positive = OnlyPositive()

Custom type casting is also available. To achieve this user should inherit abstract class Field and implement method cast:

from typing import List
from pkonfig import Field

class ListOfStrings(Field):
    def cast(self, value: str) -> List[str]:
        return value.split(",")

Available fields

Builtin Python types has appropriate Field types:

• bool -> Bool
• int -> Int
• float -> Float
• Decimal -> DecimalField
• str -> Str
• bytes -> Byte
• bytearray -> ByteArray

The only reason to use this types directly is customising field nullability and cache policy.

PathField

Basic path type that is parental for other two types and is used when you define field using pathlib.Path. This type raises FileNotFoundError exception during initialization if given path doesn't exist:

from pkonfig import Config, PathField


class AppConfig(Config):
    mandatory_existing_path = PathField()
    optional_path = PathField(missing_ok=True)

In given example field optional_path may not exist during initialization.

File

File inherits PathField but also checks whether given path is a file.

Folder

Folder inherits PathField and does checking whether given path is a folder.

EnumField

This field uses custom enum to validate input and cast it to given Enum:

from enum import Enum
from pkonfig import Config, EnumField


class UserType(Enum):
    guest = 1
    user = 2
    admin = 3


class AppConfig(Config):
    user_type = EnumField(UserType)


config = AppConfig({"user_type": "admin"})
print(config.user_type is UserType.admin)  # True

LogLevel

LogLevel field is useful to define logging level through configs. LogLevel accepts strings that define log level and casts that string to logging level integer value:

import logging
from pkonfig import Config, LogLevel


class AppConfig(Config):
    some_level = LogLevel()
    another_level = LogLevel()


config = AppConfig(
    {
        "some_level": "info",
        "another_level": "Debug",
    }
)

print(config.some_level)        # 20
print(config.another_level)     # 10

print(config.another_level is logging.DEBUG)     # True

Choice

Choice field validates that config value is a member of the given sequence and also does optional type casting:

from pkonfig import Config, Choice


class AppConfig(Config):
    one_of_attr = Choice([10, 100], cast_function=int)


config = AppConfig({"one_of_attr": "10"})
print(config.one_of_attr == 10)  # True

config = AppConfig({"one_of_attr": "2"})    # raises TypeError exception

When cast_function is not given raw values from storage are used.

DebugFlag

DebugFlag helps to set widely used debug option. DebugFlag ignores value case and treats 'true' string as True and any other value as False:

from pkonfig import Config, DebugFlag


class AppConfig(Config):
    lower_case = DebugFlag()
    upper_case = DebugFlag()
    random_string = DebugFlag()


config = AppConfig(
  {
    "lower_case": "true",
    "upper_case": "TRUE",
    "random_string": "foo",
  }
)
print(config.lower_case)        # True
print(config.upper_case)        # True
print(config.random_string)     # False

Types to Fields mapping

All fields for BaseConfig children classes are converted to descriptors internally. Class pkonfig.config.DefaultMapper defines how field types will be replaced with descriptors. This mapping is used by default:

{
    bool: Bool,
    int: Int,
    float: Float,
    str: Str,
    bytes: Byte,
    bytearray: ByteArray,
    Path: PathField,
    Decimal: DecimalField,
}

When field type is not found in this mapper it is ignored and won't be taken from storage source while resolving.

User can modify default mapper giving dictionary of types and appropriate fields:

from decimal import Decimal
from pkonfig import Config, DefaultMapper, DecimalField


class AppConfig(Config):
    _mapper = DefaultMapper({float: DecimalField})
    foo: float

config = AppConfig(dict(foo=1/3))
assert isinstance(config.foo, Decimal)  # True

Per-environment config files

When your app is configured with different configuration files and each file is used only in an appropriate environment you can create a function to find which file should be used:

from pkonfig import Env, Yaml, Config, Choice


CONFIG_FILES = {
    "prod": "configs/prod.yaml",
    "staging": "configs/staging.yaml",
    "local": "configs/local.yaml",
}


def get_config_file():
    class _Config(Config):
        env = Choice(
          ["prod", "local", "staging"], 
          cast_function=str.lower,
          default="prod"
        )
    
    _config = _Config(Env())
    return CONFIG_FILES[_config.env]

get_config_file uses environment variables and predefined config files pathes to check whether APP_ENV var is set, validate this variable and return appropriate config file name. Then actual application configuration:

from collections import ChainMap
from pkonfig import Env, Yaml, Config, Choice


CONFIG_FILES = {
    "prod": "configs/prod.yaml",
    "staging": "configs/staging.yaml",
    "local": "configs/local.yaml",
}


def get_config_file():
    class _Config(Config):
        env = Choice(
          ["prod", "local", "staging"], 
          cast_function=str.lower,
          default="prod"
        )
    
    _config = _Config(Env())
    return CONFIG_FILES[_config.env]


class AppConfig(Config):
    env = Choice(
        ["prod", "local", "staging"], 
        cast_function=str.lower,
        default="prod"
    )
    ...

storage = ChainMap(
  Env(),
  Yaml(get_config_file()),
)
config = AppConfig(storage)