envargs

envargs gives you an easy way of specifying which environment variables to look for as program arguments/config. This library aims at providing an as similar as possible experience to that of using Python's argparse package.

Installing

pip install envarguments

Why envargs?

As most containerized applications take most or at least part of their runtime config from environment variables I always seemed to reimplement the same parsing utilities for new projects. envargs.EnvParser gives you a no-dependencies quick way of specifying which variables your program needs to run, which are mandatory and which are not, and their data types. A single call to parse_env and your registered environment variables are decoded, ready to be used.

Usage

Registering interesting variables and getting a hold of them in your running program is very straight forward:

import envargs

NUM_WORKERS = "NUM_WORKERS"
RETENTION_POLICY = "RETENTION_POLICY"
RETENTION_POLICY_DEFAULT = "keep-alive"

parser = envargs.EnvParser()
parser.add_variable(NUM_WORKERS, type=int, default=1)
parser.add_variable(RETENTION_POLICY, default=RETENTION_POLICY_DEFAULT)
namespace = parser.parse_env()

num_workers = namespace.num_workers
retention_policy = namespace.retention_policy
...

Register interesting variables with add_variable and call parse_env() to obtain a populated namespace. The namespace will have one property per found registered variable, named the same as the environment variable in lower case. Variable NUM_WORKERS thus gets the field name namespace.num_workers.

Generating a description text

In addition, to further clarify how your application uses environment variables, an envargs.EnvParser provides a helpful description property that can help determine if you have set up your variables as expected. After calling add_variables on your EnvParser instance for each interesting variable, the parsers description field will contain a helpful text which describes all added variables, if they're required, their types, and their default values (if any).

import envargs
import logging

LOGGER = logging.getLogger(__name__)

env_parser = envargs.EnvParser()
env_parser.add_variable("NUM_WORKERS", type=int)
env_parser.add_variable("RETRY_COUNTER", required=False, type=int)
env_parser.add_variable("OVERLOAD_LIMIT", default="high")

LOGGER.debug(env_parser.description)

This text can be added to an argparse.ArgumentParser description to show the text easily from the commandline. Note the use of the raw formatter class to avoid stripping line breaks.

import argparse
import envargs

env_parser = envargs.EnvParser()
env_parser.add_variable("WORK_DIST_STRATEGY", default="round-robin")
env_parser.add_variable("CPU_OVERLOAD_LIMIT_PERCENT", type=int, default=80)

arg_parser = argparse.ArgumentParser(
   formatter_class=argparse.RawDescriptionHelpFormatter,
   description="""
This is the best program, ever written, clearly.

Below, some info on environment variables used \/
""" + env_parser.description)
...

Variable properties

An environment variable, when registered, can be given a set of properties, all available as keyword arguments given to EnvParser.add_variable(...). If no keyword arguments are given, just the name of the environment variable itself, it will be assumed that:

1. The variable is required (raising a RequiredError in case it's missing)
2. The variable is a string (meaning no special decoding of the string gotten from os.getenv(...)) is done
3. There is no default value

Read the docstring on EnvParser.add_variable to see all available options. Below is an example of a non-required boolean variable with a default value:

import envargs

def do_throttling(acceptor_throttling: bool):
    ...

env_parser = envargs.EnvParser()
env_parser.add_variable("ACCEPTOR_THROTTLING", type=bool, required=False, 
                        default=False)
ns = env_parser.parse_env()

do_throttling(ns.acceptor_throttling)

The above will attempt to find the environment variable ACCEPTOR_THROTTLING and if it doesn't exist, fall back on False as a default value. The namespace returned by parse_env will therefore always be populated with the acceptor_throttling property.

An optional environment variable that isn't present at the time of calling parse_env will still yield an attribute in the returmed namespace, set to None.

import envargs

env_parser = envargs.EnvParser()
env_parser.add_variable("OPTIONAL_SETTING", required=False)
namespace = env_parser.parse_env()

# Will not raise an assertion error
assert namespace.optional_setting is None

Parsing booleans

To simplify things for this library, booleans can be expressed in only a few selected ways:

TRUTH_SET = {"true", "True", "1"}
FALSE_SET = {"false", "False", "0"}

These definitions are available at envargs.envargs.TRUTH_SET/FALSE_SET. Deviating from these values for a boolean variable will result in a ParseError:

import envargs
import os

from envargs.errors import ParseError

SYSTEM_ONLINE = "SYSTEM_ONLINE"

# Set the env var to something other than what's defined in the 
# TRUTH_SET/FALSE_SET
os.environ[SYSTEM_ONLINE] = "YES"

env_parser = envargs.EnvParser()
env_parser.add_variable(SYSTEM_ONLINE, type=bool)

# This statement raises a ParseError
try:
    env_parser.parse_env()
except ParseError as e:
    print(e)