A library for simplifying the configuration of Python applications at all stages of deployment.

Opset is a config manager that let you manage your configuration via YAML file or environment variables. The general principle of Opset is that you want to hold your secrets and manage your configurations via configuration files when doing local development and via environment variables when your app is deployed. It is however possible to also handle local development through environment variables if the developer sees fit.

With Opset you define everything that can be tweaked with your application in one Pydantic model. This way the developers and integrators working with your code will know exactly what setting they can change on your code base. You can then overwrite the default config with a local config stored in a file called local.yml, this file is aimed to be used for local development by your developers and let them easily manage a configuration file that fits their development need. Finally, you can also have environment variables that have a matching name to your config that will overwrite your config, letting you use your config in a deployed environment without having your secret written down in a config file. Opset aims to reconcile the ease of use of a config file with the added security of environment variables.

This library is available on PyPI under the name Opset. You can install with pip by running pip install opset.

1. Lexicon
2. Architecture Overview
   1. Loading the config for unit tests
3. Usage Guide
   1. Making the difference between null and empty
   2. Opset + Google Cloud Secret Manager
   3. Naming your config sections
   4. Controlling your entry points
4. Example Configuration file
   1. default.yml
   2. local.yml
   3. unit_test.yml
   4. Example Logging Configuration values
   5. Log Processors
5. Support for unit tests
   1. setup_unit_test_config
   2. mock_config
6. Contributing and getting set up for local development

There are three possible config files

The content of the default config is loaded first, and if any settings are redefined in local.yml, the default values from the model are overwritten by local.yml.

Environment variables will apply after the local.yml overwrite of the config settings if they have a matching name. To do so, the environment variable must be named in the following way:

{APP_NAME_ALL_CAPS}_{SECTION}_{SETTING}

So for the application my-small-project if we wanted to overwrite the setting port from the section app, your environment variable would need to be named like this:

MY_SMALL_PROJECT_APP_PORT

It is also possible to have nested sections, so following the example above, if you wanted to override the value of api.weather.host you could do so using the following environment variable:

MY_SMALL_PROJECT_API_WEATHER_HOST

To create a new configuration you need to first implement a pydantic model based on OpsetSettingsModel to represent your configuration. It can contain sub models all based on OpsetSettingsModel. Then you can initialize your configuration by instantiating the Config class. Your config object will be available in the config attribute of the opset config.

Opset will also check for local.yml. The location needs to be specified when instantiating the Config object. Your local.yml should be added to .gitignore.

A basic Opset setup will look like this:

from opset import OpsetSettingsMainModel, Config


class MyConfig(OpsetSettingsMainModel):
    host: str
    port: int = 8080


config = Config("my-app", MyConfig, "my_app.config").config

You would then import your new config variable where needed in your app.

The Config class takes the following arguments:

The configuration is stored in YAML and follows the YAML standard. As such, it makes a distinction between null keys and empty keys.

app:
  # this setting's value is declared but not defined
  # it will be set to None when accessed unless it is overwritten in local.yml or in an environment variable
  api_key: null
  # this setting's value is set to an empty string
  log_prefix: 

The config object is initiated once you create the Config object, before that, trying to get read a value from the config will throw an exception. It is very important to have a good idea of what the entry points are in your application and to create your Config object as early as possible in your application to avoid issues.

You cannot instantiate Config more than one time, so make sure the code handling your configuration is only ran once.

You need to install opset with the extras gcp in order to use this feature.

Opset is able to fetch secrets from Google Cloud Secret Manager. You need to be authenticated using gcloud CLI or setting up a service account.

The config value should respect on of these formats

• opset+gcp://projects/<my_project>/secrets/<my_secret>
• opset+gcp://projects/<my_project>/secrets/<my_secret>/versions/<my_version>

Example

database:
  host: opset+gcp://projects/dev-3423/secrets/db_host

It is also possible to create a file .opset.yml in your project to create mapping for project name. For instance, with the following config.

gcp_project_mapping:
  dev: dev-3423

Opset will be able to map the project name like this.

opset+gcp://projects/dev/secrets/db_host -> opset+gcp://projects/dev-3423/secrets/db_host

This file is typically defined by developers for their own development and local usage of the app. This file may contain secrets and as such it must be added to the .gitignore file.

Opset also provides functionality for configuring the logging handlers for your project, this uses structlog in the background. This is provided through the aforementioned load_logging_config function. If you choose to use this functionality, you will need to add some more values to your configuration files, and you can find an example of such values here:

logging:
  date_format: "iso"  # strftime-valid date format, e.g.: "%Y-%M-%d", or "iso" to use the standard ISO format
  use_utc: True  # Use UTC timezone if true, or local otherwise
  min_level: DEBUG  # Minimum level to display log for
  colors: False  # Use colors for log display, defaults to False
  disable_processors: False  # Disables log processors (additional info at the end of the log record)
  logger_overrides: # overwrite min log level of third party loggers
    googleapiclient: ERROR
  json_format: False  # Whether the logs should be formatted as json. Defaults to False.
  json_event_key: "event"  # The key to use for the event in the json format. Defaults to "event"

Since we are using structlog you can use the Processor feature to add additional info to your log records, this can be useful to add a request ID, or the hostname of the machine to all your log records without having to pass anything to your logging calls.

To use this simply define any processors you want by inheriting from the BaseProcessor class of opset and pass an instance to the load_logging_config on your opset config call:

import logging

from flask import Flask
from opset import BaseProcessor, Config, OpsetSettingsMainModel, OpsetLoggingConfig, load_logging_config

from my_app.request_context import get_request_id


class MyConfig(OpsetSettingsMainModel):
    host: str
    port: int = 8080
    logging: OpsetLoggingConfig


class RequestContextProcessor(BaseProcessor):
    def __call__(self, logger, name, event_dict):
        event_dict["request_id"] = get_request_id()
        return event_dict


config = Config("my_app", MyConfig, "my_app.config", setup_logging=False).config  # Defer the logging setup
load_logging_config(config.logging, custom_processors=[RequestContextProcessor()])  # Pass your custom processors

app = Flask(__name__)


@app.route("/")
def root():
    logging.info("This will include the request ID!")
    return "OK"

A processor receives the logger object, the logger name and most importantly the event_dict which contains all the info of the log record. So simply add to the event_dict in your processor and return it.

In local development processors can add some unnecessary noise to the log output, so they can be disabled by setting logging.disable_processors to True in your local.yml.

By default, Opset enables the built-in HostNameProcessor, which adds the machine hostname to log records. It can be disabled by passing use_hostname_processor=False in the load_logging_config call.

Since we are using python's logging library, you can use custom log handlers to customize how and where the information is logged when using the logger.

To use this simply define any log handlers you want by inheriting from the Handler class of logging and overwriting the emit method, and pass an instance to the load_logging_config call:

import logging

from flask import Flask
from opset import Config, OpsetSettingsMainModel, OpsetLoggingConfig, load_logging_config
from logging import Handler
import json


class MyConfig(OpsetSettingsMainModel):
    host: str
    port: int = 8080
    logging: OpsetLoggingConfig


class LocalFileHandler(Handler):
    def __init__(self):
        Handler.__init__(self)

    def emit(self, record):
        """
        Will log the record in the root log.json file
        """
        with open("log.json", "w") as fp:
            json.dump(record.msg, fp)


config = Config("my_app", MyConfig, "my_app.config", setup_logging=False).config  # Defer the logging setup
load_logging_config(config.logging, custom_handlers=[LocalFileHandler()])  # Pass your custom handlers

app = Flask(__name__)


@app.route("/")
def root():
    logging.info("Log me in a local file!")
    return "OK"

The handler receives the record object, containing all the log information that was processed by the processors. The handler can chose what to do with that information, should it be to log it in a local file, send it to a blob storage, send it to an external tool (ex: Sentry)

Opset support unit testing to make sure you can handle the special cases that may come up in your application configuration during unit testing.

To setup your config for unit tests you will want to import your opset_config object and call setup_unit_test on it. This function takes a dictionary of all the configuration you want to overwrite. This call should happen only once in your unit test setup.

The mock_config contextmanager on the opset config is used to temporarily overwrite your configuration. Like setup_unit_test you pass a dictionary of the configurations you want to overwrite.

To set yourself up for development on Opset, make sure you are using poetry and simply run the following commands from the root directory:

make install