judoscale 1.1.0

Official Python adapter for Judoscale — the advanced autoscaler for Heroku

Judoscale

This is the official Python adapter for Judoscale. You can use Judoscale without it, but this gives you request queue time metrics and job queue time (for supported job processors).

It is recommended to install the specific web framework and/or background job library support as "extras" to the judoscale PyPI package. This ensures that checking if the installed web framework and/or background task processing library is supported happens at dependency resolution time.

Supported web frameworks

• Django
• Flask
• FastAPI

Supported job processors

• Celery (with Redis as the broker)
• RQ

Using Judoscale with Django

Install Judoscale for Django with:

$ pip install 'judoscale[django]'

Add Judoscale app to settings.py:

INSTALLED_APPS = [
    "judoscale.django",
    # ... other apps
]

This sets up the Judoscale middleware to capture request queue times.

Optionally, you can customize Judoscale in settings.py:

JUDOSCALE = {
    # LOG_LEVEL defaults to ENV["LOG_LEVEL"] or "INFO".
    "LOG_LEVEL": "DEBUG",

    # API_BASE_URL defaults to ENV["JUDOSCALE_URL"], which is set for you when you install Judoscale.
    # This is only exposed for testing purposes.
    "API_BASE_URL": "https://example.com",

    # REPORT_INTERVAL_SECONDS defaults to 10 seconds.
    "REPORT_INTERVAL_SECONDS": 5,
}

Once deployed, you will see your request queue time metrics available in the Judoscale UI.

Using Judoscale with Flask

Install Judoscale for Flask with:

$ pip install 'judoscale[flask]'

The Flask support for Judoscale is packaged into a Flask extension. Import the extension class and use like you normally would in a Flask application:

# app.py

from judoscale.flask import Judoscale

# If your app is a top-level global

app = Flask("MyFlaskApp")
app.config.from_object('...')  # or however you configure your app
judoscale = Judoscale(app)


# If your app uses the application factory pattern

judoscale = Judoscale()

def create_app():
    app = Flask("MyFlaskApp")
    app.config.from_object('...')  # or however you configure your app
    judoscale.init_app(app)
    return app

This sets up the Judoscale extension to capture request queue times.

Optionally, you can override Judoscale's own configuration via your application's configuration dictionary. The Judoscale Flask extension looks for a top-level "JUDOSCALE" key in app.config, which should be a dictionary, and which the extension uses to configure itself as soon as judoscale.init_app() is called.

JUDOSCALE = {
    # LOG_LEVEL defaults to ENV["LOG_LEVEL"] or "INFO".
    "LOG_LEVEL": "DEBUG",

    # API_BASE_URL defaults to ENV["JUDOSCALE_URL"], which is set for you when you install Judoscale.
    # This is only exposed for testing purposes.
    "API_BASE_URL": "https://example.com",

    # REPORT_INTERVAL_SECONDS defaults to 10 seconds.
    "REPORT_INTERVAL_SECONDS": 5,
}

Note the official recommendations for configuring Flask.

Using Judoscale with Celery and Redis

Install Judoscale for Celery with:

$ pip install 'judoscale[celery-redis]'

:warning: NOTE 1: The Judoscale Celery integration currently only works with the Redis broker.

:warning: NOTE 2: Using task priorities is currently not supported by judoscale. You can still use task priorities, but judoscale won't see and report metrics on any queues other than the default, unprioritised queue.

Judoscale can automatically scale the number of Celery workers based on the queue latency (the age of the oldest pending task in the queue).

Setting up the integration

To use the Celery integration, import judoscale_celery and call it with the Celery app instance. judoscale_celery should be called after you have set up and configured the Celery instance.

from celery import Celery
from judoscale.celery import judoscale_celery

celery_app = Celery(broker="redis://localhost:6379/0")
# Further setup
# celery_app.conf.update(...)
# ...

judoscale_celery(celery_app)

This sets up Judoscale to periodically calculate and report queue latency for each Celery queue.

If you need to change the Judoscale integration configuration, you can pass a dictionary of Judoscale configuration options to judoscale_celery to override the default Judoscale config variables:

judoscale_celery(celery_app, extra_config={"LOG_LEVEL": "DEBUG"})

:warning: NOTE: Calling judoscale_celery turns on sending task-sent events. This is required for the Celery integration with Judoscale to work.

Judoscale with Celery and Flask

Depending on how you've structured your Flask app, you should call judoscale_celery after your application has finished configuring the Celery app instance. If you have followed the Flask guide in the Flask documentation, the easiest place to initialise the Judoscale integration is in the application factory:

def create_app():
    app = Flask(__name__)
    app.config.from_object(...) # or however you configure your app
    celery_app = celery_init_app(app)
    # Initialise the Judoscale integration
    judoscale_celery(celery_app, extra_config=app.config["JUDOSCALE"])
    return app

Judoscale with Celery and Django

If you have followed the Django guide in the Celery documentation, you should have a module where you initialise the Celery app instance, auto-discover tasks, etc. You should call judoscale_celery after you have configured the Celery app instance:

from celery import Celery
from django.conf import settings
from judoscale.celery import judoscale_celery

app = Celery()
app.config_from_object("django.conf:settings", namespace="CELERY")
app.autodiscover_tasks()
# Initialise the Judoscale integration
judoscale_celery(app, extra_config=settings.JUDOSCALE)

Using Judoscale with RQ

Install Judoscale for RQ with:

$ pip install 'judoscale[rq]'

Judoscale can automatically scale the number of RQ workers based on the queue latency (the age of the oldest pending task in the queue).

Setting up the integration

To use the RQ integration, import judoscale_rq and call it with an instance of Redis pointing to the same Redis database that RQ uses.

from redis import Redis
from judoscale.rq import judoscale_rq

redis = Redis(...)
judoscale_rq(redis)

This sets up Judoscale to periodically calculate and report queue latency for each RQ queue.

If you need to change the Judoscale integration configuration, you can pass a dictionary of Judoscale configuration options to judoscale_rq to override the default Judoscale config variables:

judoscale_rq(redis, extra_config={"LOG_LEVEL": "DEBUG"})

Judoscale with RQ and Flask

The recommended way to initialise Judoscale for RQ is in the application factory:

judoscale = Judoscale()

def create_app():
    app = Flask(__name__)
    app.config.from_object("...") # or however you configure your application
    app.redis = Redis()

    # Initialise the Judoscale integration for Flask
    judoscale.init_app(app)

    # Initialise the Judoscale integration for RQ
    judoscale_rq(app.redis)

    return app

Then, in your worker script, make sure that you create an app, which will initialise the Judoscale integration with RQ. Although not required, it's useful to run the worker within the Flask app context. If you have followed the RQ on Heroku pattern for setting up your RQ workers on Heroku, your worker script should look something like this:

from rq.worker import HerokuWorker as Worker

app = create_app()

worker = Worker(..., connection=app.redis)
with app.app_context():
    worker.work()

See the run-worker.py script for reference.

Judoscale with RQ and Django

The Judoscale integration for RQ is packaged into a separate Django app.

You should already have judoscale.django in your INSTALLED_APPS. Next, add the RQ integration app judoscale.rq:

INSTALLED_APPS = [
    "judoscale.django",
    "judoscale.rq",
    # ... other apps
]

By default, judoscale.rq will connect to the Redis instance as specified by the REDIS_URL environment variable. If that is not suitable, you can supply Redis connection information in the JUDOSCALE configuration dictionary under the "REDIS" key.

Accepted formats are:

• a dictionary containing a single key "URL" pointing to a Redis server URL, or;
• a dictionary of configuration options corresponding to the keyword arguments of the Redis constructor.

JUDOSCALE = {
    # Configuring with a Redis server URL
    "REDIS": {
        "URL": os.getenv("REDISTOGO_URL")
    }

    # Configuring as kwargs to Redis(...)
    "REDIS": {
        "HOST": "localhost",
        "PORT": 6379,
        "DB": 0
    }
}

If you are using Django-RQ, you can also pull configuration from RQ_QUEUES directly:

RQ_QUEUES = {
    "high_priority": {
        "HOST": "...",
        "PORT": 6379,
        "DB": 0
    },
}

JUDOSCALE = {
    # ... other configuration options
    "REDIS": RQ_QUEUES["high_priority"]
}

:warning: NOTE: Django-RQ enables configuring RQ such that different queues and workers use different Redis instances. Judoscale currently only supports connecting to and monitoring queue latency on a single Redis instance.

Development

This repo includes a sample-apps directory containing apps you can run locally. These apps use the judoscale adapter, but they override API_BASE_URL so they're not connected to the real Judoscale API. Instead, they post API requests to https://requestinspector.com so you can observe the API behavior.

See the README in a sample app for details on how to set it up and run locally.

Contributing

judoscale uses Poetry for managing dependencies and packaging the project. Head over to the installations instructions and install Poetry, if needed.

Clone the repo with

$ git clone [e-mail]:judoscale/judoscale-python.git
$ cd judoscale-python

Verify that you are on a recent version of Poetry:

$ poetry --version
Poetry (version 1.3.1)

Install dependencies with Poetry and activate the virtualenv

$ poetry install --all-extras
$ poetry shell

Run tests with

$ python -m unittest discover -s tests