Heroku is an awesome platform for hosting web applications, that much is obvious. A few days ago I had another application to deploy on Heroku, but with a different usage profile. The application, a simple breaking news tweeting app, periodically scrapes a popular Israeli forum with breaking headlines, and tweets them - a fairly straightforward task. However, this application has no request-response cycle, and in fact has no open web gateway, just a simple task running periodically, every minute in our case.

Naturally, this task needs to run on a 24/7-available server, not just on a random desktop. Sure I have several VMs I can piggyback this task on, but I wanted to find the way to package this little task properly such that I can deploy it easily on Heroku and forget about the whole thing. Since I'm running a single process on a single Heroku dyno, if I could get it to work, it wouldn't cost a thing.

For asyncronous and scheduled tasks in Python, the obvious solution is to use Celery. The core of the setup is a single Celery worker running a periodical task. Since we only have one worker, and we can't spare another process for the Celery heartbeat process (it'd cost another Heroku dyno which isn't free), we'll use the celery worker process with the -B flag that bundles the worker and the heartbeat into one convenient process.

Celery can't work without a messaging broker, naturally with Heroku we'll use the redistogo:nano plan.

Here's the code for a simple worker, tasks.py:

import logging

from celery import Celery
from celery.task import periodic_task
from datetime import timedelta
from os import environ

REDIS_URL = environ.get('REDISTOGO_URL', 'redis://localhost')

celery = Celery('tasks', broker=REDIS_URL)


def fib(n):
    if n > 1:
        return fib(n - 1) + fib(n - 2)
    else:
        return 1


@periodic_task(run_every=timedelta(seconds=10))
def print_fib():
    logging.info(fib(30))

To wrap it up, you'll need a Procfile with a single line launching the worker:

worker: celery -A tasks worker -B --loglevel=info

I find this setup to be very convenient if I need to deploy a single recurring task, and not care at all about setting up cron jobs or manually configuring deployment environments. Heroku FTW.

All the code, as always, is in a single repo on Github: https://github.com/yuvadm/heroku-periodical. Enjoy!