Our Platforms

Praekelt uses two main platforms for the bulk of engineering work:

  1. Django

    We use Django for websites, mobi sites, responsive sites, mobi HTML5 apps.

  2. Vumi

    We use our Vumi platform for SMS, USSD and other messaging protocols.

Use of any other platform must be approved by our engineering management, and may need to be hosted separately from our usual environments, so please get this sorted out prior to commencing development on a project.

Django

You should use the latest stable release of Django as of the start of the project unless otherwise specified. Pin that version in your requirements so that the project won’t break by accidently being deployed on a newer version without testing.

We strongly recommend you use our sample django-skeleton as a starting point for Django projects, as it has the layout and configurations we use, so that deployment will be smooth.

We deploy Django in the following stack:

  • Ubuntu Server (current LTS release)
  • haproxy for load balancing where appropriate
  • nginx Web server that we use to run our Django applications in production.
  • gunicorn is the application server we’re using to service our Django applications with
  • supervisord manages your processes and allows us to ensure that your applications are a) running and b) come up at boot should the server be restarted for whatever reason.
  • postgresql for storing our data, usually for Django projects

For development, you can simplify this, and for QA we won’t bother about haproxy but the rest of the stack will be required for QA so we recommend you keep your dev environment as close to this as you can.

Notes:

  • We manage hostnames in nginx, because there may be multiple QA and live hostnames so don’t use Django’s ALLOWED_HOSTS.
  • Make use of pip and virtualenv
  • Avoid using CachedStaticFilesStorage, or generating CSS/JS automatically as this breaks load balanced environments.
  • There are specific requirements for logging. Please see the Logging section for more information.

Vumi

Vumi is a scalable messaging engine which we use for SMS, USSD and other messaging protocols.

Vumi Go is a hosted version of Vumi. Where Vumi gives you the tools to build large scale messaging applications, Vumi Go provides you with a working environment that is already integrated into numerous countries.

Apps can be written for Vumi Go, to power messaging campaigns or information systems. These apps can be written in Javascript to run in a sandboxed environment (which is our preferred option) or in Python.

See the Vumi Go documentation as well as the JS Sandbox toolkit documentation for writing apps.

Logging

All logging must happen to a standardized path, under /var/praekelt/logs. If your app has a large amount of files (celery, worker, and app, for instance), write each of these to a named path under /var/praekelt/logs/appname/foo.log, otherwise just /var/praekelt/logs/foo.log is sufficient.

Logs always end with the extension .log. .err is not valid. If you need to write out error logs, either use the name foo-error.log, or write your supervisord configuration with the redirect_stderr option.

The basis for this requirement is to ease debugging (hunting logs in 7 different directories is never fun, and causes issues when under time pressure), simplifies log rotation, and allows them to easily fit within our automatic log collection and indexing system.