Massaging the Pony: Message Queues and You

Description

Massaging the Pony: Message Queues and You

Presented by Shawn Rider

Message queues are a potential solution for any site that needs to facilitate robust asynchronous operations on your website including carrying out intensive or long-running actions or synchronizing off-site services. We will look at how PBS has used Celery and RabbitMQ to provide more reliable service and how to set up a robust message queue solution in a cloud hosting environment.

Abstract

In maintaining several different site projects written in Django over the past four years, the PBS Education team repeatedly found the need for the ability to create asynchronous processes that could handle long-running actions and mission-critical synchronization. We needed a way to handle tasks such as generating reports on large data sets and executing remote processes on fussy third-party services. The clear solution to the problems we encountered was to implement a robust message queue solution.

Message queues allow sites to execute processes outside the normal HTTP request cycle and then deliver the results of those processes in a variety of ways. After a survey of all available message queue solutions we settled on a combination of Celery and RabbitMQ to create a reliable, always-available system that can handle mission-critical tasks. We can easily develop against a message queue in our development builds, and we have an infrastucture based in a cloud hosting service that allows us to rely on the message queue without reservation.

In this talk will will look at what a message queue does for your site, how to implement a message queue in your Django code, and how to set up a reliable message queue infrastructure on your servers. We will show examples from some of our sites at PBS so it is clear how message queues work and to present some common use cases that warrant a message queue solution. We will also discuss some of the design patterns that made it easy to switch to message queue based processing for discreet tasks, as well as some of the things we learned that have made the solution even easier to implement for new tasks.