Strange Loop

Running Heroku on Heroku

The Heroku application platform makes deploying, running and scaling applications incredibly easy.

Traditionally these apps have been Ruby web applications. But as both the platform and its users mature, we are seeing the complexity of hosted apps increase, with more complex infrastructure systems running on Heroku.

Today, nobody is more interested in running infrastructure on Heroku than Heroku itself, as self-hosting offers massive benefits and is a fascinating engineering puzzle to boot.

We will first discuss the concept of self-hosting and why it such an interesting computer science problem, and a vital property of many systems. Compilers, revision control systems and application platforms all exhibit similar properties of bootstrapping, cross-compiling, and avoiding circular dependencies.

Then we will take a look at the more interesting self-hosted components of Heroku such as the distributed process management application that is responsible for keeping track of of everything running in the cloud including itself, and the distributed application compiler that used to be a server farm but now is little more than a Heroku app that can even compile itself for releasing new versions.

All of this will show how working towards a self-hosted platform results in comprehensive consistency assurance and gains in efficiency, noble goals for such a complex software system.

Noah Zoschke

Heroku, Inc.

Noah Zoschke is a lead engineer at Heroku, a Ruby cloud platform as a service. He spends his time doing infrastructure development on the Heroku runtime, a distributed code compilation, process management and process execution system responsible for running and scaling over 150,000 applications.