Platform building blocks

Before we get into the details of how to extend the Allura platform, perhaps it would be smart to explain some of the big pieces and why they are there.

We wanted Allura tools to be fast, we needed them to scale, and we had some complex requirements for data storage and extensibility. So, we needed a fast, flexible, and easy to use data persistence system.

We were very impressed by the general message architecture of Roundup, but we wanted to extend it from just email messages to include scm commits, and we added a message bus (RabbitMQ which we'll talk about in a second), to make it fast.

We were also impressed by the flexibility of Roundup's Hypertable system in allowing for ad-hock ticket schema additions.

It definitely seemed like something we wanted in a next generation forge, because we wanted app tools to be able to:

• create and version their own document types,
• extend existing document structures,
• and to mange document revisions, access control lists, and other platform level data.

In spite of the power and flexibility of the Roundup HyperTable implementation, we had some concerns about performance and scalability.

Fortunately several of the Allura authors used MongoDB in rewriting the download flow of SourceForge.net, and knew that it could handle huge loads (we saturated a 2gb network connection on the server with 6% cpu utilization).

We also knew that MongoDB's flexible replication system would allow us to build the forge in such a way that we could easily provide a package of all project data to developers concerned about lock-in.

Not only that but Rick Copeland had built a couple of custom Object Non-Relational Mappers (ONRMs?) before, including one for MongoDB, and he whipped up Ming, which backed on MongoDB and gave us exactly what we needed.

As I mentioned before we also needed a fast, flexible message bus and queuing system. RabbitMQ was (lightning) fast, (shockingly) flexible, but not super easy to use. Fortunately we didn't have to roll our own wrapper here, as the Python community already whipped up Carrot, and Celery, which made working with the RabbitMQ based AMQP bus a LOT easer.

Application Tools

Writing a tool for Allura is as simple as defining a few controllers to handle particular URL's, templates to render pages, and defining the schemas of any Allura document types that your tool requires.

When you write Allura tools, you'll get lots of stuff for free:

• Search-ability of your Artifacts
• Artifact versioning for accountability and transparency
• Ability to extend existing Artifacts
• Reuse central User/group/permission management
• A robust and flexible permissions system
• Access to a real-time event publishing system