In this post, I’m going to share my thoughts about an on-premises private cloud configuration for collaborative software development. These are just some conceptual notes, but I’ll try to double back later and provide step-by-step directions for setting up the system I’m going to describe here.
The substrate I prescribe is Joyent’s Triton, which is completely open source, so a finished product can be deployed to the Joyent public cloud or in a colocated datacenter. I won’t go into depth about standing up Triton, because Joyent has graciously described how to stand up Triton on affordable consumer hardware.
As a footnote, Triton was originally known as SmartDataCenter, so you’ll see
many references to
sdc in the context of Triton. Just build a mental regex.
UPDATE: I didn’t mean to spread FUD! While the open source version of Triton does include an operations portal, it’s worth checking out the (also) open source sdc-portal developed by Faithlife. Although it hasn’t been updated in a while, Faithlife’s sdc-portal has some very compelling features and I’d encourage anyone interested in deploying Triton to check it out.
A somewhat little known fact about Triton is that it comes with an LDAP server, which provides a perfect hook for integration with GitBucket. GitBucket is an open source clone of GitHub and by integrating with sdc-ufds (the “Unified Foundational Directory Service”), we can revoke all privileges in the system through the sdc-portal interface. Pretty sweet, huh?
It gets better. Triton exposes the entire datacenter as a single
with sdc-docker, so you can develop and deploy containerized
applications without any painful virtual machine configuration. With Triton, all
of your containers will be running at bare metal speeds and provisioned across
all available compute nodes. It isn’t actually Linux, but it’s close enough.
Can we put some icing on this cake? Okay.
How about a multi-user version of Jupyter Notebook for collaborative,
explorative development? The beautiful part about this is that once you have
Docker configured on your machine to connect to your private Triton
DOCKER_HOST you can just do some variation of:
docker run -d --name jupyter.hub jupyter/jupyterhub jupyterhub
That’s pretty much it, although there are several approaches that can be taken here for authentication (including another LDAP integration). Jupyter is no longer limited to Python, so you can also use Ruby, R, and several other languages if you so choose. You are almost certainly going to want to build your own custom docker image for Jupyter Hub to support your goals, but this is a pretty simple process. Again, I’ll try to provide more detail in another post.
Why would you bother with any of this? Well, the only money you’d spend on this environment would be on hardware that you’d own. It’s not exactly expensive to develop on the Joyent public cloud or host your private repositories on Github, but if you consider the total cost over the lifetime of a business these things can add up pretty quickly. When you start to run the numbers, there is a very real economic argument in favor of this approach.
That’s the general outline! Just food for thought.