Serverless Docker

Swarm makes it incredibly easy to run code on your infrastructure. You wrap some code up inside a Docker container, and Swarm will make it run on whatever free resources you have.

But what if those containers could also run other containers on a Swarm? We could put pieces of our application inside containers that are run on-demand on a Swarm. Docker containers can be used as functions from within other applications:

>>> import dockerrun
>>> client = dockerrun.from_env()
>>> client.run("bfirsh/leftpad", ["foo", "5"])
'  foo\n'

Take, for example, running background tasks in a web app. In a traditional architecture, you would have a set of task workers and a message queue to pass work from the web frontends to the task workers.

If your web frontends have access to a Swarm, you can run the task directly on your Swarm:

client.run("tasks/reticulate-splines", detach=True)

To read more about this, check out this blog post.

Examples

• Serverless voting app – A serverless web app
• go-dexec examples
• Funker – an example app that uses Funker to do processing in the background

## Reading

• Introducing dexec – What if the Go os/exec library could containerize?

Tools

• Funker – Functions as Docker containers
• go-dcgi – CGI, but with Docker containers

## Client libraries

• docker-py – Run Docker containers from Python apps
• dockerrun – A simpler interface for running Docker containers in Python (soon to be part of docker-py)
• go-dexec – Like Go os/exec package but for Docker
• dockerode – Run Docker containers from Node.js apps
• docker-java – Run Docker containers from Java apps

Stuff that needs working on

We need your help!

• Make this work with Docker 1.12.
• A proxy that scopes a Docker API so that containers can securely manage and run "child" containers.
• Helpers for injecting the Docker API socket into containers that are run.
• A server for running scheduled / cron jobs as Docker containers on a Swarm.