piku/README.md

piku

The tiniest Heroku/CloudFoundry-like PaaS you've ever seen, inspired by dokku.

Motivation

I kept finding myself wanting an Heroku/CloudFoundry-like way to deploy stuff on a few remote ARM boards and my Raspberry Pi cluster, but since dokku still doesn't work on ARM and even docker can be overkill sometimes, I decided to roll my own.

Project Status/ToDo:

From the bottom up:

• Prebuilt Raspbian image with everything baked in
• chroot/namespace isolation (tentative)
• Proxy deployments to other nodes (build on one box, deploy to many)
• Support Node deployments
• Support Clojure/Java deployments
• Sample Go app
• Support Go deployments
• CLI command documentation
• nginx support
• Testing with pre-packaged uWSGI versions on Debian Jessie (yes, it was painful)
• Support barebones binary deployments
• Complete installation instructions (see INSTALL.md, which also has a draft of Go installation steps)
• Installation helper/SSH key setup
• Worker scaling
• Remote CLI commands for changing/viewing applied/live settings
• Remote tailing of all logfiles for a single application
• HTTP port selection (and per-app environment variables)
• Sample Python app
• Procfile support (wsgi and worker processes for now, web processes being tested)
• Basic CLI commands to manage apps
• virtualenv isolation
• Support Python deployments
• Repo creation upon first push
• Basic understanding of how dokku works

Using piku

piku supports a Heroku-like workflow, like so:

• Create a git SSH remote pointing to piku with the app name as repo name (git remote add paas piku@server:app1)
• git push paas master your code
• piku determines the runtime and installs the dependencies for your app (building whatever's required)
  • For Python, it segregates each app's dependencies into a virtualenv
  • For Go, it defines a separate GOPATH for each app
• It then looks at a Procfile and starts the relevant workers using uWSGI as a generic process manager
• You can then remotely change application settings (config:set) or scale up/down worker processes (ps:scale) at will.

Later on, I intend to do fancier dokku-like stuff like reconfiguring nginx, but a twist I'm planning on doing is having one piku machine act as a build box and deploy the finished product to another.

Supported Platforms

piku is intended to work in any POSIX-like environment where you have Python, uWSGI and SSH, i.e.: Linux, FreeBSD, Cygwin and the Windows Subsystem for Linux.

As a baseline, this is currently being developed on an original, 256MB Rasbperry Pi Model B.

Since I have an ODROID-U2, a bunch of Pi 2s and a few more ARM boards on the way, it will be tested on a number of places where running x64 binaries is unfeasible.

But there are already a few folk using piku on vanilla x64 Linux without any issues whatsoever, so yes, you can use it as a micro-PaaS for 'real' stuff. Your mileage may vary.

Supported Runtimes

piku will support deploying apps written in Python, Go, Clojure (Java) and Node (see above).

FAQ

Q: Why piku?

A: Partly because it's supposed to run on a Pi, because it's Japanese onomatopeia for 'twitch' or 'jolt', and because I know the name will annoy some of my friends.

Q: Why Python/why not Go?

A: I actually thought about doing this in Go right off the bat, but click is so cool and I needed to have uWSGI running anyway, so I caved in. But I'm very likely to take something like suture and port this across, doing away with uWSGI altogether.

Q: Does it run under Python 3?

A: It should. click goes a long way towards abstracting the simpler stuff, and I tried to avoid most obvious incompatibilities (other than a few differences in subprocess.call and the like). However, this targets Python 2.7 first, since that's the default on Raspbian. Pull requests are welcome.

Q: Why not just use dokku?

A: I use dokku daily, and for most of my personal stuff. But the dokku stack relies on a number of x64 containers that need to be completely rebuilt for ARM, and when I decided I needed something like this (March 2016) that was barely possible - docker itself is not fully baked for ARM yet, and people are still trying to get herokuish and buildstep to build on ARM.