Skip to content
/ beehive Public

A version-controlled view of HackGT deployments.

License

Notifications You must be signed in to change notification settings

HackGT/beehive

Repository files navigation

Why?

Beehive is our deployment abstraction layer that lets us:

  1. clearly and declaratively represent the current state of our deployment,
  2. roll back to known working states if necessary,
  3. have a clear audit trail when a deployment goes awry,
  4. separate our deployment implementation details from normal developers,
  5. and most importantly minimize manual changes to the infrastructure (through SSH, etc.).

We bascially get most of this for free with git, the rest is handled by Beekeeper, a system we wrote to parse the files here and translate them into one deployment configuration.

The goal here is to let everyone be able to deploy their apps and not have one person's app destroy another! It's like each app lives in it's own world.

Sounds Great! Tell me how.

It's pretty simple, the main unit here is the domain name on which your app runs. Say you want to run an app that tells you if your phone is on silent or not on the url ismichaelsphoneoff.hack.gt, simply make a file called ismichaelsphoneoff.yaml in the top directory of this repository.

On the other hand you could provide this service for many people by making a folder called isyourphoneoff and placing a michael.yaml and an andrew.yaml file there. This would make michael.isyourphoneoff.hack.gt and andrew.isyourphoneoff.hack.gt.

NOTE: so far we can't go deeper with more subdomains, if you want support for this consider contributing ;0

Config File Format

# a simple git ref on master:
git: https://github.com/HackGT/phonehome.git

# a more complex git
git:
    remote: https://github.com/HackGT/phonehome.git
    branch: master # optional
    rev: faceb00c  # optional, takes precedence over branch

# port the service will bind to, will be exposed through the `PORT` env var
# if none is given a port will be decided for you
target_port: 3000

# the port to expose externally (for http this is port 80)
# (optional) defaults to port 80 if nothing is given.
port: 80

# a list of dependent services, currently supports `mongo` and `postgres`
# this is optional
wants:
    mongo:
        env: CUSTOM_ENV_VAR
    postgres: true # the env var will default to `POSTGRES_URL`

# a list of secrets to be given through environment variables.
# (optional) more on this in the secrets section...
secrets:
  - SESSION_SECRET

# environment variables passed to the program (optional)
env:
  EMAIL_FROM: "HackGT Team <[email protected]>"
  EMAIL_HOST: smtp.sendgrid.net
  EMAIL_PORT: 465
  PRODUCTION: true
  # turns into '["this can be","a list"]'
  WHITELIST:
    - this can be
    - a list
  # turns into '{"what":"is","going":"on"}'
  WEIRD_DATA:
    what: is
    going: on

At least one type of git is required. This will find the Docker Hub image of the same name (hackgt/phonehome) and pull it. If rev is specified it will pull the image with the tag of that rev, if branch not specified or master it will pull the latest image, if the branch is specified it will pull latest-${branch name}.

The deployment also gets these environment variables set by default:

  1. PORT the port the service should bind to.

Parting words

Hopefully this design will become more unified in the future. But for now our priority is getting to a working state that can sustain itself for a long time to come.