The Sweet Sensation of Automation

A guide to automated npm publishing and GitHub pushing using Heroku

Last year I created a little npm module called all-the-package-names. It's a simple array of all the existing package names in the npm registry. It can be used to find if a package name is available, or to find all packages with names matching a given pattern.

Here's a one-liner to try it out:

npm i -g trymodule && trymodule all-the-package-names

This module was easy to create, but it is by nature impossible to keep up to date, as hundreds of new packages are added to the npm registry every day.

I wanted to remove myself from the publishing equation, so I set up a scheduled task on Heroku that runs every hour, fetches new names from the registry, commits changes to git, and publishes to npm. To enable this, I had to figure out how to securely perform all these operations within the confines of a Heroku dyno:

This post covers the steps required in setting up such an automation scheme. It is assumed that you have basic working knowledge of Heroku, npm, and GitHub.

heroku create my-npm-bot

Unlike a typical Heroku app, this app doesn't have a web process. There's no web server listening on a port and serving requests. It's just an app on a faceless computer in the cloud that will wake up on occasion to do your bidding.

Heroku has these awesome open source things called buildpacks that give you complete control of your app's build process. A buildpack is a script that is executed on your app's code to prepare it to run on the Heroku platform. When you deploy a Node.js app, for example, Heroku's Node buildpack is used to download node and npm, run npm install, etc.

In most cases, Heroku users don't need to know about buildpacks at all. Heroku detects the language or framework of your web app and automatically chooses the right buildpack for you. But sometimes you need to exercise control over the app's build process, and that's when buildpacks come in handy.

For the purposes of this app, three buildpacks are required:

To add these three buildpacks to your app, run the following commands:

heroku buildpacks:add -i 1 https://github.com/github-modules/github-buildpack
heroku buildpacks:add -i 2 https://github.com/nice-registry/npm-buildpack
heroku buildpacks:add -i 3 heroku/nodejs

The i stands for index, allowing you to specify the order in which the buildpacks are executed. In many cases this order is important, but for our purposes the order probably doesn't matter, as none of these buildpacks are relying on functionality or build artifacts from another.

GitHub provides a simple web interface for creating auth tokens. Create a token with repo access, then add it to your Heroku app's config:

heroku config:set GITHUB_AUTH_TOKEN=YOUR_TOKEN

The next time you push your app to Heroku, the GitHub buildpack you just added will detect the presence of GITHUB_AUTH_TOKEN in your app's environment, and use it to create a .netrc file.

~/.netrc is a plaintext file used to store credentials for accessing remote machines. It's an old school unix thing, so it's widely supported by tools like curl, git, and the heroku toolbelt,

Here's what an entry for GitHub looks like:

machine github.com login YOUR_TOKEN password x-oauth-basic

If you pass the --netrc flag to curl (or -n for short), it will automatically read credentials from your ~/.netrc file, allowing you to easily make authenticated requests:

curl -n https://api.heroku.com/apps
curl -n https://api.github.com/users/zeke/repos

Now that the GitHub credentials are securely stored, git commands like git push will be automatically authenticated.

npm doesn't have a fancy web interface for creating tokens, nor does it support the .netrc format. It is possible, however, to find your personal token by looking in the .npmrc file in your home directory.

Here's a one-liner for reading that token:

cat ~/.npmrc | head -1 | sed 's/.*=//g'

Add the token to your Heroku app's config:

heroku config:set NPM_AUTH_TOKEN=YOUR_TOKEN

Like the GitHub buildpack, the npm buildpack will detect the NPM_AUTH_TOKEN environment variable and use it to write a .npmrc file into the app directory. Any npm commands that require authentication (like publish and whoami) will now be authenticated automatically when they are run within the app directory.

To figure out if everything's working, deploy your app:

git commit --allow-empty "testing npm and github auth"
git push heroku master

Then open up a remote bash shell containing your app's code and environment:

heroku run bash
cat .npmrc
cat .netrc

If everything is configured properly, you'll see your credentials in the .netrc and .npmrc files, and you should be able to run authenticated commands like npm whoami and git push.

Tip: If you try to git push your app to Heroku and get an "Everything up to date" message from git, it's because nothing in your source code has changed. To force-push the repo to Heroku, create an empty commit:

git commit --allow-empty -m "empty commit"
git push heroku master

Once the authentication is set up, the next step is to write a release script to be executed periodically on Heroku.

mkdir -p scripts
touch scripts/release
chmod +x scripts/release

Next, add the release script to package.json, so it can be executed as an npm script:

{
  "name": "all-the-package-names",
  "scripts": {
    "release": "scripts/release"
  }
}

Here's a simplified version of the release script used by all-the-package-names. For the full script, check out scripts/release on the GitHub repo.

#!/usr/bin/env bash 
 
set -x            # print each command before execution 
set -o errexit    # always exit on error 
set -o pipefail   # don't ignore exit codes when piping output 
set -o nounset    # fail on unset variables 
 
# set up the repo 
git clone https://github.com/zeke/all-the-package-names
cd all-the-package-names
npm run build
npm test
 
# bail if no changes are present 
[[ `git status --porcelain` ]] || exit
 
count=$(cat names.json | wc -l)
git add names.json
git config user.email "zeke@sikelianos.com"
git config user.name "Zeke Sikelianos"
git commit -m "$count package names"
npm version minor -m "bump minor to %s"
npm publish
git push origin master --follow-tags
 
# clean up 
cd ..
rm -rf all-the-package-names

Tip: Heroku's node buildpack will install dependencies from package.json by default. If your build script requires devDependencies to be installed too, set the following in your app environment:

heroku config:set NPM_CONFIG_PRODUCTION=false

There are a few ways to manually test a script on Heroku. You can run it remotely:

heroku run npm run release

or you can shell into your app and run one-off commands inside the shell:

heroku run bash
npm run release

Once you've got a working release script, it's time to automate!

Heroku has a very useful (and free) add-on called Scheduler that allows you to configure one-off tasks to run against your app. There's a web-based GUI for scheduling tasks, and they can be run every ten minutes, every hour, or once a day. It's like cron, but much easier to use.

To install and open Scheduler:

heroku addons:create scheduler
heroku addons:open scheduler

The scheduler interface is pretty simple. You specify a command to run and the frequency at which to run it.

Heroku Scheduler UI

From the Heroku docs on Dyno Sleeping:

Free dynos will sleep when a web dyno receives no web traffic for a period of time. In addition, if a free dyno exceeds a quota of 18 hours of activity during a 24 hour window, it will be forced to recharge.

This means you get 18 hours of compute time per day for free! And because this app doesn't have a web process, it only needs to be "awake" for a few minutes a day for the short amount of time required to run the release task every hour.

Even for apps with long-running tasks at shorter intervals, the free dyno would still probably be adequate. A task that runs every 10 minutes and takes 5 minutes to complete only amounts to 12 hours of compute time per day, which is still under the 18 hour limit.

If your build process takes a long time or needs to run more frequently, a hobby dyno is adequate for most tasks and will only set you back $7 a month.

Once your automated publishing workflow is finished, you can kick back and let the machines do the work, then watch your contribution graph turn a very dark shade of green...

GitHub Contributions