Category Archives: open source

Deploying with Ember.js: a story

What is this all about?

This is the story of how we changed our Ember.js application deployment one night while no one was looking. This is not a strictly technical story, and the details may not matter; what is important is the journey we took and the steps you can follow to get your app deploying however you want.

Table of Contents

  1. What is this all about?
  2. Some Background (feel free to skip)
  3. Our Goals (the whole point of this)
  4. Our Journey
    1. Starts with a plan on a whiteboard
    2. Leads to Ember-cli-deploy
  5. The End
  6. Afterwords

Some Background (feel free to skip)

Our application – Ilios – is an installed solution and the code is in two parts. The API and management code is written in PHP and is installed by medical schools all over the world. Each installation is a little bit different, but they all require and consume our Ember.js frontend application. We don’t ship the frontend with the API codebase, it is provide separately and updated far more frequently. In the past we have shipped an index.html file with links to our AWS Cloudfront distribution containing the assets. This way the assets are stored in a CDN where they can be loaded quickly by users and easily kept up to date by our team without needing to understand each individual installation.

Our Goals (the whole point of this)

In order to take advantage of http/2 push and service workers we need to change the way the frontend is deployed. Our goal is to ship a single archive file containing the frontend assets including the app code, service workers, images, styles etc…

This archive must be tagged in such a way that we can have many versions in the wild at the same time with a default active version. Because the code for our API does change it must be possible to activate versions in such a way that v23 is the default frontend for API v1.1 and v34 is the default frontend for API v1.3.

Once the archive is downloaded and extracted by the API server it must be simple to parse and customize the files so that scripts and styles can be sent to the browser using http/2 push LINK headers and service workers can be placed at the root of our domain (where they must be do be effective).

Because we are constantly adding new features and fixing bugs we need to be able to release versions continuously.


We want to ship a magical box to our customers and we want to ship it all the time.

Our Journey

Starts with a plan on a whiteboard


  1. Build the app
  2. Archive the build artifacts
  3. Tag the archive with a version which combines both a unique ID for the code as well as the API version it works with
  4. Upload the archive to an S3 bucket
  5. Mark the most recent version so it is easy to find


  1. Download and extract the archive at
    • a) the most recent version
    • b) a specific version from the past
    • c) dev build we want to test
  2. Parse the the index file to extract configuration and assets
  3. Serve a modified index.html file to users

Leads to Ember-cli-deploy

This isn’t a surprise ember-cli-deploy has both the builtin tools and a plugin ecosystem to make deploying Ember.js apps very manageable. All we need to do is assemble a list of plugins that can meet our needs. As is the case with many ember addons, our journey really starts with Ember Observer where we find there is a whole category dedicated to plugins for ember-cli-deploy.

Some investigation and perusal of the deploy docs and a few minutes in the #ec-deploy channel in the Ember Slack Team will lead us to some standard choices and well supported solutions.

  • ember-cli-deploy is the foundation we can build on
  • ember-cli-deploy-build builds our app
  • ember-cli-deploy-revision-data can tag and activate versions
  • ember-cli-deploy-s3 uploads our assets and makes them public in an S3 bucket

A search for ‘archive’ in Ember Observer leads us to:

  • ember-cli-deploy-archive which takes our build assets and outputs a single tar archive

These combined knock out several of our needs.

Further investigation of ember-cli-deploy-s3 leads us to it’s companion ember-cli-deploy-s3-index which not only uploads to S3, it also manages versioning and activation.

And while looking around we also stumble upon ember-cli-deploy-json-config which conveniently parses our messy index.html file and outputs nice human and machine readable JSON that will be way easier for our API server to consume.

Looks like we don’t have to write any code at all! Just install and configure some plugins just the way we want and in the end just need to run a few instal commands:

ember install ember-cli-deploy
ember install ember-cli-deploy-archive
ember install ember-cli-deploy-build
ember install ember-cli-deploy-display-revisions
ember install ember-cli-deploy-json-config
ember install ember-cli-deploy-revision-data
ember install ember-cli-deploy-s3-index

Which helpfully creates a config/deploy.js file for us with some useful defaults:

module.exports = function(deployTarget) {
  var ENV = {
    build: {}
    // include other plugin configuration that applies to all deploy targets here

  if (deployTarget === 'development') { = 'development';
    // configure other plugins for development deploy target here

  if (deployTarget === 'staging') { = 'production';
    // configure other plugins for staging deploy target here

  if (deployTarget === 'production') { = 'production';
    // configure other plugins for production deploy target here

  // Note: if you need to build some configuration asynchronously, you can return
  // a promise that resolves with the ENV object instead of returning the
  // ENV object synchronously.
  return ENV;

We just need to configure our AWS info and we will something working. Note: I removed targets here as well as comments for brevity.

module.exports = function(deployTarget) {
  var ENV = {
    build: {}
    's3-index': {
      accessKeyId: '<our-access-key>',
      secretAccessKey: '<our-secret>',
      bucket: '<our-bucket-name>',
      region: '<our-bucket-region>'

  return ENV;

Next step is to try it out:

ember deploy development  --verbose
Pipeline complete

And when we got look in our S3 bucket low and behold there is an build.tar:HASH file in there! We’re really getting somewhere.

That version is nice, but remember we really need to know what API versions this build is compatible with. Oh, I see there is a prefix option we can use.

's3-index': {
  prefix: 'v1.22'

Deploy again and we now have a bucket with

build.tar:HASH (from our first deploy)
  build.tar:NEWHASH (from our second deploy)

That is exactly what the doctor ordered. Check that off our list and lets download that archive and see what it has. Hmmm… it has an index.html file, but no index.json file. Looking back at the --verbose output we can see why

+- didBuild
|  |
|  +- archive
- saving tarball of tmp/deploy-dist to tmp/deploy-archive/build.tar
- tarball ok
|  |
|  +- json-config
- generating `tmp/deploy-dist/index.json` from `tmp/deploy-dist/index.html`
- generated: `tmp/deploy-dist/index.json`
- added `index.json` to `context.distFiles`

The build.tar file is getting generated before the index.json file has been created. Since both of these plugins are doing their work inside the didBuild hook we just need to swap the order they run in. We can find out how to do that in the docs at

Looks like we just add a pipeline object to our deploy config like:

pipeline: {
  runOrder: {
    'archive': { after: 'json-config' },

and… done! running ember deploy development again gives us exactly the build.tar we’re looking for.

The End

Seriously. That’s how the story ends. No brave journey into the darkness of code comments to figure out just how this is supposed to work. It just works. TM. It’s so simple we can just add it as a command at the end of our CI process and never think about it again.


I’m 100% head over heals in love with ember-cli-deploy. I find it incredibly freeing to be able to tinker with our deployment process to find just the right setup without needing to design and script each pice every time. The plugin ecosystem gives me the building blocks I can use to assemble our pipeline in whatever way makes sense so we can test new ideas and strategies fast and with very little friction.

I hope this was a useful overview of one deployment journey. Want to share your own? Tell me about a mistake I made or shower me with praise? Leave a comment here or find me at @iam_jrjohnson.

Improving Summon/EZproxy/Shibboleth Login Experience With Custom JavaScript

If you use Summon’s custom auth banner/VPN banner with Shibboleth and EZproxy, you may have noticed an unfortunate side effect. If you do a search, and then click the banner to log in, you will be redirected to the start page rather than back to your search results.

Fixing this means tapping into their app with custom JavaScript. The front end of Summon is built on AngularJS 1.2.27 and ships with jQuery.  Here’s how we’re currently doing it:

(function () {
  var waitingForInitialLoad = true;

  var fixBannerLink = function () {
    var links = $('.vpnBanner.customAuthBanner div a');
    if (waitingForInitialLoad && links.length === 0) {
      setTimeout(fixBannerLink, 100);
    } else {
      waitingForInitialLoad = false;
      window.addEventListener('hashchange', fixBannerLink);
    links.each(function() {
      if (/^Off the UCSF network/.test(this.text)) {
        $(this).attr('ng-href', '' + encodeURIComponent(location.href));
        $(this).attr('href', '' + encodeURIComponent(location.href));


Most embarrassing hack: The /^Off the UCSF network/ regular expression is of course matching the text of the link. You’ll need to update that to match the text in your own link. (Or find a less embarrassing hack and submit it as a pull request in the GitHub repository.)

Also, our EZproxy URL prefix is hard-coded in there a few times. Change to your own, and make sure you use the qurl URL parameter rather than url.

If you have any questions, feel free to open an issue in the GitHub repository.

Ember CLI, Heroku, and You

Warning, this is old information and way more work than you need to do. The solution we are using now is the excellent heroku-buildpack-ember-cli.

A disclaimer: This is not for use in production. Doing this for a production app would be a bad decision.

The problem: Developers on the Ilios Project need to be able to share their changes with other team members. While it is possible to deploy a static Ember CLI app nearly anywhere, we want to include our mock API so everyone is looking at the same data.

The solution: Use Heroku to host an Ember CLI application running its built in Express server.

Continue reading Ember CLI, Heroku, and You

Introducing Amalgamatic

“Search!” by Jeffrey Beall licensed CC BY-ND 2.0

Academic libraries offer many resources, but users cannot be expected to search in, say, a half dozen different interfaces to find what they’re looking for. So academic libraries typically offer federated search.

Sometimes, a solution is purchased. Many libraries, for example, use 360 Search.

Here at UCSF, we are among the libraries that have built our own federated search. Twice.

Continue reading Introducing Amalgamatic

Production Packer Passwords: Securing the Root User

Packer is a tool for creating identical machine images for multiple platforms from a single source configuration. It is mostly used to create base images for developers using Vagrant. However it’s just as useful for creating virtual machine (VM) images to deploy in production. Using Packer in this way, you can create a consistent starting point for VMs which are then provisioned further with, for example, Puppet or Chef, creating a ready-to-deploy image with your application already installed.

One minor headache for using Packer in this way is how to safely create a root account with a known password without exposing that password in configuration files.

The key to this process is hooking into the scripted install process. For Debian this is known as Preseed. Redhat calls it Kickstart. Most Packer VMs are built with some kind of Preseed/Kickstart file.

Continue reading Production Packer Passwords: Securing the Root User

See your failures: Taking screenshots with PhantomJS running Behat tests

PhantomJS is a headless browser which, when combined with Behat, can run the same tests you can run in a browser like Chrome or Firefox. Headless testing is faster because it doesn’t actually render anything for a user. This makes it ideal for rapid test driven development.  

A downside to headless testing is that when it fails it can be hard to see exactly why.  Figuring out whether it is your code or your test that is causing the issue can be extremely frustrating and you might find yourself asking; Why can’t I just see it?  The frustration mounts when the same test runs just fine via Selenium in a GUI browser.

OK, was that enough buildup?  Are you ready for the good stuff?  Right then, here we go.

Continue reading See your failures: Taking screenshots with PhantomJS running Behat tests