Workflow to the test

Time to start cracking


I’m for this exercise (see previous post) aiming for an Ember app with Node as a web backend. Simplicity will be paramount when it comes to functionality so as to focus on scripting the workflow. Be aware of that this post is more of a log of the activities, and therefore slightly lengthy and full of mistakes and dead ends.

# Install ember command-line interface
npm install -g ember-cli
# Create new ember app
ember new hello-ember
cd hello-ember

Note: Installation of ember is not part of the Friendly requirements, but I’m thinking it will not be required to run/develop the app later.

Try it out
# Start app
ember server

Opening a web browser to http://localhost:4200 shows a tawdry Welcome to Ember.js text. Not so exhilirating, but it works.

A nicer surprise is that editing the code, well, changing the hello text, and saving automatically updates my browser view. Live reload seems to be built in.

Gulp it

Since live reload seems built in, all we need to at this stage is have gulp run the ember start command, right?

# Register gulp as a development dependency
npm install gulp --save-dev
npm install gulp-shell --save-dev

Type up a complicated gulpfile.js:

var gulp = require('gulp');
var shell = require('gulp-shell')

gulp.task('up', shell.task(['ember server']));

Success! Typing gulp up starts the ember app and refreshing my browser still displays a not-so-impressive site.

Going the docker mile

Why? Primarily for later benefits, being able to use the same workflow for non-node-based applications. I also want to keep dependencies of locally installed tools/versions to a minimum and be able to use off-the-shelf dockerized services.

Not only do I want to use docker, I want to be able to start multiple docker images, such as both a web server and a database. In comes Docker Compose (formerly known as Fig). Docker allows you to start containers, Docker Compose allows you to start multiple containers with a single command as well as connect these (e.g. web container can easily access database container without reverting to DNS, fixed IP-addresses and the like).

In this case it ought to be sufficient to use a single node container to run our ember example, but for the sake of completeness I will (1) use an existing ember container (simplify our life, remember?) and use docker compose to create a standard for later projects.

Attempt #1


  image: geoffreyd/ember-cli
   - .:/usr/src/app
   - "4200:4200"
   - "35729:35729"
  command: server --watcher polling

Starting with docker-compose up. It works, but… my CPU is running haywire. Not at all my docker experience so far (being impressed by how resource efficient it seems to be). Not OK. Google seems to indicate file watching through VM (boot2docker) might not be the best choice.

Changed command: server --watcher polling to command: server. Much better. Still slow in starting, and cpu-hungry initially, but quickly settles down into non-noticeable CPU usage digit. Unfortunately, now Sublime is running away. Trying this…no good - still unacceptable slowness in the Sublime awesomeness. Not only that - live reload of ember has stopped working, together with manual reload.

Attempt #2

Instead of hurling myself into figuring out why file watching for Ember and Sublime Text has issues with Docker (or boot2docker), let’s try another approach.

This time I’ll use nodemon, a tool that will restart my node server whenever files change. Not as sexy as live reload, but it will do the job. It also has the benefit of working for both server and client updates. Fine tuning can come later, if warranted.


  image: fluciotto/nodemon
   - .:/usr/src/myapp
   - "80:3000"
  working_dir: /usr/src/myapp
  command: nodemon app.js

Note the addition of a new server script. app.js is in this case the simplest of express.js server scripts (web server that serves the static ember files):

var express = require('express');
var app = express();

app.use(express.static(__dirname + '/dist'));

app.get('*', function(req, res) {
    res.sendfile('./dist/index.html'); // load our public/index.html file

var server = app.listen(3000, function () {

  var host = server.address().address;
  var port = server.address().port;

  console.log('Example app listening at http://%s:%s', host, port);

Trying again, docker-compose up. OK, pointing my browser to boot2docker shows the now familiar greeting, using Sublime to change the greeting…and nothing happens. Hmm, looking at console output from nodemon suggests that it doesn’t monitor recursively into folders. Changing a file in the current directory confirms this, with a nice reload. Thinking one step further, I realize that I’ve pointed nodemon to the ember dist directory, if I make changes I need to trigger an ember build from my source files to the dist directory if I’m using this method. I also note that boot2docker (the virtual machine running Docker on my Mac) is pushing 70% cpu. As a comparison, replacing nodemon with standard node, boot2docker runs at a low 2% with the same example. We’re not there just yet.

Attempt #3

New tack. Using the volume feature of Docker together with monitoring file changes is likely workable, but seems to have a few pitfalls that I will leave as-is for now. Instead of sharing a file directory between the main Docker container and the host, I’ll deploy changed files to the container. My first idea is to perform a deployment as one would tend to perform to production environments, but since that would (1) require me to modify the Docker container to support some deployment mechanism, (2) would likely be slightly slower than one would want in a development workflow and (3) might work very differently for different technologies and produce different result. A simpler, I think, version would be to simply copy modified files to the container using Docker infrastructure. This is not so neat as I would like, but has the benefit of potentially using the various features of development tooling for different technologies (such as ember-cli above).

Unfortunately, Docker does not provide a copy-file-to-running-container feature out of the box. I could setup ssh on the container, but again, that would require modifying the container which I don’t like. Instead I’ll try a method suggested by some nice people on the Internet where the docker exec feature can be hacked to pipe a stream from the host to a command being executed on the container, e.g. such as writing to a file. In essence cat file | docker exec -i containername sh -c 'cat > file'.

This sadly took longer than expected (a smell), but I’ve now created a gulp module that can consume file changes and store these on a Docker container (disclaimer: will not work on Windows) and modified my gulpfile.js to (1) start by copying the dist-directory (compiled ember-app) to the container and (2) monitor for file changes to the app, when found ember build will be executed, producing modified files in the dist-directory (3) which are monitored and copied to the container.

Not as pretty as I wanted, but can surely be improved over time. Testing the setup, I can now modify files on my host, which after a few seconds (due to the build step put in between) will show in the browsed app. CPU is purring along on a sleepy level. Success, albeit way to many hours later.

See GitHub for the final solution.

  • I can use gulp up, and Ctrl-C as I wanted
  • I almost kept my Friendly requirements (currently using ember build on the host, but should be easily remedied)
  • Potential to use technology-specific features in the Docker containers
  • No modification of the Docker images necessary
  • Docker-compose in play for future additions of services
  • More complicated, and slower than wanted
  • OS-dependent code (Docker-compose only seems to have a cli, requiring use of host executed commands)
  • Going outside somewhat of both Docker and gulp paradigms

Add project to GitHub

Following the guide:

git init
git add .
git commit -m 'First commit'
git remote add origin
git remote -v
git push origin master

Deploy to staging with Travis

OK, next step. I’ve been promised a fully-automated deploy to a production-like testing environment. Chosen tooling is Heroku and Travis, the first due to familiarity, the second due to hip-factor, both due to being free and cloud-based for this particular project.

Get Travis talking
  • Sign-up with GitHub…check
  • Enable the helloember-project…check
  • .travis.yml…hmm, google? Wait a minute. I already have a travis-file in my directory, likely created when generating the ember skeleton. Oh, well…check
  • Push to GitHub (changed README)…check
  • Travis build running…check
  • Waiting…green light! For what, one may wonder - it seems there are some jshint checks being run. As I haven’t changed the ember code, I’m not that surprised that it still checks out.
Get Travis deploying
  • According to the manual, this should be a breeze. Good to know.
  • I need to install travis cli to encrypt my heroku key. Another unfriendly requirement, although hopefully a one-time-thing. I suddenly need ruby installed (or rather, verified that it is still working)
  • Trusting that travis will not secretly steal my key: travis encrypt $(heroku auth:token) --add deploy.api_key…done. My travis.yml now has an api-key in it. Adding heroku as the provider.
  • The documentation seems to suggest that there is no “deploy-to-production” button, but rather that one can use different branches (e.g. deploy-to-production = branch). Assuming that it will work that way already, and naming the master branch as staging.
  • Fingers crossed, pushing changes to GitHub…Fail. No stash found, "App not found." (wrong app "helloember-staging"?). Might be I have to create the app on Heroku first? Surprisingly no clear result from Google on this.
  • App created manually on heroku (with just name, nothing else), building again…hmm, retry did not seem to work, commiting small change instead…starting to realize that cloud-based CI requires some patience while debugging issues…Deployed successfully!
  • Browsing to…Application Error.
  • heroku logs --app helloember-stagingsh: 1: ember: not found. Ah, seems as if heroku is instructed by my package.json to start the application with ember (which one should only do in a development environment).

Let’s take a bit of a timeout, instead of just following the various guides. My app is basically an ember skeleton, which in development mode compacts into html and javascript in the dist folder. Checking the travis-file, there are no instructions on how to package the app for deployment, and there are no instructions for heroku (apart from ember server) on how to run it. Somewhere along the line, I probably should be the one to decide which web server heroku should use, and which files should be served. For the development workflow I used a Docker container with nginx to serve the dist-folder. It seems a viable simple strategy, except for nginx/Docker. Let’s switch these for a simple expressjs app instead.

  • Added before_deploy: npm build to travis.yml (e.g. ember build as specified in package.json)
  • Changed start-command in package.json to node app.js (ready-made in one of the attempts above).
  • Commit again…Improvement. The app is now running, but giving 404:s, it looks like the ember build step wasn’t executed properly.
  • Commiting explicit ember build && a ls dist…build works as expected, but still 404
  • Changing sendfile to sendFile, app.get(*) to app.get(/) and adding logs…another type of error. Sigh.
  • heroku run bash --app helloember-staging followed by ls. Hmm, no dist folder at all.
  • Adding skip_cleanup: true to the deploy section did the trick. Extra points for fixing it via my iPhone on the subway (edit online on GitHub, commit).

Deploy to production

If all is well, all I now need to do is create a new app on Heroku (check) branch to a new production branch on GitHub. Testing…works!

Verifying the workflow

  1. Inspiration hits. Run to computer. Check
  2. Open a terminal window, type cd project && subl . && gulp up to start the project environment. Check
  3. Make changes to a text file or two. Save. Check
  4. View results, e.g. open web browser to app UI or unit-test report. Check
  5. Type Ctrl-C in the terminal window to bring the environment down. Check
  6. Commit to Github. Check
  7. Verify that staging environment is updated. Check
  8. Branch and verify that production environment is updated. Check