I have been trying out Codeship and Heroku for continuous deployment of an AngularJS application I writing at the moment. The app was created using Yeoman and uses bower and grunt. Initially I thought this seemed like a really good setup as Codeship was free to use and I was quickly able to configure this to build my AngularJS project and it offered the ability to add a deployment step after the build. There were even many PaaS providers to choose from (Heroku, S3, Google App Engine etc). However I seem to have become a bit stuck with getting the app to run on Heroku.
The problem started from the fact that all the documentation suggested that I remove the /dist path from my .gitignore so that this directory is published to Heroku post build. This was mainly from documentation that talked about publishing to Heroku from a local machine, but I figure this is all Codeship is doing under the hood anyway. I didn't want to do this as I don't believe I should be checking build output into source control. The /dist folder was added to .gitignore for a good reason. Furthermore, this kind of defeats the point of having a CI server somewhat, as I might as well just push the latest build from my machine.
After some more digging around I found out that I could add a postinstall step to my packages.json file such as bower install && grunt build which would re-run the build on Heroku and hence repopulate all the bower dependencies (something else they wanted me to check in to source control!) and the dist directory.
After giving this a try it became apparent that I would need to add bower and grunt as dependencies in packages.json, which meant moving them from devDependencies which is where they should belong!
So I now seem to be stuck. All I want to do is publish my build artefacts (/dist) the dependencies (/bower_components) and the server.js file that will run the site. Does anyone know how to achieve this with Heroku and Codeship? Alternatively has anyone had any success with this using different tools. I am looking for something that is free and I am willing to accept that it will not be production stable (won't scale to multiple servers etc), but this is fine for now as all I want to do is continuously deploy the app for internal testing and to be able to share the output with non-technical members of my team so we can discuss features we'd like to prioritise etc.
Any advice would be greatly appreciated.
Thanks
I found two ways to get this to work.
Use the mbuchetics Heroku build pack. This works by basically re-building the app once it has been pushed to Heroku.
There were a few tricks I had to employ still to make this work. In Gruntfile.jstwo new tasks needed to be configured called heroku:production and heroku:development. This is what the buildpack executes to build the app. I initially just aliased the main build task, but found that the either the buildpack or Heroku had a problem with running jshint so in the end I copied the build task and took out the parts that I didn't need.
Also in packages.json I had to add this:
"scripts": {
"postinstall": "bower cache clean && bower install"
}
This made sure the bower_components were available in Heroku.
This allowed me to keep the .gitignore file in tact so that the 'binaries' in the dist directory and the dependencies in the bower_components directory were not committed into source control.
This is basically re-building the app once it is on Heroku and I generally prefer to use the same 'binaries' throughout the entire build and deployment pipeline. That way I know that the same code that was built, is the same code that was tested and is the same code that was deployed.
It also slows down the deployment as you have to wait for the app to build twice.
Not being satisfied with the fact I was building my app twice, I tried using a Custom Script pipeline in CodeShip instead of the pre-existing Heroku one. The script basically modified the .gitignore file to allow the dist folder to be committed and then pushed to the Heroku remote (which leaves the code on the origin remote unaffected by the change).
I ended up with the following bash script:
#!/bin/bash
gitRemoteName="heroku_$APP_NAME"
gitRemoteUrl="git@heroku.com:$APP_NAME.git"
# Configure git remote
git config --global user.email "you-email@example.com"
git config --global user.name "Build"
git remote add $gitRemoteName $gitRemoteUrl
# Allow dist to be pushed to heroku remote repo
echo '!dist' >> .gitignore
# Also make sure any other exclusions dont apply to that directory
echo '!dist/*' >> .gitignore
# Commit build output
git add -A .
herokuCommitMessage="Build $CI_BUILD_NUMBER for branch $CI_BRANCH. Commited by $CI_COMMITTER_NAME. Commit hash $CI_COMMIT_ID"
echo $herokuCommitMessage
git commit -m "$herokuCommitMessage"
# Must merge the last build in Heroku remote, but always chose new files in merge
git fetch $gitRemoteName
git merge "$gitRemoteName/master" -X ours -m "Merge last build and overwrite with new build"
# Branch is in detached mode so must reference the commit hash to push
git push $gitRemoteName $(git rev-parse HEAD):refs/heads/master
This only require a single build of the app and deploys the same binaries that were tested during the test phase.
I've used this script quite a few times now and it seems relatively stable. However one issue I know of is that when a new pipeline is created there will be no code on the master branch so this script fails when it tries to do the merge from the heroku remote. At the moment I get around this by doing an initial push of the master branch to Heroku before kicking off a build, but I imagine there is probably a better Git command I could run along the lines of; 'only merge this branch if it already exists'.