Building EC2 AMIs for docker-worker

You will need taskcluster credentials with suitable scopes (TODO: find out which scopes, or ideally a single role, and document it/them here) configured in TASKCLUSTER_* environment variables. You can use the taskcluster-shell tool to set them:

$ eval $(taskcluster signin)

You also need the Taskcluster team passwordstore repo properly configured and installed on your system (TODO: presumably also the pass unix utility, rather than just a clone of the repo - update docs accordingly). The deploy scripts require node version >= 12.11.0.

Execute the following command:

$ ./deploy.sh <docker-worker source code path> <build target>

To build docker-worker AMIs. The build target is either app or base. The base image is used to accelarate the process the more common app image. You can also tag a github release with:

$ ./release.sh

To be able to make a GitHub release, you'll need to set the environment variable DOCKER_WORKER_GITHUB_TOKEN with a suitable GitHub access token.

In order for the docker-worker workers of the firefox-ci-tc and community-tc taskcluster deployments to use the newly generated AMIs, the repositories ci-configuration and community-tc-config will need to be updated with the names of the newly generated AMIs.

Building under Vagrant

Install vagrant on your system, and a virtualisation backend, such as Virtual Box.

For example, on macOS:

$ brew install --cask vagrant
$ brew install --cask virtualbox

Install a vagrant provider plugin to match the backend you chose, and install the vagrant-reload plugin:

$ vagrant plugin update
$ vagrant plugin install virtualbox
$ vagrant plugin install vagrant-reload

Install VirtualBox Guest Additions plugin:

$ vagrant plugin install vagrant-vbguest

Clone taskcluster, and then clone docker-worker-deploy inside the workers directory so that docker-worker-deploy and docker-worker are sibling directories on your system, e.g.

$ git clone git@github.com:taskcluster/taskcluster.git
$ cd taskcluster/workers
$ git clone git@github.com:taskcluster/docker-worker-deploy.git

Bring up the vagrant machine:

$ cd docker-worker
$ vagrant up

Install the VirtualBox Guest Additions on the VM and restart it:

$ vagrant vbguest
$ vagrant reload default

SSH into the machine, initialise the /docker-worker-deploy directory, do some more vagrant setup(?)

$ vagrant ssh
$ cd /docker-worker-deploy

Maybe this has already run (see the Vagrantfile inside the workers/docker-worker directory of the taskcluster monorepo) but if you had any failures, you might need to run it again:

$ ./vagrant.sh

At some point I was asked if I would like to keep my existing /etc/default/docker or replace it with the version from apt package docker-ce. This may be because it took me several iterations to get things working, so this might not happen if things work first time. I chose to keep the version I had (option O) since from the diff, the package version looks like just comments, but the version on the file system contained the line:

DOCKER_OPTS="--storage-driver overlay2"

The build.sh script isn't available inside the VM since it is in the docker-worker directory and not the docker-worker-deploy directory. A mistake? Surely you aren't meant to build the docker images directly on your host?

$ ./build.sh

The next instruction said to install yarn as follows; I've no idea why this isn't part of the existing scripts. Perhaps because it is only needed if you want to test, so isn't done by default? Presumably this should be done inside the VM, rather than on the host. If this isn't done already, maybe we can just add it to the existing scripts?

$ curl -o- -L https://yarnpkg.com/install.sh | bash

Install required packages:

$ yarn install

Export credentials for running integration tests:

$ export TASKCLUSTER_ROOT_URL='......'
$ export TASKCLUSTER_CLIENT_ID='......'
$ export TASKCLUSTER_ACCESS_TOKEN='......'
$ export TASKCLUSTER_CERTIFICATE='......'   # not needed if using "permanent" credentials

Run tests:

$ ./node_modules/mocha/bin/mocha