The Service runner had a patch ( https://github.com/wikimedia/service-runner/commit/f552b183ce9beaab7dff9417cad2c9c7d90f1872 ) to support generating the Dockerfile. Should be something like:

runner generate testing

Output would be a DockerFile.

For cache of requirements, I guess it is up to the developer to define the behavior they want. But I guess for a npm repo they would want something like:

COPY package.json ./
RUN npm install

Is that part of the image building step or is that already testing the software? :-]

For production, npm install --production should be used. Since npm install installs developments dependencies as well (which we don't want to have in the final image), this should be used for testing only.

The content from the DockerFile gets data from the repository package.json section "deploy":

"deploy": {
  "node": "6.9.1",
  "target": "debian",
  "dependencies": {
    "_all": [
      "librsvg2-dev"
    ]
  }
}

So one can easily use a specific NodeJS version (in this case 6.9.1 but could be system to use the distribution one).

Dependencies can be various system packages (under _all). Can have different version based on distro name (ubuntu, debian, an example is Graphoid).

For production we have the service::packages puppet define:

class mathoid::packages {
    service::packages { 'mathoid':
        pkgs     => ['librsvg2-2'],
        dev_pkgs => ['librsvg2-dev'],
    }
}

Removing Continuous-Integration-Infrastructure , better tracked via the epic project Release Pipeline.