Months ago I decided to write my curriculum using Latex and to use Github to version it (here the repository). I said to myself that it would be beautiful to automatically create the latest cv version as a pdf file whenever I update the source files, maybe as a Github release.
I had the same desire when I started thinking about this blog: wouldn’t be amazing if at each new content, a new blog version was deployed automatically?
Well, that’s the story of this journey: from all manual to automatic flow using Github Actions.
Github Actions
Github Actions it’s a new Github feature allowing users to create CI/CD workflows within the repository.
A workflow is defined in a yaml file and it’s nothing but a list of jobs. Each job is then composed by steps. Lastly, each step is a set of actions. And it’s the action the Github atomic unit: each of these performs an operation (e.g. it checkouts the repository, it creates a release, …).
In a single repository, many workflows can exist and they should be defined in the .github/workflows
folder.
Each workflow is triggered by an event. For example, we could have a workflow triggered whenever a push is done in a branch:
name: Deploy new version # workflow name
on:
push:
branches:
- master
But what happens under the hood? Github has a pool of virtual machines (hosted on Azure). In each of these VMs are installed some special services called runners. When a workflow instance is created, jobs contained are executed by these runners.
Here the documentation.
CV e Latex
How to compile a Latex document? Here the file defining the workflow.
name: Build CV
on:
push:
tags:
- 'v*.*.*'
jobs:
release_cv:
runs-on: ubuntu-latest
steps:
...
push_to_blog:
needs: release_cv
runs-on: ubuntu-latest
steps:
...
This is an excerpt showing the yaml structure. Here we can find the workflow name (Build CV) and the triggering event (a push of a tag). Two jobs are defined as well:
release_cv
: it compiles.tex
files and it creates a new release on Githubpush_to_blog
: pushes the release to the blog repository
Actions composing these jobs are pretty straight forward. The interesting part is about the interaction between jobs. I said before how jobs are executed by (probably different) runners in (probably different) virtual machines. So, potentially, they do not share any kind of data. In this context artifacts come into play. Artifacts are nothing but objects that are saved and retrieved at runtime by different jobs.
In fact, the last action of the first job is the artifact upload:
- name: Upload to virtual env
uses: actions/upload-artifact@v1
with:
name: cv
path: cv_MarcoDenisi.pdf
and the first action of the second job is the download of the same artifact:
- name: Download from virtual env
uses: actions/download-artifact@v1
with:
name: cv
Blog e Hugo
The blog compilation and upload is quite homemade, here the complete version:
name: Deploy new version
on:
push:
branches:
- master
jobs:
deploy_blog:
runs-on: ubuntu-latest
steps:
- name: Checkout master
uses: actions/checkout@v1
- name: Build blog and deploy
run: |
... bash script
As you can see, each action can also execute simple bash scripts (omitted for the sake of brevity). Not very elegant, but it works!
Putting all together
I accomplished multiple goals:
- dismiss an old (and ugly) solution composed by Travis script and bash scripts
- learn a new tool
- keep the CV updated
- … and so keep the blog updated as well
Currently Github Actions is free for all repositories, both public and private (the latter with some limitations). I think it’s really worth start looking into it to have an integrated solution all in one place.