DEV Community

Cover image for The ultimate (free) CI/CD for your open-source projects
Yohan Lasorsa for ITNEXT

Posted on

The ultimate (free) CI/CD for your open-source projects

I spend a good part of my free time working on open-source software (OSS) on GitHub, and I do it (mostly) for fun and learning new things. But one thing that I can tell you is no fun, is losing time hunting bugs that I cannot reproduce... Or fighting with tools supposed to help you, like your continuous integration (CI) system.

Did you ever wonder on which CI/CD solution you should invest your time? You've come to the right place!

Mandatory disclosure: I now work for Microsoft, but I started thinking about CI migration way before switching job and I was NOT forced to do this in any way. My experience with OSS projects goes back far before my current job, and I tried to put my honest feedback as an OSS maintainer in this article.

TL;DR

  • If you plan to maintain an open-source project in the long run, what you must do is automate
  • The worst way of losing time is when your automated builds are not working as intended: flakiness, unreliable builds, and cross-platforms issues can (will) drive you crazy
  • Pragmatism should be your main driver here, seek to get the job done for the least time invested

Based on this, I found that Azure Pipelines currently offers the best time investment/benefits ratio in my opinion.

That may probably change once GitHub Actions CI/CD solution becomes available for everyone (planned for November, 13th). But until then I will stick with a tried & tested solution that allows me spending more time with my loved ones than fighting with build issues 🍸

What will you learn here?

  • How to setup Azure Pipelines for an open-source project hosted on GitHub (Node.js in this case, but it works for any stack)
  • Why it can help you save some of your limited OSS working time

If you don't care for the back story, you can jump straight into action.

The full story

It has been a few years now that I have been maintaining multiple open-source projects on GitHub, and through the years I have come to this conclusion:

  • 🔥 Being a maintainer is hard (no kidding!)
  • ⏲ It takes a lot of your free time (that you might sometimes prefer spending with you family and friends)
  • 😭 Cutting corners to gain time in short-term almost always results in a lot of time lost in the long-term (like with any software project)

Jokes aside, the last point is easy to overlook and I fell into it too many times by not adding proper tests or CI, like:

"I'm doing this mostly for myself, but I'll share it in case it can be useful to others"

Oops, did not plan for that 😱

BIG MISTAKE. Once issues and PRs come, it's too late.
Reviewing, testing and deploying fixes and new features from your contributors becomes a black hole of time lost.

Free time in black hole

Black hole by Event Horizon Telescope Collaboration, CC BY 4.0 // Altered with my awesome Paint skills

After a few projects though, it's probable that you don't make this mistake anymore and that you put in place a minimal CI pipeline as I did 😎

Then it was fine. Somehow.

A few years ago, there was not so much choice if you wanted a free hosted CI solution for an open-source project. I was happy some solutions existed, given that it was, well... free.

That's how I ended up using most popular solutions such as Travis CI, AppVeyor and CircleCI. Each one meets a specific need, but not any of these are perfect, unfortunately.

Service OSS Tier
Travis CI Simplest and most common way to get started with a CI on GitHub. 5 concurrent jobs shared between ALL your repositories
Linux + Mac builds
AppVeyor For a long time, it was the only CI platform providing a free tier for Windows builds. 1 concurrent job
Windows + Linux builds.
CircleCI The most flexible solution of these 3 as you can use your own container, but it's also the most complex to setup 1 concurrent job
1000 builds minutes/month
Linux only builds

Red builds and issues

For most simple projects using any of these solutions might be just enough. But over time I ran into a bunch of problems, growing with the complexities of my projects.

The build matrix

First comes the problem of the environment: I develop on Mac OS, the CI runs on Linux... and issues appear on Windows. That's how I ended up using a combo of Travis CI + AppVeyor, like many projects.

Then there's the issue of being compatible with multiple platform versions. For example, a project supporting Node.js >= 8 would be wise to test on all LTS version along with the latest stable version: that would mean at this time 8.x, 10.x and 12.x.

Nearly all CI services support a build matrix definition for these use cases, so that's fine.

But when you start having many projects and each one have multiple builds, total build time starts to become an issue: I already had to wait for days (!) to be able to merge some PRs because many came at the same time on different projects (and all were sharing the same worker limit). When your pipeline is already quite long (like 1h for a PR build), it can get problematic really fast.

Uncertainty is not welcome

Finally, there's the most dreaded issue that took countless hours of my life for nothing: flaky builds 😱.

Fighting the build matrix

Sometimes, for no good reasons, one or many of your builds fail. And you (or your contributors) lose time investigating the issue, discovering that the CI service is the problem, not the code.

If it's only once in the while and you're suspicious, you just restart the build and it passes. But sometimes not, making you spending precious time trying to fix the CI because:

  • An unfortunate update (bug) was pushed (shit happens, but once it was there for WEEKS)
  • Build container images were tweaked (and your CI goes 💥)
  • "We are currently experiencing technical difficulties"
  • No reason... 😞

And over time this makes you wary of your own CI, which is not a good thing.

I know I may have been asking too much and most OSS projects are just fine using these solutions. And to be completely fair, given that all these services are provided for free it's hard to complain.

Meet Azure Pipelines

Some time ago I discovered that JHipster had migrated all its CI to Azure Pipelines. I knew this project for a long time and its CI complexity is nothing to laugh at, with a lot of combination paths and huge build times. So this got me interested and I tried to migrate some projects of my own to test it out.

Well, I was surprised to see that Azure Pipelines gives you quite a lot for free compared to other solutions:

  • It can target any OS (Windows/Mac/Linux)
  • It's fast (10 parallels workers per organization for OSS tier )
  • It seems more reliable (from my own experience and discussions with JHipster maintainers)
  • It's quite easy to set up and get working (a bit more complex than Travis though, but with way more flexibility)

I started by migrating the tests of a simple Node.js project of mine. The existing Travis/AppVeyor combo was replaced with a single Azure Pipeline to test on Windows/Mac/Linux, as done in this PR. I had to disable git autocrlf option for Prettier check to be happy on Windows, but besides that the migration was trivial.

After this first success I moved on to migrate the much more complex and demanding CI of generator-ngx-rocket, and here is the result.

Before After
Service Travis + AppVeyor + Circle CI with custom Docker image Azure Pipelines
Combinations 40 builds on Travis (Linux * 2 Node versions)
20 builds on AppVeyor (Windows)
1 build on CircleCI (Android)
83 builds (Linux/Windows * 2 Node versions + Mac/Android)
Build time ~1h for a PR (~50min Travis, ~1h AppVeyor, ~5min CircleCI) ~1h for a PR (~30min if limited to the previous Travis scenario)

The migration was not trivial this time as I had to use a template to generate the combinations matrix, but the result is great:

  • Only one config to manage (and no need to build an extra Docker image for Android)
  • More combinations tested for the same time
  • Some bonus like publication of Android built APKs directly as a build artifact

After this experience, I can say that I'm all for Azure Pipeline now and I will continue to migrate my other repositories 👍

Hey wait, but what about the new GitHub Actions for CI/CD?!

I give you that it looks AWESOME, especially the OSS tier 😍
Once it becomes available for everyone on November 13th, it will probably become the best solution out there (and this article will be updated in that case 😉).

Two things to still consider Azure Pipelines though:

  • If you are not invited as one of the beta testers, you won't be able to use GitHub Actions for now. But in the meantime, you probably want a working CI solution!
  • GitHub Actions for CI/CD is based on parts of the Azure Pipelines stack, so migration should be a breeze if you are already using the latter

Let's do it!

At this point, you may want to try it out for yourself. We will go through the steps to put in place an Azure Pipeline CI for your Node.js project (don't go away if you use another stack, it's not much different and I will give some pointers 😉).

You can see an example result on this repository.

Requirements:

  • An Azure DevOps account (you can use your GitHub account to log in)
  • A GitHub repository with some tests you want to automate

1. Prepare your repository

First, you have to prepare your test scripts for the CI.

For a Node.js project, it's usually done by adding a test script in your package.json file:

{
  "scripts": {
    "test": "jest"
  }
}
Enter fullscreen mode Exit fullscreen mode

In this example, Jest will be used to run the unit tests.

You can test your script by running npm test, to make sure everything works fine.

2. Create the Azure DevOps project

In your Azure DevOps account, create a new project and give it a name:

create project

3. Setup the pipeline

After that, go to the Pipelines section on the left:

select pipelines

Then click New pipeline to start the setup process:

new pipeline

Choose GitHub (YAML), then follow the process to link your account with Azure DevOps:

choose github

Select your repository, then authorize the app on GitHub.

After that, choose the first Node.js option (or any other relevant option for your tech stack):

choose nodejs

For other stacks and languages like C#, Java, Go, Python and so on you should take a look at the appropriate page in the ecosystem section of the docs. You will get specific hints at how you can set up your build with your favorite tech.

You can find a lot of examples YAML files in this repository.

4. Customize the YAML

You will then be presented with this YAML file in the editor:

# Node.js
# Build a general Node.js project with npm.
# Add steps that analyze code, save build artifacts, deploy, and more:
# https://docs.microsoft.com/azure/devops/pipelines/languages/javascript

trigger:
  - master

pool:
  vmImage: 'ubuntu-latest'

steps:
  - task: NodeTool@0
    inputs:
      versionSpec: '10.x'
    displayName: 'Install Node.js'

  - script: |
      npm install
      npm run build
    displayName: 'npm install and build'
Enter fullscreen mode Exit fullscreen mode

For now, you only need to change the last part to run the command npm test instead of npm run build:

- script: |
    npm install
    npm test
  displayName: 'npm install and test'
Enter fullscreen mode Exit fullscreen mode

Click Save and run, then choose whether to commit directly on master or create a new branch with a PR.

I suggest you create a new branch with a PR at this stage so that you can experiment with the build configuration before merging it in your repository.

After that click Save and run again, then check that everything runs fine.

5. Adding a build matrix

Now you already have a working CI, but one of the benefits of using Azure Pipelines is that you can easily add cross-platform builds so let's do that.

Choose Edit pipeline in the menu to go back to the YAML editor:

choose edit pipeline

First add this matrix at the top of your file:

strategy:
  matrix:
    linux-node-10:
      imageName: 'ubuntu-latest'
      nodeVersion: '10.x'
    linux-node-latest:
      imageName: 'ubuntu-latest'
      nodeVersion: '>=12.x'
    mac-node-10:
      imageName: 'macos-latest'
      nodeVersion: '10.x'
    mac-node-latest:
      imageName: 'macos-latest'
      nodeVersion: '>=12.x'
    windows-node-10:
      imageName: 'windows-latest'
      nodeVersion: '10.x'
    windows-node-latest:
      imageName: 'windows-latest'
      nodeVersion: '>=12.x'
Enter fullscreen mode Exit fullscreen mode

This will allow your tests to run on Windows, Linux and Mac OS using 2 different Node.js versions each time, for a total of 6 builds.

Then update the pool section to use the image name of the current build:

pool:
  vmImage: $(imageName)
Enter fullscreen mode Exit fullscreen mode

Finally update the node task to do the same for the Node.js version:

- task: NodeTool@0
  inputs:
    versionSpec: $(nodeVersion)
  displayName: 'Install Node.js'
Enter fullscreen mode Exit fullscreen mode

Click Save, then Run to test the updated pipeline and you're done.

If you created a branch with a PR when saving the pipeline configuration the first time, you also need to go on GitHub and merge this PR first to activate the pipeline for new commits and PRs.

Don't forget to take a look at the full documentation, to see all the tasks and features available.

6. Add a status badge (optional)

build succeeded badge

I highly recommend that you add a status badge on your GitHub repository so that people get a positive signal on your project:

  • It's maintained and you care for its quality
  • Contributors can push PRs with confidence

To do so, go back to your pipelines builds and click Status badge on the menu:

click status badge

Then copy the markdown code using the button, and paste it at the top of your readme.md file at the root of your repository:

copy markdown

Once it's committed to the master branch it will become visible on the GitHub documentation of your project. The badge status will be updated automatically according to the last commit pushed on master.

To complete this status badge with some others, take a look at https://shields.io.
You can even build your own: badges are awesome

7. Enjoy your evenings

That's it!

Once a new commit or PR will be pushed to your repository, the CI will take care of it and show the result directly on GitHub:

pull request status checks

Now you can relax and enjoy your free time again 🍻.

For additional peace of mind, I suggest you enable required status check with your new CI so only PRs passing the selected checks are allowed to be merged.

Don't forget to share your feedback and experience in comments!

Going further

Of course, there are many more things that you can do to simplify your life as a maintainer and I will share them in later posts. But I think CI is one of the first things you should put in place in any project.

There's also much more that you can do with Azure Pipelines (did I mention CD somewhere?) but let's keep that for later 😉.


Follow me on Twitter, I would be happy to discuss and take your suggestions!

Top comments (20)

Collapse
 
jeastham1993 profile image
James Eastham

Love this! I've recently found Azure pipelines and migrated all our company repo's over to Azure Dev Ops repos to make the most of it.

Going to connect some of my personal GitHub projects to it as well, it really is a fantastic tool though.

Collapse
 
marceliwac profile image
Marceli-Wac

That's all great and the article is very comprehensive! However, with a lot of experience in CI/CD using Gitlab's toolchain I feel like omitting Gitlab in this piece is quite ominous. Especially given that it's not only free but also supports free hosting, operations support etc.

Collapse
 
sinedied profile image
Yohan Lasorsa

You are right as Gitlab is a good alternative to GitHub, but please see my previous answer regarding Gitlab hosted CI: as good as it is, its hosted runners are limited to Linux so it's not an option for cross-platform OSS projects. Though, you can still use Gitlab hosting with Azure Pipeline, maybe I should mention it in the article? 😉
I admit I only focused on solutions for GitHub since most OSS projects are hosted there. When I will update the article regarding GitHub actions it will be a good time to put Gitlab in the comparison, thanks for the feedback.

Collapse
 
marceliwac profile image
Marceli-Wac

I've read your comment before and agree with the limitations, just thought it still deserves a mention :D

Collapse
 
codercatdev profile image
Alex Patterson

So I use Azure pipelines at work and love how easy they are to setup using the interface. However for my personal stuff I still love using GCloud. Maybe it is just because I like to stick within the GCloud/Firebase system as a whole.

Collapse
 
sinedied profile image
Yohan Lasorsa

I never quite looked at Cloud Build because from seeing the docs it seems limited to 120min a day for its free tier, which would not fit my needs. Also the docs only mention Linux and Docker, so no MacOS/Windows support. I might try it though in place of Travis for simple projects for comparison 😉

Collapse
 
anoff profile image
Andreas Offenhaeuser

Given the alternatives you provided I agree with pipelines being the best.
However I think gitlab or drone have way better (simpler) UX for running simple jobs that can be done in docker.

One thing I really dislike about pipelines is the fact that it does not have real scheduled jobs. Yes you can configure them in yaml but it's not guaranteed that they run (once per login)

Collapse
 
sinedied profile image
Yohan Lasorsa

I'm not sure I understand what issues you had with scheduled builds, can you give me more details? You can use the always: true setting to force scheduled builds to run, otherwise if any build (manual, commit or PR) is started within the timeframe of the schedule, the scheduled build will be ignored.

Collapse
 
anoff profile image
Andreas Offenhaeuser

I tried to benchmark runtime with a 2 hourly job that didn't run over night.

twitter.com/acanthamoeba/status/11...

There are so many things in pipelines that work non intuitive that I actually prefer the other services I mentioned.

Thread Thread
 
sinedied profile image
Yohan Lasorsa

Wow I was not aware of that, it's really a weird behavior 😕
Thanks for sharing this, I will push it to the team in charge of it and see what I can do.
It's easy to find a workaround (like with a scheduled Azure function call or even with a simple curl scheduled by Travis 😄) but it kinda defeats the point of having it simply exposed in the YAML... I hope GitHub actions will be better in this regard.

Collapse
 
punkdata profile image
Angel Rivera

Full disclosure I'm a dev advocate for CircleCI. I'm interested in learning more about why you view it as "more complex" than the other options?
Also CircleCI now supports Linux, Docker, macOS/IOS & Windows based executors which provides cross platform support. CircleCI is also very supportive of open source projects and provide very generous free CI/CD resources. Check them out.
circleci.com/open-source/

Collapse
 
sinedied profile image
Yohan Lasorsa

IIRC when we put up our CircleCI setup it took us almost 2 whole days to have it working completely (including the setup of the custom docker image to build for Android though). I only helped with few parts, but compared to Travis/AppVeyor/Azure Pipelines the syntax feels weird in some points, like the alias stuff (looks like C++ 😄). I also remember we had some difficulties when trying to get the cache and workflow working properly.
But to be honest, one excellent thing though is that we never had to update the config once it was working (only the Docker image) which is a very good point!
As for OSS projects, unless I'm wrong only Linux workers are provided by default. You have to ask for a limited access to MacOS workers and I could not find anything about Windows, which is not the best if you quickly want to spin up some tests to see if it fit your needs IMHO.

Collapse
 
paulasantamaria profile image
Paula Santamaría

Great post! Can't wait to read about CD in azure pipelines 👍

Collapse
 
mokenyon profile image
Morgan Kenyon

Great article! I need to implement setup a pipeline. Thanks for putting this together!!!

Collapse
 
chrisachard profile image
Chris Achard

Interesting! Do you have any details about how this might change when Github CI comes out? I feel like that may really disrupt the current status quo.

Collapse
 
sinedied profile image
Yohan Lasorsa

I got my GitHub Actions invite yesterday so I will test it out thoroughly until its GA, and update the article if needed. What I can already say is that it's similar to Azure Pipelines feature wise, but being completely integrated with GitHub is a big advantage.

Collapse
 
chrisachard profile image
Chris Achard

Ah great. Thanks!

Collapse
 
gdledsan profile image
Edmundo Sanchez

I am looking at GCP Cloud build with 2500 re minutes per month, I would love your opinions on this

Collapse
 
dariusx profile image
Darius

I recently learnt that Gitlab has introduced Pipeline features.

Collapse
 
sinedied profile image
Yohan Lasorsa

I heard good things about it, though I found it mostly interesting for on-premise solutions as Gitlab hosted CI runners are limited to Linux. You need to bring your own MacOS/Windows agents if you need it, so it's not ideal for cross-platform OSS projects.