I needed to deploy a node web application into Azure; as always, I like to keep costs low, so I used the free GitHub action minutes I have as a GitHub pro member. And I was in for a shock, with the first build taking 41 minutes and 30 seconds when using the default template for deploying a Web Application to Azure.

Task Time
Setup Job 2 seconds
Run actions/checkout@v3 2 seconds
Set up Node.js 2 seconds
Yarn install, build and test 1 minute 28 seconds
Upload artifact for deployment job 39 minutes 38 seconds
Post set up Node.js 2 seconds
Post Run actions/checkout@v3 0 seconds
Complete job 0 seconds


There is no way I am going to wait 40 minutes for a build, so it was time to dive into what the default workflow was doing.

Improving the upload job speed

GitHub Actions was uploading each file one by one, all 60,000 of them. The cynical side of me thought it was one way for GitHub to make money as you pay per minute.

I could get around that by zipping the code and making the following changes to the pipeline: 

- name: Zip artifact for deployment
    run: zip release.zip ./* -r

- name: Upload artifact for deployment job
    uses: actions/upload-artifact@v3
    with:
    name: node-app
    path: release.zip


With this change, I now zip all the files together and then tell the upload artifact action to only upload the one zipped file. I also needed to update the release pipeline to unzip the file later. So what were the results:

Task Time
Setup Job 2 seconds
Run actions/checkout@v3 5 seconds
Set up Node.js 1 seconds
Yarn install, build and test 1 minutes 33 seconds
Zip artifact 31 seconds
Upload artifact for deployment job 1 minutes 32 seconds
Post set up Node.js 12 seconds
Post Run actions/checkout@v3 0 seconds
Complete job 0 seconds


Total time: 4 minutes and 2 seconds, a 90.28% decrease.

Reducing the size of the package

Now that I am only uploading one file, the next step is to ensure I am only uploading the files I need. I created a new task which will delete unnecessary files: 

 - name: Remove files
    run: |
    rm -r .husky
    rm -r assets
    rm -r src
    rm -r tests
    rm jest.config.js
    rm readme.md


That leaves me with the built code, node modules and the package.json file deployed into Azure. The time savings are below:

Task Time
Setup Job 1 seconds
Run actions/checkout@v3 2 seconds
Set up Node.js 6 seconds
Yarn install, build and test 50 seconds
Delete files 0 seconds
Zip artifact 11 seconds
Upload artifact for deployment job 30 seconds
Post set up Node.js 0 seconds
Post Run actions/checkout@v3 0 seconds
Complete job 0 seconds


Total time: 1 minute 48 seconds, a 55.37% decrease.

Bonus tidy up

As I was in the area, I also checked to see what the build package looked like after it was built. It contained the source code and the tests. I made a tweak to the tsconfig.json and removed even more files.

There was no drastic reduction in build time; however, it felt nice to know the build was clean.

Conculsion

In this extreme example, I shaved around 40 minutes off the build process, mainly because the default GitHub action template isn’t performant. There is some advice I would give to those reading which is:

Check the size of your package today

An efficient build process means that you are waiting around less. A shorter feedback loop will help you build and test code faster. Think back to 2022; if each build you did took 30 seconds less, how many hours would you have saved?