The perfect Vercel + GitHub Actions deployment pipeline

June 10, 2021

As a Laravel developer, I haven't had much of a need for a serverless platform like Vercel. Recently though, I've been working on an API that provides VS Code compatible syntax highlighting and Vercel was the perfect choice for hosting it: fast, secure, and reliable.

Standard Deploying

You can deploy your Vercel app in a couple different ways. One is to deploy from your local machine using the Vercel CLI, by just calling vercel --prod. This makes your app live on the internet in less than a minute, which is kind of insane.

The other option is to hook up your GitHub repo to Vercel and let Vercel deploy every time you push to main.

Both of these options are extremely quick, but make me nervous. I didn't have any confidence that I wasn't deploying a broken build.

Perfect Deploying

In my mind, the perfect strategy would:

  • Run automatically, not be dependent on a human
  • Run my test suite, only deploying if it passes
  • Deploy a preview to Vercel
  • Run the tests again, against the Vercel preview
  • Deploy to production if the tests pass again.

This means that my tests would actually run against my API live on Vercel. It's wonderful to test everything, but it gives me so much more confidence to know that after everything is deployed, it's all still going to work.

It may seem silly, but there is a wide chasm between a local machine and a fully deployed API on Vercel. There is a lot that could go wrong!

The GitHub Actions Workflow

Since the code is all hosted on GitHub, GitHub Actions is the perfect way to implement this strategy. To get a GitHub Action up and running, you just need a .yml file in the ./github/workflows/ folder.

We'll call ours deploy.yml. We're going to walk through it step by step, but here is an overview of it first, with the steps collapsed:

./github/workflows/deploy.yml

name: Deploy Torchlight Engine
on: [ push, pull_request ]
jobs:
vercel:
runs-on: ubuntu-latest
steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}
Code highlighting powered by torchlight.dev (A service I created!)

The first part of the workflow defines a bit of meta for us:

  • the name of the workflow
  • when the action will run (push or pull request)
  • the job definitions

In our case we have one job, named vercel and it runs on the latest Ubuntu.

./github/workflows/deploy.yml

name: Deploy Torchlight Engine
on: [ push, pull_request ]
jobs:
vercel:
runs-on: ubuntu-latest

Building the World

The first few steps set up our environment and get us ready to be able to do our work. We checkout the repo, restore NPM cache if we can, install Node.js, and then run our own npm run bundle script to get our application ready to be deployed.

(For more about the --frozen-lockfile bit, check out my other post on Reliably Building Assets.)

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}

Running the Tests

Before we deploy anything we want to run the tests locally. If the tests don't pass locally they definitely aren't going to pass on Vercel, so we can save ourselves some time. Vercel also has limits on concurrent builds, so it doesn't make sense to clog up the pipeline with builds that we know we are going to abandon.

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}

Deploying to Staging

If the local tests pass, then we want to go ahead and deploy a staging site to Vercel. We use a pre-built Vercel GitHub Action to help us out here, so that all we need to do is pass in some secrets.

Check out the GitHub docs for information on how to configure your repo secrets.

This action outputs a preview-url where your app can be reached on Vercel, which is what we're going run our tests against in the next step.

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}

Testing Against Staging

Now we're going to test against the fresh deploy we just made to Vercel. The first thing we do is define a new env variable called VERCEL_URL. The value of this URL come from the previous step (deploy-vercel-staging) and is one of the outputs listed in the docs of the Vercel Action.

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}
Code highlighting powered by torchlight.dev (A service I created!)

It will change based on the name of your app, but it will inject an environment variable similar to the following:

VERCEL_URL=https://your-app-git-refs-headsmain-hammerstone.vercel.app

With that env variable set, we can read it in our test runner just like every other env var, and switch the endpoint against which we're testing:

tests/runner.js

let remote = process.env.VERCEL_URL;
 
if (remote) {
console.log(`Testing against remote url: ${remote}.`);
} else {
console.log('Testing locally only.');
}
 
// Run your tests locally or aginst the preview URL.

With our test runner modified, we can simply run the exact same command to test against staging: npm run test

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}

If you check the logs of your GitHub Action, you'll see something similar to this:

Run npm run test
 
> @ test /home/runner/work/torchlight-engine/torchlight-engine
> node ./tests/run.js
 
************************
* Running Tests *
************************
 
Testing against remote url: https://torchlight-XXXX-hammerstone.vercel.app.
 
681ms [PASS] /tests/fixtures/summary-detail-with-class.txt
695ms [PASS] /tests/fixtures/yaml-with-config.txt
734ms [PASS] /tests/fixtures/focus-line-ranges.txt
755ms [PASS] /tests/fixtures/blade.txt
717ms [PASS] /tests/fixtures/min-dark-has-git-diff-colors.txt
770ms [PASS] /tests/fixtures/summary-detail.txt
708ms [PASS] /tests/fixtures/relative-double-range.txt
750ms [PASS] /tests/fixtures/focus-lines.txt
740ms [PASS] /tests/fixtures/relative-range-positive.txt
760ms [PASS] /tests/fixtures/relative-range-negative.txt
698ms [PASS] /tests/fixtures/no-lang.txt
755ms [PASS] /tests/fixtures/relative-double-range-2.txt
814ms [PASS] /tests/fixtures/moonlight-ii.txt
846ms [PASS] /tests/fixtures/yaml-ranges-were-out-of-order.txt
777ms [PASS] /tests/fixtures/solarized-light-has-line-number-colors.txt
842ms [PASS] /tests/fixtures/adds-classes.txt
940ms [PASS] /tests/fixtures/cpp-comment-is-removed.txt
[..... Lots more tests]
Done.

We've now successfully run tests against the environment that we actually care about.

The last thing we have to do is deploy to production. Remember that the entire job will fail if one of the steps fail, so we don't have to check for that.

This is almost identical to the Deploy to Staging action, save for two things. The first is that we add an if condition into the step definition to only deploy to production if we're on the main branch. We'll run tests against staging on all branches, but we obviously only ever want to deploy to production for our main branch.

Make sure that you do not have the setting enabled in Vercel that auto-deploys your main branch. We will be manually deploying from this action!

The second thing we add is the vercel-args: '--prod', which is what tells Vercel to make this a production deployment.

steps:
- name: Checkout Code ...
uses: actions/checkout@v2
 
- name: Cache Dependencies ...
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14 ...
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build ...
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally ...
run: npm run test
 
- name: Deploy to Staging ...
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel ...
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production ...
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}

All Tests Pass

Now every time you push to main, you will see the following results if everything passes:

Go forth and merge with confidence!


Here is the entire, uncollapsed YAML file in case you want to copy paste it:

name: Deploy Torchlight Engine
on: [ push, pull_request ]
jobs:
vercel:
runs-on: ubuntu-latest
steps:
- name: Checkout Code
uses: actions/checkout@v2
 
- name: Cache Dependencies
uses: actions/cache@v1
with:
path: ~/.npm
key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
restore-keys: ${{ runner.os }}-node-
 
- name: Setup Node 14
uses: actions/setup-node@v2
with:
node-version: '14'
 
- name: Build
run: |
yarn install --frozen-lockfile
npm run bundle
 
- name: Run Tests Locally
run: npm run test
 
- name: Deploy to Staging
id: deploy-vercel-staging
uses: amondnet/vercel-action@v20
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
scope: ${{ secrets.VERCEL_ORG_ID }}
 
- name: Run Tests Against Vercel
env:
VERCEL_URL: ${{ steps.deploy-vercel-staging.outputs.preview-url }}
run: npm run test
 
- name: Deploy to Production
uses: amondnet/vercel-action@v20
id: deploy-vercel-production
if: github.event_name == 'push' && github.ref == 'refs/heads/main'
with:
vercel-token: ${{ secrets.VERCEL_TOKEN }}
vercel-org-id: ${{ secrets.VERCEL_ORG_ID }}
vercel-project-id: ${{ secrets.VERCEL_PROJECT_ID_TL_ENGINE }}
vercel-args: '--prod'
scope: ${{ secrets.VERCEL_ORG_ID }}
Me

Thanks for reading! My name is Aaron and I write, make videos , and generally try really hard .

If you ever have any questions or want to chat, I'm always on Twitter.

If you want to give me money (please do), you can buy my course on SQLite at HighPerformanceSQLite.com or my course on screencasting at Screencasting.com . On the off chance you're a sophomore at Texas A&M University, you can buy my accounting course at acct229.com .

You can find me on YouTube on my personal channel . If you love podcasts, I got you covered. You can listen to me on Mostly Technical .