I need to automate deploys on Azure Devops for a project I’m working. I want it to work as cheaply as possible so serverless or PaaS was desirable. Read on to see the azure-pipelines.yaml file I used to make this work.

I’ll break down each step of the build and publish process with some comments.

The trickiest bits were deploying to each service.

You can skip to the end 👇 and find the entire file for easy copy pasting at the bottom.

Code Structure

I keep the code for my project in a mono repo on guthub. The structure looks like this

server
--src
--dist (this is build output)
client
--src
--build (this is build output)

Setting up Infrastructure

I used mongoDB Atlas for my datastore but I’ll leave that up to you. Essentially your server needs to be able to talk to whatever datastore you chose.

You will need a couple of things on Azure. It’s best if you set up a new dedicated resource group for this project.

Add a Linux based App Service. I just used an F1 (free) size. call this ‘myAppServer’ or similar. This is where graphQL will live. You might need to whitelist the “Outgoing IP address” on mongo or whatever your database is on.
Add a storage account. Scroll down to “Static Website” in the menu. Enable it. Add “index.html” as the index document name. This passes server routing requests to our single page app.

Setting up Azure DevOps pipeline

Go to Azure Devops / pippelines and add a new pipeline. Select your yaml supported repository and set up a node project or whatever. It will probably fail on the first run but that’s all good. It will ask you to link it to Azure Portal. This is super handy and means we don’t have to authenticate to run the build pipeline.

Azure will add a file azure-pipelines.yaml to your repo root. This is what we want to edit. The pipeline will always run the latest yaml definition.

Build outline

Setup some variables
Build the client and server
Test the code
Publish results
Copy required files to archive
Publish the archive
🚢 the client to our static site
🚀 the server to our App Service

I won’t go over each of the variables in the first step. Hit me up on twitter if it’s not clear which is which.

Building client and server

Here we just navigate to the relevant folders and build the code. I run the tests here too but you could do that in a different script. I’m just being lazy I guess.

I do set some environment variables here.

CI=true changes the way tests are run and published. It also enforces linting by failing the build on warnings.

REACT_APP_GRAPHQL_URL=‘myurl’ is a setting for the client application so it hits the correct production server. Otherwise the client you download would try to hit localhost. This is specific to my application though you probably need something similar. Doing this in ENV variables helps enforce the 12 factor application principals.

- script: |
     cd server
     yarn install
     npm run build --if-present
     npm run test --if-present
     cd ../client
     yarn install
     npm run build --if-present
     npm run test --if-present
   displayName: 'npm install, build and test'
   env:
     CI: true
     REACT_APP_GRAPHQL_URL: '$(serverUrl)'

Publish test results

This is pretty standard. Just list the specific paths to any test results. There is limited support for formats in azure pipelines so you might have to add a formatter to your test framework. I chose junit format here because jest has a formatter for it.

- task: PublishTestResults@2
  condition: succeededOrFailed()
  inputs:
    testRunner: JUnit
    testResultsFiles: '$(System.DefaultWorkingDirectory)/server/junit.xml'
  displayName: 'Publish test results'

Copying the server files to folder for archiving

The server is a standard Apollo graphQL server. It is not a static site so you have to copy over the source, the package.json and the lock file (I use yarn for this project).

We will install all the required packages when deploying the site later.

I just used the standard azure file locations for this build. I’m going to skip the client build because it’s the same with different paths.

 - task: CopyFiles@2
    inputs:
      SourceFolder: '$(System.DefaultWorkingDirectory)'
      Contents: |
        $(System.DefaultWorkingDirectory)/server/dist/**/*
        $(System.DefaultWorkingDirectory)/server/package.json
        $(System.DefaultWorkingDirectory)/server/yarn.lock
      TargetFolder: '$(Build.ArtifactStagingDirectory)/output'
    displayName: 'Copy SERVER files to artifact directory'

Archiving and publishing the artefacts

I won’t paste the code for this here - it’s all standard from any Azure sample. We package the code and publish so we can download later in case something is broken and we need to test.

By publishing now we could also utilize the Azure DevOps “Releases” product in the future which is triggered when new artefacts are published.

Deploying the client build

Azure provides a special “Copy to Azure” task but the does not work on Linux build agents. Instead we use the azure CLI to do this for us.

Use the bash client. We’re on Linux and I this will be there for sure.

“$web” is the default storage container for static sites on Azure blobs. The $ is a special character in bash so we have to escape it with ”\”. The Azure variables ”$(XXX)” are replaced before running on bash so won’t be a problem.

That’s it for the client. Static sites are awesome!

- task: AzureCLI@2
  inputs:
    azureSubscription: '$(azureSubscription)'
    scriptType: 'bash'
    scriptLocation: 'inlineScript'
    inlineScript: 'az storage blob upload-batch -d "\$web" --account-name "$(clientBlobAccountName)" -s "$(System.DefaultWorkingDirectory)/unzip/$(Build.BuildId)/client/build"'

Deploying the server build

The deploy of the server to app service is a bit more straightforward because the Azure task works on Linux agents.

Make sure the appType matches the type you added to Azure earlier. You must set the runtime stack to be the same as the runtime specified in your package.json (if you have done that)

We have to install any dependencies before running the container so we add a script to go into the required directory and yarn install.

Make sure you have copied over the lock file or yarn/npm will get different versions of your dependencies than you tested with!

After we install deps we simply run our application using node.

- task: AzureRmWebAppDeployment@4
  inputs:
    ConnectionType: 'AzureRM'
    azureSubscription: '$(azureSubscription)'
    appType: 'webAppLinux'
    WebAppName: '$(serverWebAppName)'
    packageForLinux: '$(System.ArtifactsDirectory)/drop/$(Build.BuildId).zip'
    RuntimeStack: 'NODE|10.16'
    StartupCommand: 'cd server && node dist/index.js'
    ScriptType: 'Inline Script'
    InlineScript: |
      cd server
      yarn install

Full azure-pipelines.yaml

# Deploy a graphql server and a static react app to azure serverless environment

trigger:
  - master

variables:
  # Subscription needs a storage account and a website
  azureSubscription: 'YOUR_SUB'

  # Needs to be a linux app service and there are a bunch of env variables that need to be set on it
  serverWebAppName: 'YOUR_SERVER_NAME'

  # This needs to have a static website setup with the default container ($web)
  clientBlobAccountName: 'YOUR_STORAGE_NAME'

  # This is provided to the client app so it knows how to hit the right server
  serverUrl: 'YOUR_SERVER_URL'

  # We build on linux to remove any dependancies on windows stuff / can move to GCP or AWS
  vmImageName: 'ubuntu-latest'

stages:
  - stage: Build
    displayName: Build stage
    jobs:
      - job: Build
        displayName: Build
        pool:
          vmImage: $(vmImageName)

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

          - script: |
              cd server
              yarn install
              npm run build --if-present
              npm run test --if-present
              cd ../client
              yarn install
              npm run build --if-present
              npm run test --if-present
            displayName: 'npm install, build and test'
            env:
              CI: true
              REACT_APP_GRAPHQL_URL: '$(serverUrl)'

          - task: PublishTestResults@2
            condition: succeededOrFailed()
            inputs:
              testRunner: JUnit
              testResultsFiles: '$(System.DefaultWorkingDirectory)/server/junit.xml'
            displayName: 'Publish test results'

          - task: CopyFiles@2
            inputs:
              SourceFolder: '$(System.DefaultWorkingDirectory)'
              Contents: |
                $(System.DefaultWorkingDirectory)/server/dist/**/*.js
                $(System.DefaultWorkingDirectory)/server/package.json
                $(System.DefaultWorkingDirectory)/server/yarn.lock
              TargetFolder: '$(Build.ArtifactStagingDirectory)/output'
            displayName: 'Copy SERVER files to artifact directory'

          - task: CopyFiles@2
            inputs:
              SourceFolder: '$(System.DefaultWorkingDirectory)'
              Contents: |
                $(System.DefaultWorkingDirectory)/client/build/**/*
              TargetFolder: '$(Build.ArtifactStagingDirectory)/output'
            displayName: 'Copy CLIENT files to artifact directory'

          - task: ArchiveFiles@2
            displayName: 'Archive files'
            inputs:
              rootFolderOrFile: '$(Build.ArtifactStagingDirectory)/output'
              includeRootFolder: false
              archiveType: zip
              archiveFile: $(Build.ArtifactStagingDirectory)/$(Build.BuildId).zip
              replaceExistingArchive: true

          - task: PublishPipelineArtifact@0
            displayName: 'Publish server pipeline artifacts'
            inputs:
              targetPath: $(Build.ArtifactStagingDirectory)/$(Build.BuildId).zip

  - stage: Deploy
    displayName: Deploy stage
    dependsOn: Build
    condition: succeeded()
    jobs:
      - deployment: Deploy
        displayName: Deploy Production
        environment: 'production'
        pool:
          vmImage: $(vmImageName)
        strategy:
          runOnce:
            deploy:
              steps:
                - task: DownloadPipelineArtifact@1
                  displayName: 'Download Pipeline Artifacts'
                  inputs:
                    buildType: 'current'
                - task: ExtractFiles@1
                  inputs:
                    archiveFilePatterns: '$(System.ArtifactsDirectory)/drop/$(Build.BuildId).zip'
                    destinationFolder: '$(System.DefaultWorkingDirectory)/unzip/$(Build.BuildId)'
                    cleanDestinationFolder: true

                - task: AzureCLI@2
                  inputs:
                    azureSubscription: '$(azureSubscription)'
                    scriptType: 'bash'
                    scriptLocation: 'inlineScript'
                    inlineScript: 'az storage blob upload-batch -d "\$web" --account-name "$(clientBlobAccountName)" -s "$(System.DefaultWorkingDirectory)/unzip/$(Build.BuildId)/client/build"'
                - task: AzureRmWebAppDeployment@4
                  inputs:
                    ConnectionType: 'AzureRM'
                    azureSubscription: '$(azureSubscription)'
                    appType: 'webAppLinux'
                    WebAppName: '$(serverWebAppName)'
                    packageForLinux: '$(System.ArtifactsDirectory)/drop/$(Build.BuildId).zip'
                    RuntimeStack: 'NODE|10.16'
                    StartupCommand: 'cd server && node dist/index.js'
                    ScriptType: 'Inline Script'
                    InlineScript: |
                      cd server
                      npm install

Hope that helps! Let me know if you have any trouble.

How to configure a pipeline on Azure Devops to build and deploy serverless graphql and react client