GitHub Actions: Automatically rerun failed jobs

rerun-workflow-test.yml

name: Rerun workflow test

on:
  workflow_dispatch:
    inputs:
      maxAttempts:
        required: true
        type: number
        default: 3
      rerunFailedOnly:
        required: true
        type: boolean
        default: 'false'

env:
  # You can hardcode this instead of passing it in from inputs
  maxAttempts: ${{ github.event.inputs.maxAttempts }}
  rerunFailedOnly: ${{ github.event.inputs.rerunFailedOnly }}

jobs:
  attempt_job:
    name: "Attempt ${{ github.run_attempt }}"
    runs-on: ubuntu-latest
    outputs:
      nextAttempt: ${{ steps.set_output_variables.outputs.nextAttempt }}
      maxAttempts: ${{ steps.set_output_variables.outputs.maxAttempts }}
    steps:
      - name: Set output variables
        id: set_output_variables
        shell: pwsh
        run: |
          $attempt = [int]$env:GITHUB_RUN_ATTEMPT
          $maxAttempts = [int]$env:maxAttempts
          $nextAttempt = $attempt + 1
          echo "::set-output name=nextAttempt::$nextAttempt"
          echo "::set-output name=maxAttempts::$maxAttempts"
        # Fails every attempt except the last
      - name: Run test script
        shell: pwsh
        run: |
          $attempt = [int]$env:GITHUB_RUN_ATTEMPT
          $maxAttempts = [int]$env:maxAttempts

          if($attempt -eq $maxAttempts) {
            Write-Host "Succeeded on attempt $attempt"
          } else {
            Write-Error "Failed on attempt $attempt"
            exit 1
          }

  rerun_workflow_job:
    name: Rerun ${{ github.run_id }}
    runs-on: ubuntu-latest
    needs: attempt_job
    if: failure() && ( needs.attempt_job.outputs.nextAttempt <= needs.attempt_job.outputs.maxAttempts )
    env:
      attempt: ${{ needs.attempt_job.outputs.nextAttempt }}
    steps:
      - name: "Trigger Attempt ${{ needs.attempt_job.outputs.nextAttempt }}"
        shell: pwsh
        env:
          # GITHUB_TOKEN with actions: write will NOT work
          # PAT Token must have workflow scope
          GH_CLI_TOKEN: ${{ secrets.PAT_TOKEN }}
        run: |
          echo "Running gh auth login..."
          $env:GH_CLI_TOKEN | gh auth login --with-token
          if(!$?) { exit 1 }

          echo "Triggering rerun attempt $env:nextAttempt"

          echo "
          gh workflow run rerun-workflow.yml
            --raw-field runId=$env:GITHUB_RUN_ID
            --raw-field rerunFailedOnly=$env:rerunFailedOnly
            --repo $env:GITHUB_REPOSITORY
          "

          gh workflow run rerun-workflow.yml `
            --raw-field runId=$env:GITHUB_RUN_ID `
            --raw-field rerunFailedOnly=$env:rerunFailedOnly `
            --repo $env:GITHUB_REPOSITORY
          if(!$?) { exit 1 }

rerun-workflow.yml

name: Rerun workflow

on:
  workflow_dispatch:
    inputs:
      runId:
        required: true
        type: string
      rerunFailedOnly:
        required: false
        type: boolean
        default: 'false'

env:
  runId: ${{ github.event.inputs.runId }}
  rerunFailedOnly: ${{ github.event.inputs.rerunFailedOnly }}

jobs:
  retry_workflow_job:
    name: Rerun ${{ github.event.inputs.runId }}
    runs-on: ubuntu-latest
    steps:
      # Upgrade gh cli to use the --failed parameter (added in version 2.6.0 (2022-03-15))
      # This takes 1m 30s
      - name: Upgrade gh cli to latest stable
        if: github.event.inputs.rerunFailedOnly == 'true'
        shell: pwsh
        run: |
          gh --version
          brew update
          brew doctor
          brew install gh
          gh --version
      - name: "Rerun ${{ github.event.inputs.runId }}"
        shell: pwsh
        env:
          # GITHUB_TOKEN with actions: write will NOT work
          # PAT Token must have workflow scope
          GH_CLI_TOKEN: ${{ secrets.PAT_TOKEN }}
        run: |
          echo "Running gh auth login..."
          $env:GH_CLI_TOKEN | gh auth login --with-token
          if(!$?) { exit 1 }

          echo "Retrying run..."
          $rerunFailedOnly = [Convert]::ToBoolean($env:rerunFailedOnly)
          if($rerunFailedOnly) {
            echo "gh run rerun $env:runId --failed --repo $env:GITHUB_REPOSITORY"
            gh run rerun $env:runId --failed --repo $env:GITHUB_REPOSITORY
          } else {
            echo "gh run rerun $env:runId --repo $env:GITHUB_REPOSITORY"
            gh run rerun $env:runId --repo $env:GITHUB_REPOSITORY
          }
          if(!$?) { exit 1 }