deploy.yml

name: Deploy to Railway

on:
  # Manual-only prod deploys. The automatic push-to-main trigger was removed
  # so merges (including migration PRs such as v141 / #3717) cannot ship to
  # Railway without an explicit workflow_dispatch. Operators must type
  # confirm=deploy-prod when running the workflow. No GitHub Environment
  # protection rules are used here (avoids paid settings).
  # The wait-for-ci job below still blocks any SHA without a green Backend
  # CI Status Gate. The deploy job uses concurrency group deploy-main with
  # cancel-in-progress: false so a prod migration mid-run is never cancelled.
  workflow_dispatch:
    inputs:
      confirm:
        description: 'Type deploy-prod to confirm a production Railway deploy (migrations included)'
        required: true
        type: string

permissions:
  checks: read
  contents: read

jobs:
  # Fail closed unless the operator typed the confirm phrase. A wrong/empty
  # value must not silently skip into a green workflow.
  confirm-gate:
    name: Confirm production deploy
    runs-on: ubuntu-latest
    timeout-minutes: 5
    steps:
      - name: Require confirm=deploy-prod
        env:
          CONFIRM: ${{ github.event.inputs.confirm }}
        run: |
          if [ "$CONFIRM" != "deploy-prod" ]; then
            echo "::error::Refusing deploy: confirm input must be exactly 'deploy-prod'."
            exit 1
          fi
          echo "Confirm input accepted; proceeding to CI wait."

  # Gate: manual deploys must wait for a green CI Status Gate check on the
  # selected ref before Railway deployment can start.
  wait-for-ci:
    name: Wait for CI
    needs: [confirm-gate]
    runs-on: ubuntu-latest
    timeout-minutes: 50
    steps:
      - name: Require backend CI green
        uses: lewagon/wait-on-check-action@369769072fe522a3a8a85c03c96af1e5242a1994  # v1.9.1
        with:
          ref: ${{ github.sha }}
          running-workflow-name: 'Wait for CI'
          repo-token: ${{ secrets.GITHUB_TOKEN }}
          # The deploy ships the backend monolith image, so it must wait on the
          # BACKEND gate specifically. Backend and frontend used to share the
          # name "CI Status Gate", so a frontend-only SHA (whose only run was
          # frontend-ci) satisfied this wait with the backend never verified.
          # The gates are now "Backend CI Status Gate" / "Frontend CI Status
          # Gate"; matching the backend name means a SHA with no backend run has
          # no matching check, so this step times out (checks-discovery-timeout)
          # instead of passing — deploy does not proceed. (#2537)
          # GitHub Checks API returns job names, not "workflow / job" format.
          # Backend CI can take longer than 15m to create its summary gate, so
          # keep discovery patient.
          check-name: 'Backend CI Status Gate'
          wait-interval: 30
          checks-discovery-timeout: 2700
          # Only a passing backend gate clears the deploy. `skipped` is NOT in
          # this list (the action default is `success,skipped`), so a skipped
          # gate cannot green-light a deploy.
          allowed-conclusions: success
          # If no "Backend CI Status Gate" check exists for the dispatched SHA
          # (e.g. a frontend-only SHA never ran backend-ci), fail rather than
          # treat "no matching check" as success — deploy must not proceed on a
          # ref the backend never verified (#2537).
          fail-on-no-checks: true

  deploy:
    name: Deploy RegEngine Service
    needs: [wait-for-ci]
    # Serialize prod deploys; never cancel mid-migration (Railway migrate).
    concurrency:
      group: deploy-main
      cancel-in-progress: false
    runs-on: ubuntu-latest
    timeout-minutes: 30
    steps:
      - name: Checkout code
        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1  # v6

      - name: Setup Node.js
        uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1  # v6
        with:
          node-version-file: .nvmrc

      - name: Install Railway CLI
        run: |
          npm install -g "@railway/cli@4.44.0"
          railway --version

      - name: Validate Railway deployment secrets
        id: validate_railway
        env:
          RAILWAY_API_TOKEN: ${{ secrets.RAILWAY_API_TOKEN }}
          RAILWAY_PROJECT_ID: ${{ secrets.RAILWAY_PROJECT_ID }}
          RAILWAY_ENVIRONMENT_ID: ${{ secrets.RAILWAY_ENVIRONMENT_ID }}
          RAILWAY_SERVICE_ID: ${{ secrets.RAILWAY_SERVICE_ID }}
        run: |
          missing=()
          [ -z "$RAILWAY_API_TOKEN" ] && missing+=("RAILWAY_API_TOKEN")
          [ -z "$RAILWAY_PROJECT_ID" ] && missing+=("RAILWAY_PROJECT_ID")
          [ -z "$RAILWAY_ENVIRONMENT_ID" ] && missing+=("RAILWAY_ENVIRONMENT_ID")
          [ -z "$RAILWAY_SERVICE_ID" ] && missing+=("RAILWAY_SERVICE_ID")

          if [ ${#missing[@]} -gt 0 ]; then
            echo "::error::Railway deployment secrets are missing: ${missing[*]}"
            exit 1
          fi

      # Stamp the commit into the build context. .dockerignore excludes .git,
      # so without this the image has no way to say which commit it is —
      # which is what made the class of failure below undetectable.
      - name: Stamp build identity
        run: |
          printf '%s\n' "${{ github.sha }}" > build-sha.txt
          echo "Stamped build-sha.txt with ${{ github.sha }}"

      - name: Deploy to Railway
        env:
          # Use the account/workspace token path. The previous project token
          # secret was unauthorized for this project/service.
          RAILWAY_API_TOKEN: ${{ secrets.RAILWAY_API_TOKEN }}
          RAILWAY_PROJECT_ID: ${{ secrets.RAILWAY_PROJECT_ID }}
          RAILWAY_ENVIRONMENT_ID: ${{ secrets.RAILWAY_ENVIRONMENT_ID }}
          RAILWAY_SERVICE_ID: ${{ secrets.RAILWAY_SERVICE_ID }}
        run: |
          railway up \
            --ci \
            --project "$RAILWAY_PROJECT_ID" \
            --environment "$RAILWAY_ENVIRONMENT_ID" \
            --service "$RAILWAY_SERVICE_ID"

      # `railway up --ci` returns once the upload is accepted, not once the new
      # image is serving. railway.toml sets restartPolicyType = "ON_FAILURE"
      # with a /readiness healthcheck, so a rollout that fails its healthcheck
      # reverts to the previous image *after* this workflow has already gone
      # green. Nothing downstream could notice: every probe asserts behaviour
      # an older image satisfies perfectly.
      #
      # This closes that hole by asserting identity rather than behaviour —
      # the deployed /health must report the commit we just shipped.
      - name: Verify the deployed commit is the one we shipped
        env:
          BACKEND_URL: https://regengine-production.up.railway.app
          EXPECTED_SHA: ${{ github.sha }}
        run: |
          set -uo pipefail

          # Rollout budget. railway.toml allows healthcheckTimeout=300 plus up
          # to 3 ON_FAILURE restarts, so a slow-but-succeeding rollout must not
          # be called a failure here.
          deadline=$(( SECONDS + 900 ))
          attempt=0
          last_sha=""
          last_code=""

          while [ "$SECONDS" -lt "$deadline" ]; do
            attempt=$(( attempt + 1 ))
            body=$(mktemp)
            last_code=$(curl -sS --max-time 20 -o "$body" -w '%{http_code}' "${BACKEND_URL}/health" || true)
            last_code="${last_code:-000}"

            if [ "$last_code" = "200" ]; then
              # `// empty` also covers a payload with no build key at all,
              # which is what an image predating build stamping returns.
              last_sha=$(jq -r '.build.commit_sha // empty' "$body" 2>/dev/null || true)
              if [ "$last_sha" = "$EXPECTED_SHA" ]; then
                rm -f "$body"
                echo "Deployed commit matches: ${EXPECTED_SHA}"
                exit 0
              fi
            fi
            rm -f "$body"

            echo "attempt ${attempt}: status=${last_code} deployed_sha=${last_sha:-(none)} expected=${EXPECTED_SHA}"
            sleep 15
          done

          echo "::error::Deploy did not converge. /health last reported status=${last_code} commit_sha=${last_sha:-(none)}, expected ${EXPECTED_SHA}."
          # Three distinct causes, three distinct messages. A single generic
          # failure string would send whoever reads this at 3am down the wrong
          # path, which is most of what makes an alert useless.
          if [ "$last_code" != "200" ]; then
            echo "::error::/health never returned 200 (last status ${last_code}). The service is not serving — check the Railway deploy logs for a crashed or unhealthy rollout."
          elif [ -z "$last_sha" ]; then
            echo "::error::/health answered but carried no commit_sha. Either the rollout reverted to an image predating build stamping, or build-sha.txt did not reach the image."
          else
            echo "::error::A different commit is serving — the rollout reverted (railway.toml restartPolicyType=ON_FAILURE) or another deploy raced this one."
          fi
          exit 1