check-deployed-app.yml

# Loads the deployed app in a real browser and asks whether it draws a trail.
#
# Tier 2 of #431, tracked in #467. `check-deployment.yml` (tier 1) proves the
# bucket will answer a browser; only this proves the APP works with it.
#
#   check-deployment.yml    daily, no downloads   can a browser REACH the data
#   this                    daily, one page load  does the APP draw a trail
#   smoke-published.yml     weekly, ~18 MB        is what is published CORRECT
#
# WHY IT IS NOT REDUNDANT WITH TIER 1. Tier 1 cannot see a build that shipped
# without its data URL, a service worker serving a stale shell, a bundle that
# throws before its first fetch, or a CORS rule that passes a bare GET and
# fails the app's real request. Every one of those is "the map is broken" and
# none is visible to anything that is not a browser.
#
# AND WHY THE CHECK IS NOT ONLY THE TWO ASSERTIONS #431 ASKED FOR. Measured
# while building it: with the data source made unreachable, "no *No trail
# line* flag" and "no data-error notice" BOTH still passed, because the app
# raises them only once a download has failed and nothing had got far enough
# to fail. So the script also asserts the page really received the data, and
# that nothing it asked the bucket for was refused - the second is the #427
# shape seen from the app's side.
#
# IT HOLDS NO CREDENTIALS. Same posture as its two siblings: public HTTPS
# reads of the deployed site and the public bucket, nothing that could change
# what a hiker gets.
#
# IT DOES NOT FAIL THE RUN. GitHub emails on every scheduled failure, so a
# real outage would send one a day until it was filtered. The tracking issue
# is the signal - opened green->red, closed red->green.

name: Check deployed app

on:
  schedule:
    # 09:30 UTC daily - after tier 1 at 09:15, so if the bucket is refusing
    # browsers that job says so first and this one is not a second alarm for
    # the same fault.
    - cron: "30 9 * * *"
  workflow_dispatch:
    inputs:
      url:
        description: "App URL to load. Defaults to the GitHub Pages deployment."
        type: string
        required: false

permissions:
  contents: read
  issues: write

concurrency:
  group: check-deployed-app
  cancel-in-progress: false

env:
  ISSUE_TITLE: "Deployed app does not draw a trail"
  ISSUE_LABEL: deployment-health
  # The origin the app is actually served from since #733. Checking the old
  # project-site URL instead would still pass - it 301s here - while proving
  # nothing about the host a hiker is on.
  DEFAULT_URL: https://ourhike.org/app/

jobs:
  check:
    runs-on: ubuntu-latest
    timeout-minutes: 20
    defaults:
      run:
        working-directory: client

    steps:
      - name: Is there a deployment to check
        id: can
        working-directory: .
        env:
          OVERRIDE: ${{ inputs.url }}
          CONFIGURED: ${{ vars.DATA_BASE_URL }}
        run: |
          BASE="${CONFIGURED}"
          if [ -z "$BASE" ]; then
            echo "::warning::No DATA_BASE_URL is set, so the deployed app has no data source to load. See LAUNCH_CHECKLIST.md step 2."
            echo "go=false" >> "$GITHUB_OUTPUT"
          else
            echo "go=true" >> "$GITHUB_OUTPUT"
            echo "base=$BASE" >> "$GITHUB_OUTPUT"
            echo "url=${OVERRIDE:-$DEFAULT_URL}" >> "$GITHUB_OUTPUT"
          fi

      - uses: actions/checkout@v4
        if: steps.can.outputs.go == 'true'

      - uses: actions/setup-node@v4
        if: steps.can.outputs.go == 'true'
        with:
          node-version: "22"
          cache: npm
          cache-dependency-path: client/package-lock.json

      - name: Install dependencies
        if: steps.can.outputs.go == 'true'
        run: npm ci

      # Playwright pins a Chromium build per release and refuses to launch a
      # different one, so the browser is installed rather than assumed. Only
      # chromium: the check is about one deployment, not cross-browser
      # behaviour, and the other two would triple the install for nothing.
      - name: Install Chromium
        if: steps.can.outputs.go == 'true'
        run: npx playwright install --with-deps chromium

      - name: Load the deployed app
        if: steps.can.outputs.go == 'true'
        env:
          APP_URL: ${{ steps.can.outputs.url }}
          DATA_BASE_URL: ${{ steps.can.outputs.base }}
        # set +e / PIPESTATUS rather than trusting the pipe: the step's exit
        # status is tee's under the default shell, so a crash of the checker
        # was invisible here (#655, same fix as check-deployment.yml).
        # --exit-zero reserves non-zero for exactly that crash.
        run: |
          set +e
          node scripts/check-deployed-app.mjs \
            --url "$APP_URL" \
            --data-base "$DATA_BASE_URL" \
            --json deployed-app.json \
            --exit-zero | tee deployed-app.txt
          status=${PIPESTATUS[0]}
          set -e
          if [ "$status" -ne 0 ]; then
            echo "::error::check-deployed-app.mjs exited $status, which --exit-zero reserves for a crash."
            exit "$status"
          fi

      - name: Write the job summary
        if: always() && steps.can.outputs.go == 'true'
        run: |
          {
            echo "## Deployed app"
            echo
            if [ -f deployed-app.txt ]; then
              echo '```'
              cat deployed-app.txt
              echo '```'
            else
              echo "The check did not produce a report. See the step log above."
            fi
          } >> "$GITHUB_STEP_SUMMARY"

      # Its own issue title, sharing the label with the other two health
      # checks: they can be red for unrelated reasons and would otherwise
      # overwrite each other's findings, while the shared label keeps all
      # three findable at once.
      - name: Open, update, or close the tracking issue
        if: steps.can.outputs.go == 'true'
        uses: actions/github-script@v7
        env:
          RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
        with:
          script: |
            const fs = require('fs')
            const trackingIssue = require(`${process.env.GITHUB_WORKSPACE}/.github/scripts/tracking-issue.js`)
            const verdict = JSON.parse(fs.readFileSync('client/deployed-app.json', 'utf8'))

            const failed = verdict.failed ?? []

            await trackingIssue({github, context, core}, {
              label: process.env.ISSUE_LABEL,
              title: process.env.ISSUE_TITLE,
              checkedAt: verdict.checked_at,
              keys: failed.map(f => f.check),
              healthy: failed.length === 0,
              allClear: `The deployed app loads its map data again, as of ${verdict.checked_at}. Closing.`,

              render: (firstSeen, cell) => [
                `**A browser loading ${verdict.url} does not get a working map.** ${failed.length} check(s) failed.`,
                '',
                'This is the app rather than the bucket. If "Check deployment" is also red, fix that one first — it is the same fault seen earlier in the chain.',
                '',
                '| check | first seen | detail |',
                '|---|---|---|',
                ...failed.map(f => `| \`${f.check}\` | ${firstSeen[f.check]} | ${cell(f.detail)} |`),
                '',
                `Last checked ${verdict.checked_at} · [run](${process.env.RUN_URL})`,
                '',
                '---',
                '_Generated by [Claude Code](https://claude.ai/code)_',
              ].join('\n'),
            })