e2e.yml

name: E2E

# Playwright end-to-end tests. Boots the full docker-compose stack
# (backend + frontend + runner image), then drives the app via headless
# Chromium. OpenAI is mocked at the network layer — no real API hits from CI.
#
# Kept as a separate workflow (not a job in ci.yml) because this one needs
# Docker + the full runner image build, which is materially slower than the
# lint/typecheck/unit jobs. Keeping them apart means a broken e2e run does
# not block the fast-feedback signal on unit tests.

on:
  push:
    branches: [main]
    paths-ignore:
      - '**.md'
      - 'docs/**'
      - '.github/ISSUE_TEMPLATE/**'
      - 'LICENSE'
  pull_request:
    paths-ignore:
      - '**.md'
      - 'docs/**'
      - '.github/ISSUE_TEMPLATE/**'
      - 'LICENSE'

concurrency:
  group: e2e-${{ github.ref }}
  cancel-in-progress: true

jobs:
  e2e:
    name: Playwright (chromium)
    runs-on: ubuntu-latest
    # Sharding splits the spec set across N parallel jobs (each its
    # own ubuntu-latest runner with its own docker-compose stack), so
    # we get effective N×2 = 8 parallelism without saturating a single
    # runner. Larger runners aren't free on the Free plan even for
    # public repos, so this is the right shape for our cost ceiling.
    # Each shard's wall-clock should land ~10-12 min instead of the
    # 25-min ceiling 2 workers were pushing.
    strategy:
      fail-fast: false
      matrix:
        shard: [1, 2, 3, 4]
    # Bumped from 15 → 25 in Phase 22C: full e2e sweep grew past the 15-min
    # ceiling once the marketing specs landed. Sharding (Phase 27-v2.2)
    # means each shard runs ~1/4 of the suite, so 25 stays generous.
    timeout-minutes: 25
    # Phase 18d: cloud-only Supabase model. These feed both docker compose
    # (which reads process env for ${VAR} interpolation in docker-compose.yml)
    # and the Playwright specs (fixtures + auth.spec.ts read process.env).
    # Values come from the codetutor-dev project — see docs/DEVELOPMENT.md.
    env:
      SUPABASE_URL: ${{ secrets.SUPABASE_URL }}
      SUPABASE_ANON_KEY: ${{ secrets.SUPABASE_ANON_KEY }}
      SUPABASE_SERVICE_ROLE_KEY: ${{ secrets.SUPABASE_SERVICE_ROLE_KEY }}
      VITE_SUPABASE_URL: ${{ secrets.VITE_SUPABASE_URL }}
      VITE_SUPABASE_ANON_KEY: ${{ secrets.VITE_SUPABASE_ANON_KEY }}
      DATABASE_URL: ${{ secrets.DATABASE_URL }}
      BYOK_ENCRYPTION_KEY: ${{ secrets.BYOK_ENCRYPTION_KEY }}
      # Phase 20-P3: e2e-only bearer for /api/metrics. Not a real secret —
      # the backend validates against this exact string, and the metrics
      # spec uses it to prove the authed path works. Local runs can set
      # METRICS_TOKEN in .env to the same value, or leave it blank (in
      # which case the spec's token branch skips).
      METRICS_TOKEN: e2e-test-metrics-token
      # Per-shard test-user email namespace — see e2e/fixtures/auth.ts
      # `workerEmail`. Without this, shard 1's `e2e-w0-${pid}@…` could
      # collide with shard 2's worker 0 if they happen to roll the
      # same suffix. Pinning to the matrix shard guarantees each
      # shard's auth-teardown scope is its own.
      E2E_USER_SUFFIX: shard-${{ matrix.shard }}
      # Phase 20-P3: Playwright cycles test pages faster than `pagehide`-driven
      # session cleanup reaches the backend, so prod-tight caps (2/user) block
      # the next test for the full sweeper interval (~45 s) and manifest as
      # "Run Code button disabled 30 s" flakes that pass on retry. The auth
      # fixture now explicitly ends each test's session in afterEach (see
      # e2e/fixtures/auth.ts), which drops the per-user leak rate dramatically;
      # the loose cap below is belt-and-suspenders for any session id the
      # sniffer misses (early-failed tests, mid-flight rebinds). Prod posture
      # (config.ts defaults 2/20/8) is unaffected.
      MAX_SESSIONS_PER_USER: "60"
      MAX_SESSIONS_GLOBAL: "200"
      DOCKER_EXEC_CONCURRENCY: "16"
    steps:
      - uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4

      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
        with:
          node-version: '20'
          cache: npm
          cache-dependency-path: e2e/package-lock.json

      - name: Boot docker-compose stack
        run: docker compose up -d backend frontend

      - name: Install e2e deps
        working-directory: e2e
        run: npm ci

      - name: Cache Playwright browsers
        uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4
        with:
          path: ~/.cache/ms-playwright
          key: ${{ runner.os }}-playwright-${{ hashFiles('e2e/package-lock.json') }}
          restore-keys: |
            ${{ runner.os }}-playwright-

      - name: Install Playwright browsers
        working-directory: e2e
        run: npx playwright install --with-deps chromium

      - name: Run Playwright tests
        working-directory: e2e
        # --shard partitions specs across the matrix — each runner
        # picks up ~1/4 of the suite and runs it in parallel with the
        # other shards. Each shard has its own docker-compose stack +
        # 2 Playwright workers, so the effective parallelism is 8
        # without any single runner being starved.
        run: npx playwright test --shard=${{ matrix.shard }}/4

      - name: Dump docker-compose logs on failure
        if: failure()
        run: docker compose logs --no-color --tail=300

      - name: Upload Playwright report
        if: always()
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
        with:
          # Shard-scoped artifact name — actions/upload-artifact@v4
          # rejects duplicate names across matrix jobs, so each shard
          # uploads its own report blob.
          name: playwright-report-shard-${{ matrix.shard }}
          path: e2e/playwright-report
          retention-days: 14
          if-no-files-found: ignore

      - name: Upload Playwright traces/videos
        if: failure()
        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
        with:
          name: playwright-traces-shard-${{ matrix.shard }}
          path: e2e/test-results
          retention-days: 14
          if-no-files-found: ignore

      - name: Tear down docker-compose
        if: always()
        run: docker compose down -v