Flight-Ops/.gitea/workflows/ci.yml

name: CI

on:
  push:
    branches: [master]
  pull_request:
    branches: [master]
  workflow_dispatch:

jobs:
  ci:
    name: build + unit + e2e
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
        uses: actions/checkout@v4

      # The runner provides a working docker CLI out of the box (full-dind) — no endpoint
      # resolution needed. Job and app containers share the daemon, so put them on one
      # user-defined network with a pinned subnet; the app gets a STATIC IP (172.39.0.10,
      # declared in docker-compose.ci.yml) because Chromium got ERR_SSL_PROTOCOL_ERROR via
      # the DNS alias while node-side HTTP to the same name worked — bypassing DNS removes
      # that variable entirely. A stale network with a different subnet is recreated.
      - name: Prepare e2e network
        run: |
          set -u
          SUBNET=172.39.0.0/24
          if docker network inspect flight-ci-e2e >/dev/null 2>&1; then
            cur=$(docker network inspect flight-ci-e2e --format '{{(index .IPAM.Config 0).Subnet}}')
            if [ "$cur" != "$SUBNET" ]; then
              echo "stale flight-ci-e2e with subnet $cur — recreating with $SUBNET"
              docker network rm flight-ci-e2e || true
              docker network create --subnet "$SUBNET" flight-ci-e2e
            fi
          else
            docker network create --subnet "$SUBNET" flight-ci-e2e
          fi
          if ! docker network connect flight-ci-e2e "$(hostname)"; then
            echo "::error::failed to attach job container '$(hostname)' to flight-ci-e2e — the job must run in a container on the shared daemon (check runner config)" >&2
            exit 1
          fi
          docker network inspect flight-ci-e2e --format 'containers on flight-ci-e2e: {{range .Containers}}{{.Name}} {{end}}'
          echo "E2E_BASE_URL=http://172.39.0.10:3000" >> "$GITHUB_ENV"

      - name: Setup Node
        uses: actions/setup-node@v4
        with:
          node-version: 22

      # bun drives installs (repo ships bun.lock, no package-lock.json); pinned to 1.4.2.
      # Retried: the runner's outbound network/DNS is flaky (transient EAI_AGAIN seen in CI).
      - name: Install bun
        run: |
          set -u
          ok=0
          for i in 1 2 3; do
            if npm i -g bun@1.4.2; then ok=1; break; fi
            echo "::warning::bun install attempt $i/3 failed (likely transient network) — retrying in 5s" >&2
            sleep 5
          done
          [ "$ok" = "1" ] || { echo "::error::bun install failed after 3 attempts" >&2; exit 1; }
          bun --version

      # Job-level dep cache via the runner's built-in cache server. continue-on-error on
      # both steps: an unreachable/broken cache server must never break the run.
      - name: Restore deps cache
        uses: actions/cache@v4
        continue-on-error: true
        with:
          path: |
            node_modules
            $HOME/.bun/install/cache
          key: ${{ runner.os }}-deps-bun-${{ hashFiles('bun.lock') }}

      # Retried: the runner's outbound network/DNS is flaky (transient EAI_AGAIN seen in CI).
      - name: Install dependencies
        run: |
          set -u
          ok=0
          for i in 1 2 3; do
            if bun install --frozen-lockfile; then ok=1; break; fi
            echo "::warning::dependency install attempt $i/3 failed (likely transient network) — retrying in 5s" >&2
            sleep 5
          done
          [ "$ok" = "1" ] || { echo "::error::dependency install failed after 3 attempts" >&2; exit 1; }

      # No explicit save step: caches marked by the restore steps are uploaded automatically at job end.

      # Map e2e tests abort ALL *.openfreemap.org traffic (page.route → route.abort()), assuming
      # styles/fonts/sprites are SELF-HOSTED from /styles, /fonts, /sprites — locally that pack is
      # produced by `npm run assets:download` and reaches the app via the compose volume mount.
      # CI must reproduce it or lib/mapStyles.ts probes /map-assets-manifest.json, finds nothing,
      # falls back to remote OFM styles — which the tests abort → style never loads (the
      # deterministic map.spec.ts wheel-zoom + style-switcher failures). The script needs no env
      # vars/args and writes into ./public (git-ignored, docker-excluded; the app container gets
      # it via the volume mount in docker-compose.ci.yml).
      - name: Restore map assets cache
        uses: actions/cache@v4
        continue-on-error: true
        with:
          path: public/
          key: ${{ runner.os }}-map-assets-${{ hashFiles('scripts/download-map-assets.mjs') }}

      # Retried: the runner's uplink is flaky (same transient EAI_AGAIN pattern as dep install).
      # The job-end cache upload seeds public/ for future runs; a partial pack without
      # map-assets-manifest.json is harmless — the app probes that manifest and falls back to
      # remote styles when it's missing.
      - name: Download map assets (retries)
        run: |
          set -u
          ok=0
          for i in 1 2 3 4 5 6; do
            if npm run assets:download; then ok=1; break; fi
            echo "::warning::map assets download attempt $i/6 failed (likely transient network) — retrying in 5s" >&2
            sleep 5
          done
          [ "$ok" = "1" ] || { echo "::error::map assets download failed after 6 attempts" >&2; exit 1; }

      - name: Build
        run: npm run build

      - name: Unit tests
        run: npx vitest run

      # Browsers run locally in CI (PW_REMOTE_BROWSER=0 below): the shared remote browser
      # server at 192.168.1.4 cannot reach a stack inside dind, and local chromium keeps
      # the job self-contained.
      # Browser binaries (~187 MiB) cached across runs: the runner's uplink is flaky and
      # cdn.playwright.dev downloads time out mid-transfer; once seeded, installs skip the download.
      - name: Restore browser cache
        uses: actions/cache@v4
        continue-on-error: true
        with:
          path: $HOME/.cache/ms-playwright
          key: ${{ runner.os }}-pw-browsers-${{ hashFiles('package.json') }}

      # Retried: cdn.playwright.dev is flaky from this runner — Playwright's internal 5-retry
      # window (~40-90s) is often not enough, but runs get through with a longer one. Progress
      # accumulates across attempts: playwright skips already-installed browsers.
      - name: Install browsers (local)
        run: |
          set -u
          ok=0
          for i in 1 2 3 4 5 6; do
            if npx playwright install --with-deps chromium; then ok=1; break; fi
            echo "::warning::browser install attempt $i/6 failed (flaky cdn.playwright.dev) — retrying in 5s" >&2
            sleep 5
          done
          [ "$ok" = "1" ] || { echo "::error::browser install failed after 6 attempts" >&2; exit 1; }

      # Cache the base image on the shared daemon up front: a single loud failure here
      # beats a 6-minute BuildKit "failed to solve" when the daemon's outbound net is down.
      - name: Pull e2e base image
        run: docker pull node:22-slim

      # Pre-build the e2e image with retries (flaky runner network); the E2E step's own
      # build then becomes a fast no-op. Plain progress keeps BuildKit output visible.
      - name: Build e2e image (retries)
        run: |
          set -u
          ok=0
          for i in 1 2 3; do
            if docker compose -f docker-compose.ci.yml -p flight-ci build --progress=plain; then ok=1; break; fi
            echo "::warning::e2e image build attempt $i/3 failed (likely transient network) — retrying in 5s" >&2
            sleep 5
          done
          [ "$ok" = "1" ] || { echo "::error::e2e image build failed after 3 attempts" >&2; exit 1; }

      # E2E against an isolated stack on the shared daemon — never the production instance.
      # docker-compose.ci.yml joins the flight-ci-e2e network prepared above; project name
      # flight-ci keeps it distinct from production. E2E_DEPLOY_CMD makes the config's
      # webServer block deploy + wait for readiness (http://app:3000 via service DNS);
      # E2E_COMPOSE_ARGS points state.spec.ts' mid-run `docker compose restart flight` at
      # the same file + project. E2E_BASE_URL comes from GITHUB_ENV (Prepare e2e network).
      - name: E2E (isolated stack in dind)
        env:
          PW_REMOTE_BROWSER: "0"
          # Build separately with plain progress so BuildKit step output is visible in the
          # job log (auto progress hides it). Playwright's webServer needs a long-running
          # command — `up -d` returns immediately and triggers "exited early"; `logs -f`
          # keeps it alive, streams app logs into CI output for debugging, and is killed with
          # the job (an orphaned `logs -f` dies on its own when cleanup removes the container).
          E2E_DEPLOY_CMD: "docker compose -f docker-compose.ci.yml -p flight-ci build --progress=plain && docker compose -f docker-compose.ci.yml -p flight-ci up -d && docker compose -f docker-compose.ci.yml -p flight-ci logs -f"
          E2E_COMPOSE_ARGS: "-f docker-compose.ci.yml -p flight-ci"
          # Chromium honors proxy env vars on Linux (node-side HTTP does not) — keep the
          # in-network app host out of any proxy path.
          NO_PROXY: app,localhost,127.0.0.1,.kkhost.pl
          no_proxy: app,localhost,127.0.0.1,.kkhost.pl
        # Connectivity diagnostics now run post-deploy in e2e/ci-diagnostics.ts (globalSetup).
        run: npx playwright test

      # The daemon is shared with production — never leave our stack or network behind.
      # -v drops the named DB volume so every run seeds fresh (no schema drift between PRs).
      - name: Cleanup e2e stack + network
        if: always()
        run: |
          docker compose -f docker-compose.ci.yml -p flight-ci down --remove-orphans -v || true
          docker network remove flight-ci-e2e 2>/dev/null || true