screentinker/.github/workflows/ci.yml
screentinker 8cb67122ad
Fix a load-time crash that took down any install collecting statistics (#279)
1.9.35 could not start with TELEMETRY_COLLECTOR=1. It threw before listening:

  ReferenceError: Cannot access 'db' before initialization
      at server.js:986

and systemd restarted it in a loop. Production was down until it was rolled back.

The mount passed the module-scope `db` to the collector's factory, but that binding is
declared ~275 lines further down. The inline handler this replaced only touched `db`
inside a request callback — which runs long after the binding exists — so moving the same
reference into a factory argument turned a lazy read into an eager one. Now resolved as
`require('./db/database').db`, the way every neighbouring call site in that region does it.

WHY NOTHING CAUGHT IT. The block is gated on a flag that only the statistics-collecting
deployment sets. It had therefore never executed in CI, on alpha, or in any test — 1676
tests, four green jobs, a clean alpha deploy, and the crashing line had still never run.
The unit tests mount the router directly and pass a db, which is precisely the part that
was fine.

So the boot smoke now boots WITH the collector enabled and asserts its routes answer:
/api/public/stats returns the expected shape, and a malformed report is refused with 400.
Booting alone would not be enough — the collector could mount and be broken.

Confirmed by reproduction: the released code fails to boot under that flag, and this does
not. 1676/1676 pass.
2026-08-14 12:17:41 -05:00

163 lines
6.8 KiB
YAML

name: CI
on:
push:
branches: [main]
pull_request:
branches: [main]
workflow_dispatch:
permissions:
contents: read
# main gets frequent pushes - cancel an in-flight run when a newer commit
# (or rerun) supersedes it, per ref.
concurrency:
group: ci-${{ github.ref }}
cancel-in-progress: true
jobs:
test:
name: Unit tests (node --test)
runs-on: ubuntu-latest
defaults:
run:
working-directory: server
steps:
- uses: actions/checkout@v6
with:
# The player-parity baselines judge a claim against the SHIPPED source
# (`git show <latest tag>:…`), because a baseline describes what an
# UN-UPDATED display can do. The default shallow checkout has no tags, so
# the suite silently fell back to the working tree and the biconditionals
# inverted: fixing a player's payload bug made CI demand a baseline change
# for displays that cannot possibly have the fix yet. Green locally, red
# here, for a reason found nowhere in the diff.
fetch-depth: 0
- uses: actions/setup-node@v6
with:
node-version: '20'
cache: npm
cache-dependency-path: server/package-lock.json
- run: npm ci
- run: npm test
openapi:
name: OpenAPI spec lint
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: actions/setup-node@v6
with:
node-version: '20'
- name: Lint the public API spec
run: npx --yes @redocly/cli@latest lint docs/openapi.yaml
# Contract integrity: the spec documents ONLY the token-reachable public surface.
# A JWT-only router (admin/auth/provision/...) appearing here is a security flag,
# not a convenience - fail loudly. (The runtime partition test is a separate suite
# that will cross-check the spec against the live mount list.)
- name: Assert spec is public-only
run: |
BAD=$(grep -oE '^ /(admin|auth|workspaces|ai|provision|white-label|status|subscription|stripe|teams|player-debug|contact|tokens)\b' docs/openapi.yaml || true)
if [ -n "$BAD" ]; then echo "::error::JWT-only path(s) leaked into the public spec:"; echo "$BAD"; exit 1; fi
if grep -qE 'unassigned|/prune' docs/openapi.yaml; then echo "::error::token-denied endpoint present in public spec"; exit 1; fi
echo "OK: spec is public-only"
android-test:
name: Android unit tests (Kotlin schedule evaluator vectors)
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: actions/setup-java@v4
with:
distribution: temurin
java-version: '17'
# Cache ~/.gradle/caches AND ~/.gradle/wrapper. The wrapper cache holds the
# Gradle distribution (gradle-8.5-bin.zip); without it every run re-downloaded
# ~130MB from services.gradle.org, and a transient reset there failed the job
# before any test ran (org.gradle.wrapper.Install.forceFetch). On a cache hit the
# distribution is already present, so the download — and its flake — is skipped.
cache: gradle
- uses: android-actions/setup-android@v3
# ScheduleEvalTest reads the SHARED shared/schedule-vectors.json (wired via
# the test task in app/build.gradle.kts), so a ScheduleEval.kt change that
# breaks the contract fails here.
- name: Kotlin evaluator vector conformance
working-directory: android
run: ./gradlew :app:testDebugUnitTest --no-daemon
smoke:
name: Boot smoke + version check
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: actions/setup-node@v6
with:
node-version: '20'
cache: npm
cache-dependency-path: server/package-lock.json
- name: Install deps
working-directory: server
run: npm ci
# Boot against a fresh SQLite db (clean checkout = no db yet). SELF_HOSTED
# makes the first user an admin with no billing. No certs present, so the
# server listens on plain HTTP at :3001. Background it and wait until it
# answers.
- name: Boot server
working-directory: server
env:
SELF_HOSTED: 'true'
# Boot WITH the collector on. This block is config-gated and only the
# statistics-collecting deployment sets the flag, so it had never executed in CI,
# on alpha, or in any test - and a load-time crash inside it took production down
# while every check was green. Code only one deployment runs is exactly the code
# CI has to execute.
TELEMETRY_COLLECTOR: '1'
run: |
node server.js > "$RUNNER_TEMP/server.log" 2>&1 &
echo $! > "$RUNNER_TEMP/server.pid"
for i in $(seq 1 30); do
curl -sf http://localhost:3001/api/status >/dev/null && exit 0
sleep 1
done
echo "server did not come up within 30s:"; cat "$RUNNER_TEMP/server.log"; exit 1
# Assert the public status endpoint is healthy and reports exactly the
# VERSION file - this is what proves the single-source-of-truth wiring.
- name: Assert /api/status ok and version matches VERSION
run: |
STATUS="$(curl -sf http://localhost:3001/api/status)"
echo "status: $STATUS"
EXPECTED="$(cat VERSION)"
REPORTED="$(echo "$STATUS" | jq -r .version)"
echo "VERSION file: $EXPECTED reported: $REPORTED"
test "$(echo "$STATUS" | jq -r .status)" = "ok"
test "$REPORTED" = "$EXPECTED"
echo "OK: status ok, version $REPORTED matches VERSION"
# Booting is not enough on its own - the collector could be mounted and broken. Prove
# the routes it adds actually answer, so a fault inside that block fails here rather
# than on the single deployment that turns it on.
- name: Assert the collector routes answer when enabled
run: |
STATS="$(curl -sf http://localhost:3001/api/public/stats)"
echo "stats: $STATS"
test "$(echo "$STATS" | jq -r 'has("screens") and has("installs")')" = "true"
REPORT="$(curl -s -o /dev/null -w '%{http_code}' -X POST \
-H 'Content-Type: application/json' -d '{"bad":1}' \
http://localhost:3001/api/telemetry/report)"
echo "malformed report -> HTTP $REPORT"
test "$REPORT" = "400"
echo "OK: collector mounted and answering"
- name: Stop server
if: always()
run: kill "$(cat "$RUNNER_TEMP/server.pid")" 2>/dev/null || true
# TODO (deferred - needs a tag earlier than HEAD, so meaningful from v1.8.0 on):
# upgrade-path job. Restore a db created by the previous tagged release, boot
# the current code against it, and assert migrations complete and /api/status
# is healthy. Add once a prior release tag exists.