Files
Stirling-PDF/.github/workflows/docker-compose-tests.yml
T
Ludy ec3de16c08 ci: centralize Gradle caching across GitHub Actions workflows (#7546)
## Summary

This pull request restructures Gradle dependency caching across the
GitHub Actions workflows.

The central `gradle-cache-prime` job is responsible for preparing the
shared backend Gradle cache. Reusable workflows restore that shared
cache without writing to the same key, while independently triggered
workflows use isolated cache namespaces.

## What changed

### Shared Gradle cache

- Added a stable `gradle-v1-` cache namespace for the shared backend
cache.
- The cache key includes the runner OS, runner architecture, JDK
version, and the relevant Gradle configuration files.
- The cache key is calculated before Gradle runs and reused for the
later save step.
- The prime job performs a lookup first and resolves backend
dependencies only when the exact cache is missing.
- This prevents Gradle or Spotless changes during the prime step from
producing a different save key from the key used by downstream jobs.

### Reusable workflows

- Backend, OpenAPI, license, Docker, E2E, and migration workflows
restore the shared cache instead of writing to the shared key.
- The backend build matrix includes `matrix.jdk-version` in its cache
key.
- Enterprise, Tauri, and generated-model workflows support the
`use_shared_cache` boolean input.
- When `use_shared_cache` is enabled, those workflows restore the shared
cache.
- When it is disabled, they use workflow-specific cache namespaces.

### Independent workflows

Independent workflows now use separate cache prefixes, including:

- `gradle-license-report-v1-`
- `gradle-swagger-v1-`
- `gradle-push-docker-v1-`
- `gradle-tauri-releases-v1-`
- `gradle-deploy-pr-v1-`
- `gradle-playwright-e2e-v1-`
- `gradle-generated-models-v1-`

This prevents them from creating or affecting the shared backend cache
before the prime job.

### Build and E2E flow

- Removed the `-PnoSpotless` option from the central Gradle
dependency-resolution command.
- Removed the separate Gradle dependency prime/retry logic from the live
E2E workflow.
- Connected the Tauri build and generated-models check to the central
cache-prime job.

## Motivation

Previously, multiple workflows could use and save the same Gradle cache
key independently. The first workflow to save the cache could therefore
determine its contents, even if it had resolved a different or
incomplete set of dependencies.

The cache key was also evaluated after some Gradle tasks had run. If
Gradle or Spotless modified a file covered by `hashFiles(...)`, the save
key could differ from the restore key used by downstream jobs.

This change gives the shared cache a single owner, isolates
workflow-specific caches, and makes cache usage deterministic across the
CI pipeline.

## Expected result

- `gradle-cache-prime` is the single writer for the shared backend
Gradle cache.
- Downstream jobs restore the same cache without competing cache writes.
- Independently triggered workflows remain isolated through their own
cache namespaces.
- Changes to the monitored Gradle configuration files produce a new
cache key.
- The normal Gradle/Spotless path is included when the shared cache is
populated.

## Validation

- Compared the cache key expressions and `hashFiles(...)` inputs across
the affected workflows.
- Verified that the central restore and save steps use the same
precomputed key.
- CI should confirm that the prime job populates the shared cache and
downstream workflows only restore it.

## Checklist

- [ ] I have read the [Contribution
Guidelines](https://github.com/Stirling-Tools/Stirling-PDF/blob/main/CONTRIBUTING.md)
- [ ] I have performed a self-review of my changes
- [ ] I have run the relevant CI checks
- [ ] I have tested the workflow changes
2026-08-18 19:25:55 +01:00

176 lines
7.1 KiB
YAML

name: Docker Compose Cucumber tests
# Reusable workflow called from build.yml when project / docker / testing
# sources change. Boots the docker-compose stack and runs the cucumber
# scenarios in testing/cucumber.
on:
workflow_call:
inputs:
docker-base-changed:
description: "Whether the docker base image changed (forwarded from files-changed)."
required: false
type: string
default: "false"
permissions:
contents: read
jobs:
docker-compose-tests:
environment: ci-unsigned
runs-on: ubuntu-latest
permissions:
actions: write
contents: read
checks: write
steps:
- name: Harden Runner
uses: step-security/harden-runner@b09bb98e06d4d774595224525879c09bc6e98c40 # v2.20.1
with:
egress-policy: audit
- name: Checkout Repository
uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
- name: Restore cache Gradle User Home
uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
with:
path: |
~/.gradle/caches
~/.gradle/wrapper
key: gradle-v1-${{ runner.os }}-${{ runner.arch }}-jdk-25-${{ hashFiles('gradle/wrapper/gradle-wrapper.properties', 'gradle/libs.versions.toml', 'buildSrc/**', 'settings.gradle', 'build.gradle', 'app/**/build.gradle', 'gradle/**/*.gradle') }}
- name: Set up JDK 25
uses: actions/setup-java@b6effb05e454b25005698d916606bdc6ffcbf961 # v5.7.0
with:
java-version: "25"
distribution: "temurin"
# When the PR changes the base image, test.sh builds it locally
# (stirling-pdf-base:local) into the daemon image store. A buildx
# container builder can't see that store, so skip it here and let
# `docker buildx build` fall back to the default docker driver, which
# resolves the local base. The gha cache backend is also skipped (its
# runtime token isn't exposed) since the docker driver can't use it.
- name: Set up Docker Buildx
if: inputs.docker-base-changed != 'true'
uses: docker/setup-buildx-action@bb05f3f5519dd87d3ba754cc423b652a5edd6d2c # v4.2.0
# Expose ACTIONS_RUNTIME_TOKEN / ACTIONS_RESULTS_URL for docker buildx type=gha cache backend.
- name: Expose GitHub runtime for Buildx cache
if: inputs.docker-base-changed != 'true'
uses: crazy-max/ghaction-github-runtime@04d248b84655b509d8c44dc1d6f990c879747487 # v4.0.0
- name: Install Docker Compose
run: |
sudo curl -SL "https://github.com/docker/compose/releases/download/v2.39.4/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
sudo chmod +x /usr/local/bin/docker-compose
- name: Install uv
uses: astral-sh/setup-uv@c771a70e6277c0a99b617c7a806ffedaca235ff9 # v9.0.0
with:
enable-cache: true
cache-dependency-glob: |
engine/pyproject.toml
engine/uv.lock
- name: Install Cucumber and coverage dependencies
run: |
uv sync --project engine --locked --group cucumber --group tools
- name: Extract JaCoCo agent for cucumber coverage
# Stages build/jacoco/jacocoagent.jar where the coverage override
# file bind-mounts it into the cucumber container. The agent jar
# never goes into the published image - this is host-only.
run: ./gradlew copyJacocoAgent -PnoSpotless
- name: Run Docker Compose Tests
run: |
chmod +x ./testing/test_webpages.sh
chmod +x ./testing/test.sh
chmod +x ./testing/test_disabledEndpoints.sh
./testing/test.sh
env:
MAVEN_USER: ${{ secrets.MAVEN_USER }}
MAVEN_PASSWORD: ${{ secrets.MAVEN_PASSWORD }}
MAVEN_PUBLIC_URL: ${{ secrets.MAVEN_PUBLIC_URL }}
DOCKER_BASE_CHANGED: ${{ inputs.docker-base-changed }}
# Tells test.sh to layer testing/compose/docker-compose-coverage.override.yml
# over the cucumber compose so the container starts with the
# JaCoCo agent attached via JAVA_CUSTOM_OPTS.
STIRLING_PDF_TEST_COVERAGE: "1"
- name: Generate cucumber JaCoCo report
# `if: always()` so a behave failure still produces partial
# coverage from whatever endpoints did run. The exec file only
# exists when the container shut down cleanly - guard so the step
# is silent on the (rare) crash path.
if: always()
id: cucumber-coverage
run: |
if [ -s testing/cucumber-coverage/cucumber.exec ]; then
./gradlew jacocoReportFromExec \
-PexecFile=testing/cucumber-coverage/cucumber.exec \
-PreportDir=build/reports/jacoco/cucumber \
-PnoSpotless
echo "report=true" >> "$GITHUB_OUTPUT"
else
echo "::warning::No cucumber .exec at testing/cucumber-coverage/cucumber.exec (container may have crashed before flushing)"
echo "report=false" >> "$GITHUB_OUTPUT"
fi
- name: Cucumber coverage step summary
if: always() && steps.cucumber-coverage.outputs.report == 'true'
run: |
uv run --project engine --locked --group tools python scripts/coverage-summary.py \
--title "Cucumber (docker) JaCoCo coverage" \
--jacoco "cucumber=build/reports/jacoco/cucumber/jacocoTestReport.xml" \
--github-step-summary
- name: Upload cucumber JaCoCo report
if: always() && steps.cucumber-coverage.outputs.report == 'true'
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: jacoco-cucumber-${{ github.run_id }}
path: build/reports/jacoco/cucumber/
retention-days: 7
- name: Upload raw cucumber .exec for aggregate merge
# Picked up by the coverage-aggregate workflow via the
# `jacoco-exec-*` artifact name pattern.
if: always() && steps.cucumber-coverage.outputs.report == 'true'
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: jacoco-exec-cucumber
path: testing/cucumber-coverage/cucumber.exec
retention-days: 7
if-no-files-found: warn
- name: Upload Cucumber Report
if: always()
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: cucumber-report
path: testing/cucumber/report.html
retention-days: 7
if-no-files-found: warn
- name: Upload Test Reports
if: always()
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: docker-compose-test-reports
path: testing/reports/
retention-days: 7
if-no-files-found: warn
- name: Cucumber Test Report
if: always()
uses: dorny/test-reporter@a43b3a5f7366b97d083190328d2c652e1a8b6aa2 # v3.0.0
with:
name: Cucumber Tests
path: testing/cucumber/junit/*.xml
reporter: java-junit
fail-on-error: false