# Dockerfile.visual-runner
FROM mcr.microsoft.com/playwright:v1.40.0-jammy

# Install application-specific fonts
COPY ./fonts /usr/share/fonts/custom
RUN fc-cache -f -v

# Set environment variables for deterministic rendering
ENV PLAYWRIGHT_BROWSERS_PATH=/ms-playwright
ENV FONTCONFIG_PATH=/etc/fonts
ENV NODE_OPTIONS="--max-old-space-size=4096"

# Pre-install Playwright browsers to avoid runtime downloads
RUN npx playwright install --with-deps chromium

// visual.config.ts
import { defineConfig, devices } from '@playwright/test';

export default defineConfig({
  testDir: './visual-suites',
  fullyParallel: true,
  forbidOnly: !!process.env.CI,
  retries: process.env.CI ? 2 : 0,
  workers: process.env.CI ? 1 : undefined,
  use: {
    trace: 'on-first-retry',
    screenshot: 'only-on-failure',
    viewport: { width: 1280, height: 720 },
    launchOptions: {
      args: [
        '--font-render-hinting=none',
        '--disable-gpu',
        '--disable-software-rasterizer'
      ]
    }
  },
  projects: [
    {
      name: 'chromium-stable',
      use: { ...devices['Desktop Chrome'] }
    }
  ],
  reporter: process.env.CI ? 'html' : 'list',
  outputDir: './visual-output'
});

d acyclic graph (DAG) that skips unnecessary artifact downloads and enforces execution order.

# .gitlab-ci.yml (excerpt)
stages:
  - build
  - deploy-review
  - validate-ui
  - cleanup

variables:
  VISUAL_IMAGE: "$CI_REGISTRY_IMAGE/visual-runner:1.0.0"
  REVIEW_URL: "https://${CI_ENVIRONMENT_SLUG}.${CI_DEFAULT_DOMAIN}"

build-app:
  stage: build
  image: node:20-slim
  script:
    - npm ci
    - npm run build
  artifacts:
    paths:
      - dist/
    expire_in: 1 hour

deploy-review-env:
  stage: deploy-review
  image: alpine:latest
  script:
    - echo "Deploying to ${REVIEW_URL}"
  environment:
    name: review/$CI_COMMIT_REF_SLUG
    url: $REVIEW_URL
    on_stop: teardown-review-env

validate-visual-regression:
  stage: validate-ui
  image: $VISUAL_IMAGE
  needs:
    - job: deploy-review-env
      artifacts: false
  script:
    - npx playwright test --config=visual.config.ts
  artifacts:
    when: always
    paths:
      - visual-output/
    reports:
      junit: visual-output/results.xml
    expire_in: 30 days
  allow_failure: true

teardown-review-env:
  stage: cleanup
  when: manual
  script:
    - echo "Tearing down review environment"

Rationale:

• needs: [deploy-review-env] ensures the UI validation job starts immediately after deployment finishes, without waiting for other parallel jobs.
• artifacts: false in the needs block prevents unnecessary artifact downloads, cutting pipeline latency.
• allow_failure: true keeps the pipeline green during the stabilization phase. Flip to false once false positives drop below 2%.
• reports: junit enables GitLab to parse test results directly in the Merge Request widget.

Pitfall Guide

1. Caching Baseline Images

Explanation: GitLab CI cache is scoped to pipeline runs and branch names. Caching baseline PNGs means they are discarded after each pipeline, forcing regeneration and breaking version control. Fix: Store baselines in the repository. Enable Git LFS (git lfs track "*.png") to prevent history bloat. Never add baseline directories to the cache key.

2. Ignoring Protected Variable Scope

Explanation: GitLab CI/CD variables marked as "Protected" are only injected into pipelines running on protected branches. Feature branches will fail to authenticate with cloud services or internal APIs. Fix: Either protect your feature branches, or create a separate unprotected variable group for visual testing credentials. Validate variable availability with echo $VAR_NAME in a debug job.

3. Skipping Environment Stabilization

Explanation: Headless browsers inherit system font metrics and GPU rasterization rules. Shared runners change hardware profiles between jobs, causing pixel-level drift. Fix: Use a versioned Docker image with pinned fonts, disabled GPU acceleration, and explicit anti-aliasing flags. Rebuild the image only when browser versions or font families change.

4. Blocking Merge Requests Prematurely

Explanation: Enforcing visual tests on day one guarantees pipeline failures due to baseline mismatches and environmental noise. Teams quickly disable the job entirely. Fix: Start with allow_failure: true. Run the job in non-blocking mode for 2–3 weeks. Collect false positive data, tune thresholds, and switch to blocking mode only when the failure rate stabilizes below 5%.

5. Misusing dependencies vs needs

Explanation: The dependencies keyword downloads artifacts from all previous jobs in the stage, regardless of relevance. This adds unnecessary I/O and extends pipeline duration. Fix: Use needs to declare explicit job dependencies. Set artifacts: false when you only need execution order, not file transfer. This enables DAG execution and reduces CI wait times by 30–40%.

6. Underestimating Memory Footprint

Explanation: Image comparison algorithms load full-resolution bitmaps into RAM. Running multiple workers simultaneously triggers OOM kills on shared runners. Fix: Limit parallelism to workers: 1 for visual jobs. Increase Node.js heap space via NODE_OPTIONS="--max-old-space-size=4096". For suites exceeding 50 snapshots, migrate to self-managed runners with 8GB+ RAM.

7. Generating Baselines Locally

Explanation: Local machines have different DPI scaling, font smoothing, and browser extensions. Baselines generated locally will fail in CI, creating a false sense of security. Fix: Create a manual trigger job in GitLab CI that runs npx playwright test --update-snapshots. This guarantees baselines are captured in the exact same environment used for validation.

Production Bundle

Action Checklist

• Freeze rendering environment: Build a versioned Docker image with pinned fonts, browser version, and headless flags.
• Enable Git LFS: Track all baseline PNGs to prevent repository bloat and merge conflicts.
• Configure DAG execution: Replace dependencies with needs and disable unnecessary artifact downloads.
• Set non-blocking mode: Deploy visual tests with allow_failure: true for the first 14 days.
• Tune retry logic: Apply retries: 2 and workers: 1 to absorb transient rendering noise.
• Route artifacts correctly: Publish HTML reports and JUnit XML to GitLab for MR widget integration.
• Create baseline regeneration job: Add a manual trigger job that runs --update-snapshots in CI.
• Monitor memory usage: Track job termination codes; migrate to self-managed runners if OOM errors exceed 10%.

Decision Matrix

Scenario	Recommended Approach	Why	Cost Impact
Small team, limited infra	Percy / Cloud SaaS	Zero environment management, vendor handles drift	Per-snapshot pricing scales linearly
Strict compliance, air-gapped	Containerized Playwright + Self-Managed Runners	Full control over rendering, no external dependencies	Higher upfront runner cost, predictable CI minutes
Rapid prototyping, frequent UI changes	BackstopJS + GitLab Artifacts	Fast setup, readable HTML reports, easy baseline review	Moderate maintenance overhead, intermittent project updates
Enterprise scale, custom design system	Custom Docker Image + Playwright + LFS	Deterministic diffs, version-controlled baselines, parallelizable	Initial image build time, but lowest long-term CI cost

Configuration Template

# .gitlab-ci.yml - Visual Regression Pipeline
stages:
  - build
  - deploy-review
  - validate-ui
  - cleanup

variables:
  VISUAL_IMAGE: "$CI_REGISTRY_IMAGE/visual-runner:1.0.0"
  REVIEW_URL: "https://${CI_ENVIRONMENT_SLUG}.${CI_DEFAULT_DOMAIN}"
  PLAYWRIGHT_JUNIT_OUTPUT: "visual-output/results.xml"

build-application:
  stage: build
  image: node:20-slim
  script:
    - npm ci --prefer-offline
    - npm run build
  artifacts:
    paths:
      - dist/
    expire_in: 1 hour

deploy-review-instance:
  stage: deploy-review
  image: alpine:latest
  script:
    - echo "Provisioning review environment at ${REVIEW_URL}"
  environment:
    name: review/$CI_COMMIT_REF_SLUG
    url: $REVIEW_URL
    on_stop: destroy-review-instance

run-visual-validation:
  stage: validate-ui
  image: $VISUAL_IMAGE
  needs:
    - job: deploy-review-instance
      artifacts: false
  script:
    - npx playwright test --config=visual.config.ts
  artifacts:
    when: always
    paths:
      - visual-output/
    reports:
      junit: $PLAYWRIGHT_JUNIT_OUTPUT
    expire_in: 30 days
  allow_failure: true

destroy-review-instance:
  stage: cleanup
  when: manual
  script:
    - echo "Decommissioning review environment"

Quick Start Guide

1. Create the runner image: Copy the Dockerfile example, add your application fonts, and push it to your GitLab container registry with a version tag.
2. Initialize Playwright config: Save the visual.config.ts template in your repository root. Adjust viewport dimensions and retry counts to match your design system.
3. Add the CI job: Paste the run-visual-validation job into your .gitlab-ci.yml. Ensure it references your custom image and declares a needs dependency on your deployment job.
4. Generate first baselines: Trigger the pipeline manually. Once it completes, run npx playwright test --update-snapshots inside the CI environment (or via a dedicated manual job) to capture the initial reference set.
5. Commit and monitor: Push the baseline PNGs to Git LFS. Keep allow_failure: true active for two weeks. Review artifact reports in Merge Requests, tune thresholds, and switch to blocking mode once false positives stabilize.