Reduce GitHub Actions costs through workflow optimizations

[ScottWilliamAnderson]

Summary

Optimize GitHub Actions workflows to reduce billable runner minutes while maintaining all existing functionality and leaving room for future enhancements.

Background

Current workflow analysis reveals several cost optimization opportunities:

• Multiple workflows run on every PR regardless of what files changed
• Playwright tests use 4 parallel shards (4× runner cost)
• test-docker-compose duplicates testing already covered by Playwright
• No Docker layer caching between workflow runs

Proposed Changes

1. Remove test-docker-compose from PRs (keep on master only)

Estimated savings: ~3-5 min per PR

Change the trigger from:

on:
  push:
    branches: [master]
  pull_request:
    types: [opened, synchronize]

to:

on:
  push:
    branches: [master]

Rationale: Playwright tests already validate the full Docker stack works. Running both is redundant for PRs.

---

2. Reduce Playwright shards from 4 to 2

Estimated savings: ~50% of Playwright costs

Update the matrix in playwright.yml:

strategy:
  fail-fast: false
  matrix:
    shardIndex: [1, 2]  # was [1, 2, 3, 4]
    shardTotal: [2]      # was [4]

Update artifact download pattern accordingly.

Tradeoff: Tests will take ~2x longer to complete, but cost half as much.

---

3. Add path filtering to lint-backend.yml

Estimated savings: Skip ~2-3 min on frontend-only PRs

Add a changes job similar to what playwright.yml already uses:

jobs:
  changes:
    runs-on: ubuntu-latest
    outputs:
      backend: ${{ steps.filter.outputs.backend }}
    steps:
      - uses: actions/checkout@v5
      - uses: dorny/paths-filter@v3
        id: filter
        with:
          filters: |
            backend:
              - backend/**
              - pyproject.toml
              - .github/workflows/lint-backend.yml

  lint-backend:
    needs: [changes]
    if: ${{ needs.changes.outputs.backend == 'true' }}
    # ... existing job config

Add an alls-green job for branch protection compatibility.

---

4. Add path filtering to test-backend.yml

Estimated savings: Skip ~5-8 min on frontend-only PRs

Same pattern as above:

filters: |
  backend:
    - backend/**
    - docker-compose*.yml
    - .github/workflows/test-backend.yml

---

5. Add path filtering to generate-client.yml

Estimated savings: Skip ~3-4 min when only frontend/docs change

Only run when backend API-related files change:

filters: |
  api:
    - backend/app/api/**
    - backend/app/models.py
    - backend/app/main.py
    - scripts/generate-client.sh
    - .github/workflows/generate-client.yml

---

6. (Optional/Future) Add Docker layer caching

Potential additional savings: 1-3 min per Docker build

Consider adding Docker layer caching using docker/build-push-action with GitHub Actions cache:

- uses: docker/build-push-action@v5
  with:
    context: .
    cache-from: type=gha
    cache-to: type=gha,mode=max

This is more complex to implement but would speed up Docker builds across all workflows.

---

Checklist

• Update test-docker-compose.yml - remove PR trigger
• Update playwright.yml - reduce to 2 shards
• Update lint-backend.yml - add path filtering + alls-green job
• Update test-backend.yml - add path filtering + alls-green job
• Update generate-client.yml - add path filtering + alls-green job
• Update branch protection rules if needed (require the new alls-green jobs)
• (Optional) Add Docker layer caching to workflows

Estimated Impact

Scenario	Current Minutes (est.)	After Optimization (est.)
Backend-only PR	~35-45 min	~20-25 min
Frontend-only PR	~35-45 min	~10-15 min
Full-stack PR	~35-45 min	~25-30 min
Push to master	~40-50 min	~40-50 min (unchanged)

Note: Actual savings depend on your PR patterns. If most PRs touch both frontend and backend, savings will be closer to 30-40%. If many PRs are frontend-only or backend-only, savings could reach 60-70%.

[ScottWilliamAnderson]

I think this is better suited for testing in a private fork first, I can reopen this once I have a working version.