Optimize Docker build performance for CI/CD

[manavgup]

Problem

Docker builds in GitHub Actions are taking 15+ minutes, making deployments very slow.

Evidence

• IBM Code Engine staging deployment stuck on backend image build for 16+ minutes
• Frontend and backend images both building from scratch each time
• No effective layer caching being utilized

Root Causes

1. Dockerfile inefficiencies: Not optimized for layer caching
2. Large dependency installations: Poetry/npm installs happening in every build
3. No multi-stage build optimization: Could separate build and runtime stages better
4. Rebuilding instead of reusing: Not leveraging pre-built images from publish workflow

Proposed Solutions

Short-term (Quick Win)

• Modify IBM Code Engine workflow to use pre-built images from GHCR
• Only rebuild when Dockerfile changes
• Use latest tag from publish.yml workflow

Medium-term (Dockerfile Optimization)

1. Backend Dockerfile.backend:

   • Better layer ordering (dependencies before code)
   • Multi-stage builds (builder + runtime)
   • Leverage buildkit cache mounts for Poetry

2. Frontend Dockerfile.frontend:

   • Better layer ordering (package.json before code)
   • Multi-stage builds (build + nginx)
   • Leverage buildkit cache mounts for npm

Long-term (Advanced)

• Consider building in IBM Cloud Container Registry (faster network)
• Implement Docker buildx bake for parallel builds
• Use GitHub Actions cache more effectively

Success Criteria

• Deployment time reduced from 15+ minutes to under 5 minutes
• Builds only when code/dependencies change
• Effective layer caching working

Related

• IBM Code Engine deployment workflow: .github/workflows/ibm-code-engine-staging.yml
• Backend Dockerfile: backend/Dockerfile.backend
• Frontend Dockerfile: frontend/Dockerfile.frontend