Release Management

Versioning

Semantic Versioning — MAJOR.MINOR.PATCH

Current: v3.5.3 (March 2026) — Python 3.11.14, sklearn 1.8.0, LightGBM 4.6+, FinBERT, PySpark 4.1

Release Checklist

• [ ] All tests passing, coverage >= 85% (actual: 90–98%, 395+ tests)
• [ ] Security scans clean (Gitleaks, Bandit, Trivy)
• [ ] Model metrics meet thresholds (AUC > 0.80, R² > 0.75, Acc > 0.85)
• [ ] Docker images built and pushed to Artifact Registry
• [ ] Integration tests pass
• [ ] Documentation updated

Release Process

1. Version bump → Update __init__.py, pyproject.toml across projects
2. Test → pytest --cov, Docker build, integration tests
3. Tag & push → git tag -a vX.Y.Z && git push --tags
4. Build images → Push to Artifact Registry (us-central1-docker.pkg.dev)
5. Deploy → kubectl rollout restart deployment -n ml-portfolio
6. Verify → Health checks + smoke tests on GKE

Rollback

kubectl rollout undo deployment/<service> -n ml-portfolio

Docker Registry

Images pushed to GCP Artifact Registry: - us-central1-docker.pkg.dev/ml-portfolio-duque-om-202602/ml-portfolio-images/bankchurn:latest - us-central1-docker.pkg.dev/ml-portfolio-duque-om-202602/ml-portfolio-images/nlpinsight:latest - us-central1-docker.pkg.dev/ml-portfolio-duque-om-202602/ml-portfolio-images/chicagotaxi:latest

Emergency Procedures

1. Immediate rollback: kubectl rollout undo deployment/<service> -n ml-portfolio
2. Hotfix branch from main, fix, test, merge
3. Post-mortem within 24 hours

Audit Trail

• All releases tagged in Git
• Docker images immutable in Artifact Registry
• MLflow tracks all model versions and parameters
• CI/CD logs retained for 90 days

---

Last Updated: March 2026 — v3.5.3