How to Run Python in Production

a year ago

Python is dominant for data analysis and machine learning in production.
Recommended package managers: poetry and uv (uv is fast but not yet stable).
Use linters like autoflake, flake8, ruff, isort, and pylint to catch errors in dynamically typed Python code.
Prevent secret leaks with tools like gitguardian, gitleaks, or noseyparker.
Use pre-commit hooks for enforcing code quality.
For web services, FastAPI + gunicorn offers 3X throughput over Flask + gunicorn.
Use data-classes or pydantic for data handling and helper classes for pure functions.
Avoid Python's async and multi-threading due to GIL issues; use multiple processes instead.
Use pip-audit and deptry for dependency management and cleanup.
Ensure legal compliance with licensecheck to avoid problematic library licenses.
Deploy with Docker, avoiding Alpine-based images due to glibc vs musl issues.
Use CPU-only PyTorch for non-GPU deployments to reduce size.
Compile code during Docker builds for faster boot times.
Download external dependencies at build time to avoid delays and ensure hermetic builds.
Run Docker containers as non-root users to reduce attack surface.

Hasty Briefsbeta