Deployment

A FastAPI application has no build step: you deploy it by running it on an ASGI server. This chapter covers running a production server with fastapi run and Uvicorn, choosing the number of worker processes, building a container image, and what to watch for behind a reverse proxy.

Running in production with fastapi run

fastapi dev is for development; in production, use fastapi run. It turns off auto-reload and by default listens on port 8000 on all network interfaces (0.0.0.0).

# a single process
fastapi run app/main.py

# another port, with four worker processes
fastapi run app/main.py --port 80 --workers 4

fastapi run uses Uvicorn under the hood. Running Uvicorn yourself gives the same result; in that case you point it at the app as module:variable instead of a file path.

uvicorn app.main:app --host 0.0.0.0 --port 8000 --workers 4

Workers and process management

A single Python process effectively uses one CPU core, so to use several cores you start several processes with --workers. Start around the number of cores and tune with load testing. Each worker has its own memory, so keep an eye on memory use if your app loads large models.

If you deploy straight onto a server, use a process manager such as systemd so the app restarts after a reboot or a crash. On platforms like Kubernetes, where the cluster manages how many containers run, the usual approach is a single process per container and scaling by adding replicas.

Building a container image

Containers let you reproduce the runtime environment exactly. The approach recommended in the official docs starts from the official Python image, installs dependencies first and copies the code afterwards, so that changing only the code reuses the cached dependency layer.

FROM python:3-slim

WORKDIR /code

COPY ./requirements.txt /code/requirements.txt
RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt

COPY ./app /code/app

CMD ["fastapi", "run", "app/main.py", "--port", "80"]

In a real project, pin the image tag to the same Python version you develop with. Write CMD in the exec (array) form shown above so that stop signals reach the app and your lifespan cleanup code actually runs.

docker build -t hello-fastapi .
docker run -d --name hello -p 8000:80 --env-file .env hello-fastapi

Keep .env out of the image and supply it at run time with --env-file or your platform's secrets feature. Listing .venv, .env and __pycache__ in .dockerignore keeps the image small.