Server deployment is a complex area, that will depend on what kind of service you're deploying Uvicorn onto. As a general rule, you probably want to: * Run `uvicorn --reload` from the command line for local development. * Run `gunicorn -k uvicorn.workers.UvicornWorker` for production. * Additionally run behind Nginx for self-hosted deployments. * Finally, run everything behind a CDN for caching support, and serious DDOS protection. ## Running from the command line Typically you'll run `uvicorn` from the command line. ```bash $ uvicorn main:app --reload --port 5000 ``` The ASGI application should be specified in the form `path.to.module:instance.path`. When running locally, use `--reload` to turn on auto-reloading. The `--reload` and `--workers` arguments are **mutually exclusive**. To see the complete set of available options, use `uvicorn --help`. See the [settings documentation](../settings.md) for more details on the supported options for running uvicorn. ## Running programmatically To run directly from within a Python program, you should use `uvicorn.run(app, **config)`. For example: ```py title="main.py" import uvicorn class App: ... app = App() if __name__ == "__main__": uvicorn.run("main:app", host="127.0.0.1", port=5000, log_level="info") ``` The set of configuration options is the same as for the command line tool. Note that the application instance itself *can* be passed instead of the app import string. ```python uvicorn.run(app, host="127.0.0.1", port=5000, log_level="info") ``` However, this style only works if you are not using multiprocessing (`workers=NUM`) or reloading (`reload=True`), so we recommend using the import string style. Also note that in this case, you should put `uvicorn.run` into `if __name__ == '__main__'` clause in the main module. !!! note The `reload` and `workers` parameters are **mutually exclusive**. ## Using a process manager Running Uvicorn using a process manager ensures that you can run multiple processes in a resilient manner, and allows you to perform server upgrades without dropping requests. A process manager will handle the socket setup, start-up multiple server processes, monitor process aliveness, and listen for signals to provide for processes restarts, shutdowns, or dialing up and down the number of running processes. ### Built-in Uvicorn includes a `--workers` option that allows you to run multiple worker processes. ```bash $ uvicorn main:app --workers 4 ``` Unlike gunicorn, uvicorn does not use pre-fork, but uses [`spawn`](https://docs.python.org/3/library/multiprocessing.html#contexts-and-start-methods), which allows uvicorn's multiprocess manager to still work well on Windows. The default process manager monitors the status of child processes and automatically restarts child processes that die unexpectedly. Not only that, it will also monitor the status of the child process through the pipeline. When the child process is accidentally stuck, the corresponding child process will be killed through an unstoppable system signal or interface. You can also manage child processes by sending specific signals to the main process. (Not supported on Windows.) - `SIGHUP`: Gracefully restart the workers one at a time with no dropped requests. Fresh workers pick up new code on disk. - `SIGTTIN`: Increase the number of worker processes by one. - `SIGTTOU`: Decrease the number of worker processes by one. ### Gunicorn !!! warning The `uvicorn.workers` module is deprecated and will be removed in a future release. You should use the [`uvicorn-worker`](https://github.com/Kludex/uvicorn-worker) package instead. ```bash python -m pip install uvicorn-worker ``` Gunicorn is probably the simplest way to run and manage Uvicorn in a production setting. Uvicorn includes a gunicorn worker class that means you can get set up with very little configuration. The following will start Gunicorn with four worker processes: `gunicorn -w 4 -k uvicorn.workers.UvicornWorker` The `UvicornWorker` implementation uses the `uvloop` and `httptools` implementations. To run under PyPy you'll want to use pure-python implementation instead. You can do this by using the `UvicornH11Worker` class. `gunicorn -w 4 -k uvicorn.workers.UvicornH11Worker` Gunicorn provides a different set of configuration options to Uvicorn, so some options such as `--limit-concurrency` are not yet supported when running with Gunicorn. If you need to pass uvicorn's config arguments to gunicorn workers then you'll have to subclass `UvicornWorker`: ```python from uvicorn.workers import UvicornWorker class MyUvicornWorker(UvicornWorker): CONFIG_KWARGS = {"loop": "asyncio", "http": "h11", "lifespan": "off"} ``` ### Supervisor To use `supervisor` as a process manager you should either: * Hand over the socket to uvicorn using its file descriptor, which supervisor always makes available as `0`, and which must be set in the `fcgi-program` section. * Or use a UNIX domain socket for each `uvicorn` process. A simple supervisor configuration might look something like this: ```ini title="supervisord.conf" [supervisord] [fcgi-program:uvicorn] socket=tcp://localhost:8000 command=venv/bin/uvicorn --fd 0 main:App numprocs=4 process_name=uvicorn-%(process_num)d stdout_logfile=/dev/stdout stdout_logfile_maxbytes=0 ``` Then run with `supervisord -n`. ## Running behind Nginx Using Nginx as a proxy in front of your Uvicorn processes may not be necessary, but is recommended for additional resilience. Nginx can deal with serving your static media and buffering slow requests, leaving your application servers free from load as much as possible. In managed environments such as `Heroku`, you won't typically need to configure Nginx, as your server processes will already be running behind load balancing proxies. The recommended configuration for proxying from Nginx is to use a UNIX domain socket between Nginx and whatever the process manager that is being used to run Uvicorn. If using Uvicorn directly you can bind it to a UNIX domain socket using `uvicorn --uds /path/to/socket.sock <...>`. When running your application behind one or more proxies you will want to make sure that each proxy sets appropriate headers to ensure that your application can properly determine the client address of the incoming connection, and if the connection was over `http` or `https`. For more information see [Proxies and Forwarded Headers](#proxies-and-forwarded-headers) below. Here's how a simple Nginx configuration might look. This example includes setting proxy headers, and using a UNIX domain socket to communicate with the application server. It also includes some basic configuration to forward websocket connections. For more info on this, check [Nginx recommendations](https://nginx.org/en/docs/http/websocket.html). ```conf http { server { listen 80; client_max_body_size 4G; server_name example.com; location / { proxy_set_header Host $http_host; proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for; proxy_set_header X-Forwarded-Proto $scheme; proxy_set_header Upgrade $http_upgrade; proxy_set_header Connection $connection_upgrade; proxy_redirect off; proxy_buffering off; proxy_pass http://uvicorn; } location /static { # path for static files root /path/to/app/static; } } map $http_upgrade $connection_upgrade { default upgrade; '' close; } upstream uvicorn { server unix:/tmp/uvicorn.sock; } } ``` Uvicorn's `--proxy-headers` behavior may not be sufficient for more complex proxy configurations that use different combinations of headers, or where the application is running behind more than one intermediary proxying service. In those cases, you might want to use an ASGI middleware to set the `client` and `scheme` dependent on the request headers. ## Running behind a CDN Running behind a content delivery network, such as Cloudflare or Cloud Front, provides a serious layer of protection against DDoS attacks. Your service will be running behind huge clusters of proxies and load balancers that are designed for handling huge amounts of traffic, and have capabilities for detecting and closing off connections from DDoS attacks. Proper usage of cache control headers can mean that a CDN is able to serve large amounts of data without always having to forward the request on to your server. Content Delivery Networks can also be a low-effort way to provide HTTPS termination. ## Running with HTTPS To run uvicorn with https, a certificate and a private key are required. The recommended way to get them is using [Let's Encrypt](https://letsencrypt.org/). For local development with https, it's possible to use [mkcert](https://github.com/FiloSottile/mkcert) to generate a valid certificate and private key. ```bash $ uvicorn main:app --port 5000 --ssl-keyfile=./key.pem --ssl-certfile=./cert.pem ``` ### Running gunicorn worker It's also possible to use certificates with uvicorn's worker for gunicorn. ```bash $ gunicorn --keyfile=./key.pem --certfile=./cert.pem -k uvicorn.workers.UvicornWorker main:app ``` ### Customizing the SSL context For TLS scenarios that the `--ssl-*` flags don't cover (e.g., mutual TLS, custom `SSLContext.options`, bumping `minimum_version`, loading certificates from memory), pass an `ssl_context_factory` to `uvicorn.run()` or `Config`. The factory receives the `Config` instance and a `default_ssl_context_factory` callable that builds the standard context from the `ssl_*` settings on `Config`. Use it to start from uvicorn's default and mutate it, or ignore it and build your own context from scratch - the `ssl_*` settings are only consumed by the default factory, so if you don't call it they're effectively unused. ```python import ssl from collections.abc import Callable import uvicorn from uvicorn.config import Config def ssl_context_factory(config: Config, default_ssl_context_factory: Callable[[], ssl.SSLContext]) -> ssl.SSLContext: context = default_ssl_context_factory() context.minimum_version = ssl.TLSVersion.TLSv1_3 return context uvicorn.run( "main:app", ssl_keyfile="key.pem", ssl_certfile="cert.pem", ssl_context_factory=ssl_context_factory, ) ``` The factory is called inside each worker process, so it works with `--reload` and `--workers > 1`. The factory itself must be picklable in those modes (a top-level function is fine; lambdas and local closures are not). The `ssl_*` settings on `Config` are only consumed by `default_ssl_context_factory()`; if you build the context yourself without calling it, those settings are ignored. ## Proxies and Forwarded Headers When running an application behind one or more proxies, certain information about the request is lost. To avoid this most proxies will add headers containing this information for downstream servers to read. Uvicorn currently supports the following headers: - `X-Forwarded-For` ([MDN Reference](https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/X-Forwarded-For)) - `X-Forwarded-Proto`([MDN Reference](https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/X-Forwarded-Proto)) Uvicorn can use these headers to correctly set the client and protocol in the request. However as anyone can set these headers you must configure which "clients" you will trust to have set them correctly. Uvicorn can be configured to trust IP Addresses (e.g. `127.0.0.1`), IP Networks (e.g. `10.100.0.0/16`), or Literals (e.g. `/path/to/socket.sock`). When running from CLI these are configured using `--forwarded-allow-ips`. !!! Warning "Only trust clients you can actually trust!" Incorrectly trusting other clients can lead to malicious actors spoofing their apparent client address to your application. A proxy chain may send a header once per hop rather than as a single comma-separated value. Repeated `X-Forwarded-For` fields are combined in order (as the equivalent comma-separated list, [RFC 9110, 5.3](https://www.rfc-editor.org/rfc/rfc9110#section-5.3)), while for `X-Forwarded-Proto` the last field is used. For more information, check [`ProxyHeadersMiddleware`](https://github.com/Kludex/uvicorn/blob/main/uvicorn/middleware/proxy_headers.py). ### Client Port Currently if the `ProxyHeadersMiddleware` is able to retrieve a trusted client value then the client's port will be set to `0`. This is because port information is lost when using these headers. ### UNIX Domain Sockets (UDS) Although it is common for UNIX Domain Sockets to be used for communicating between various HTTP servers, they can mess with some of the expected received values as they will be various non-address strings or missing values. For example: - when NGINX itself is running behind a UDS it will add the literal `unix:` as the client in the `X-Forwarded-For` header. - When Uvicorn is running behind a UDS the initial client will be `None`. ### Trust Everything Rather than specifying what to trust, you can instruct Uvicorn to trust all clients using the literal `"*"`. You should only set this when you know you can trust all values within the forwarded headers (e.g. because your proxies remove the existing headers before setting their own).