Skip to content

Web Server

https://github.com/tiangolo/fastapi/issues/2062

if using docker with kuberntes or some load balancer, i'll recommend: CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8080"]

After trying

  • hypercorn and gunicorn with 1 or 4 workers,

  • hypercorn with trio or default,

  • gunicorn with uvicorn worker I came to the conclusion that with the above configuration I got the best results when receiving many requests on endpoints that make several requests.

uvicorn

uvicorn is an asynchronous ASGI web server. It uses uvloop if it can. Uvicorn supports HTTP/1.1 and WebSockets, but does not (yet) support HTTP/2.

cli

uvicorn 'my_fastapi.fileapi:app' --host '127.0.0.1' --port 7000 --loop 'uvloop' --workers 1 --root-path '/api' --reload

import uvicorn
uvicorn.run(
    'my.api.main:app',
    host='127.0.0.1',
    port=7000,
    reload=False,
)

Settings: https://www.uvicorn.org/settings/

  • --loop: The uvloop implementation provides greater performance, but is not compatible with Windows or PyPy.

  • Underwindows windows can use loop asyncio.

hypercorn

Hypercorn is an asynchronous ASGI web server. It can use uvloop as Uvicorn does, or use other event loops. It supports HTTP/2.

Config: https://pgjones.gitlab.io/hypercorn/how_to_guides/configuring.html

https://levelup.gitconnected.com/deploy-fastapi-with-hypercorn-http-2-asgi-8cfc304e9e7a

hypercorn 'my_fastapi.fileapi:app' --bind 127.0.0.1:5000 --worker-class 'uvloop' --workers 2 --root-path '/api' --reload --debug

#import uvloop
import asyncio
from hypercorn.config import Config
from hypercorn.asyncio import serve

import os
os.environ['HYPERCORN_WORKERS'] = '2'
os.environ['HYPERCORN_DEBUG'] = 'False'
os.environ['HYPERCORN_ROOT_PATH'] = 'c:/api'

config = Config()
config.bind = ['127.0.01:8000']
#config.worker_class = uvloop.Loop # uvloop only for linux
#config.use_reload = auto_reload
config.accesslog = '-'
asyncio.run(serve(app, config))

more example

import click
#import uvloop
import asyncio
from hypercorn.config import Config
from hypercorn.asyncio import serve

from fastapi import FastAPI
from fastapi.middleware.gzip import GZipMiddleware

from . import __version__
from .routers import (
    dev,
    tst,
)

app = FastAPI(
    title='Dev REST API',
    version=__version__,
    docs_url='/',
)

app.add_middleware(GZipMiddleware, minimum_size=100000)

app.include_router(sales.router, prefix='/sales', tags=['Sales'])
app.include_router(marketing.router, prefix='/marketing', tags=['Marketing'])

import os
os.environ['HYPERCORN_WORKERS'] = '2'
os.environ['HYPERCORN_DEBUG'] = 'False'
os.environ['HYPERCORN_ROOT_PATH'] = 'c:/test/api'

@click.command()
@click.option('--host', type=click.STRING, default='127.0.0.1')
@click.option('--port', type=click.INT, default=7000)
@click.option('--auto-reload/--no-auto-reload', default=True)
def cli(**kwargs):
    host = kwargs['host']
    port = kwargs['port']
    auto_reload = kwargs['auto_reload']
    config = Config()
    config.bind = [f'{host}:{port}']
    #config.worker_class = uvloop.Loop # uvloop only for linux
    #config.use_reload = auto_reload
    #config.accesslog = '-'
    asyncio.run(serve(app, config))

gunicorn

gunicorn is a synchronous web server.