Nginx + Gunicorn is what we use. It has worked great for us. We have also tried uWSGI, a waitress for our production and Nginx + Gunicorn have been sure shot winners in terms of scalability and performance.

This is our gunicorn.conf that ensures gunicorn does not cause memory leaks

    import multiprocessing
    import os
    import random
    from getenv import env

    def pre_fork(server, worker):
       f = '/tmp/app-initialized'
       open(f, 'w').close()

    bind = 'unix:///tmp/nginx.socket'
    workers = os.environ.get('WEB_CONCURRENCY', 
    (multiprocessing.cpu_count() * 1 - 1))

    MAX_GUNICORN_REQUESTS = env('MAX_GUNICORN_REQUESTS', 100)

    if MAX_GUNICORN_REQUESTS:
       max_requests = MAX_GUNICORN_REQUESTS

    MAX_REQUESTS_JITTER = env('MAX_REQUESTS_JITTER', 25)

    if MAX_REQUESTS_JITTER:
        max_requests_jitter = env('MAX_REQUESTS_JITTER', 25)

Nginx is the one I see the most these days.

Having multiple instances behind a reverse proxy allows you to do rolling deploys, which will have no downtime.

For a while I did nginx with gunicorn. No downtime rolling deploys. Just need a git pull and the proper kill signal.

Now I use kubernetes with uwsgi. Once you are kubernetes any app server works, so went uwsgi because it is very very simple.

I currently use Apache with mod_wsgi. New updates involve just touching the .wsgi file. schema changes are a bit more involved obviously.