Django channels sounds like a great idea and I really like the conceptual step it made with ASGI, but in practice I found it to be riddled with problems.
For instance, the posix IPC implementation of ASGI is very buggy and "not officially supported". The other implementation uses Redis (so, yet another component to take care of) and is quite brittle as well. We've noticed severe slowdowns of downloading images when using the Daphne web server in development mode, many issues running out of memory, clogged queues and so on. This is also very difficult to debug, due to the many moving parts, and of course it only happens on production, so good luck trying to reproduce it locally.
For this project (which is still in production), we ended up falling back to a standard Gunicorn web server setup using traditional WSGI to handle regular HTTP traffic, and relying on channels only for websockets. For new projects, we've moved to a tiny standalone gevent-based websockets server (https://pypi.org/project/high-templar/) which we deploy next to a standard gunicorn stack. It is much simpler and much more stable. Unfortunately, it's something we've developed in-house, so if there are issues we have to fix them ourselves. There've been some problems with gevent which apparently have been fixed in the latest git repo, but there hasn't been a release since last year. So many issues....
For instance, the posix IPC implementation of ASGI is very buggy and "not officially supported". The other implementation uses Redis (so, yet another component to take care of) and is quite brittle as well. We've noticed severe slowdowns of downloading images when using the Daphne web server in development mode, many issues running out of memory, clogged queues and so on. This is also very difficult to debug, due to the many moving parts, and of course it only happens on production, so good luck trying to reproduce it locally.
For this project (which is still in production), we ended up falling back to a standard Gunicorn web server setup using traditional WSGI to handle regular HTTP traffic, and relying on channels only for websockets. For new projects, we've moved to a tiny standalone gevent-based websockets server (https://pypi.org/project/high-templar/) which we deploy next to a standard gunicorn stack. It is much simpler and much more stable. Unfortunately, it's something we've developed in-house, so if there are issues we have to fix them ourselves. There've been some problems with gevent which apparently have been fixed in the latest git repo, but there hasn't been a release since last year. So many issues....