If you have something built in Flask that you want to run on your own VPS rather than Heroku or another IaaS, it's incredibly easy using gunicorn or uwsgi with nginx. gunicorn is a bit faster and easier to get started with out of the two.
I've been using nginx for reverse proxying mostly, but what's the purpose of using something like gunicorn or uwsgi over using just bare Flask? I'm not very familiar with best practices in the area. Is the goal to filter requests for performance or something else?
The server config in the gist is definitely just a proxy, forwarding all of the requests through to gunicorn. Many people add more to the configuration block, to do something like serve static assets with nginx rather than their Flask application.
"While lightweight and easy to use, Flask’s built-in server is not suitable for production as it doesn’t scale well and by default serves only one request at a time. Some of the options available for properly running Flask in production are documented here."
In short, the Flask server is not really written or tuned for production, so we use gunicorn or wsgi which are WSGI containers that are production ready :)
So I can deploy a Facebook Messenger bot on commodity hardware and services; yet I cannot run Facebook Messenger on my mobile browser or write a compatible Messenger app for iOS or Android?
How do you handle the FB API concurrency limits? I'm not familiar with the messenger API, but the marketing API is pretty strict about not allowing more than a few requests (2? 5?) per second.
Once I hit that, I hashed out a scheme where I serialize and queue outgoing requests (SQS), and then execute them from a lambda that is triggered by a 1 second timer. At this point, I was in a crunch for what was supposed to be a fairly simple task, and modern AWS is somewhat of an unfamiliar environment for me, so I just deployed it as a classic Node worker on Heroku.
Are you using Javascript for that? I thought the API needed to have a server that's always running so the chats can be instant, but is it possible to do that with Lambda?
I'm currently using Modulus and I think it will get pretty expensive if I keep making more bots, so a cheaper way to do it would be nice!
Lambda functions initialize in <10ms when warm, <100ms otherwise, in my experience. They seem to stay warm for 15 minutes or so after a cold start.
The excellent Serverless framework is pure JS, but you can also deploy jars to Lambda, which lets you use any JVM language. Clojure Lambdas have a more extreme profile: ~500ms for a cold start, <1ms once warm. You could keep it warm with a timer event source, which could do double duty as a watchdog / heartbeat thing.
Here's super rough instructions in a gist, and you can find plenty of great guides on Google of course: https://gist.github.com/mcescalante/5db616b9a826605f1df35f79...