fossasia/open-event-server

As mentioned above, we are running production eventyay.com on just a 2GB RAM server, so it is scalable to very low requirements. We are running a dev version on just free version of heroku with 512 MB RAM, so it is certainly possible, but as you can guess, you'll get really poor performance and be pushing the boundary of memory. Unless extremely simple, you can't expect an interpreted language's application to run in under 512 MB RAM and give production performance, that can't happen unless you write a native service or in a compiled language like go. And anyway, even if the server took just 10 MB of RAM, the performance on heroku would be extremely bad, not at all recommended for production, they throttle the CPU cycles of even the 1 CPU core they provide in free instance. So, no parallelization and even worse performance. All requests will block until the currently running request is completed, unless you run gunicorn with async worker like eventlet, which has introduced some problems with our payments integrations in past.

In production, web process takes 400-450 MB RAM, you can reduce it by reducing the amount of gunicorn workers from 4 to lower. This will again trade off performance for memory. celery takes 300 MB. And you should know that Python garbage collection is far from state of art so it accumulates memory over time. On a fresh start, clery takes 150 MB RAM and web takes 200 MB for 1 instance (which is what runs on our dev instance). So, you can get away with it on 500 MB plan if you run 1 instance of web and celery and regularly restart instances like heroku does after some time. But, you'll get horrible performance. At least 1 GB RAM is required to even run default production config as it takes about 500-600 MB RAM after some time.