Demo of django app that loads large data only once.

To run:

• install requirements.txt
• gunicorn in this directory
• Test: /usr/bin/time -f %e curl http://localhost:8000

To see the most important lines: git grep DEMO

• largemem/views.py allocates a large array.
• It also has a view that returns a random value from it. The view waits 5 seconds to simulate computation time (and to verify that indpendent requests are being served by different workers)
• gunicorn imports proj/wsgi.py at load time thanks to preload_app=True
• proj/wsgi.py imports views.py which allocates the array
• gunicorn does fork() to make other processes, which doesn't copy memory: Linux does "copy on write" so the large array is only in memory once.
• Now, multiple workers can serve requests at the same time, with only one copy of the array in memory.

Other notes:

• Nothing here should be special to Django or gunicorn... but the web software has to be made in a way that forks all from one process and lets you import at the start.