Does granian support a pre-fork model?

[beaugunderson]

Our application is very large and we're operating in memory-constrained environments. Because of this we use a pre-fork model to benefit from copy-on-write. Is this possible with granian?

[gi0baro]

Nope. Each worker needs a clean interpreter state on boot.

The only implementation that shares the application between workers is the freethreaded one (Python 3.13t), but that's also currently quite experimental.

[beaugunderson]

A-ha. Dang, I really wanted to try granian for our main application but that's a 1.1gb USS increase for four workers:

# granian with 4 workers
$ smem
  PID User     Command                         Swap      USS      PSS      RSS
  156 root     granian asginl 0.0.0.0:8000        0    13820    16771    33380
  175 root     granian asginl 0.0.0.0:8000        0   339632   347512   381496
  177 root     granian asginl 0.0.0.0:8000        0   335112   343640   379008
  180 root     granian asginl 0.0.0.0:8000        0   337292   345873   381364
  183 root     granian asginl 0.0.0.0:8000        0   337252   345831   381304
                                                     1363108
# gunicorn + uvicorn with 4 workers
$ smem
  PID User     Command                         Swap      USS      PSS      RSS
  113 root     gunicorn: master [app.asgi:        0    26376    87634   318888
  139 root     gunicorn: worker [app.asgi:        0    27100    83085   300900
  142 root     gunicorn: worker [app.asgi:        0    15008    74218   298420
  145 root     gunicorn: worker [app.asgi:        0    15016    74219   298420
  148 root     gunicorn: worker [app.asgi:        0   113168   159315   346680
                                                      196668