Gunicorn worker fails to boot due to NotImplementedError

[codebanesr]

Description:
I'm encountering a NotImplementedError: unsupported platform which causes my Gunicorn worker process to fail. This seems to be originating from the httpx/trio libraries. Has anyone else experienced similar issues?

Error Traceback:

[2024-02-18 21:08:17 -0800] [42822] [ERROR] Exception in worker process
Traceback (most recent call last):
 File "/opt/homebrew/lib/python3.11/site-packages/gunicorn/arbiter.py", line 609, in spawn_worker
  worker.init_process()
 File "/opt/homebrew/lib/python3.11/site-packages/gunicorn/workers/ggevent.py", line 147, in init_process
  super().init_process()
 File "/opt/homebrew/lib/python3.11/site-packages/gunicorn/workers/base.py", line 134, in init_process
  self.load_wsgi()
 File "/opt/homebrew/lib/python3.11/site-packages/gunicorn/workers/base.py", line 146, in load_wsgi
  self.wsgi = self.app.wsgi()
        ^^^^^^^^^^^^^^^
  ... (Remainder of traceback) 
NotImplementedError: unsupported platform

Environment:

• OS: Mac M2 [darwin]
• Python: 3.11.7
• Gunicorn: version 21.2.0
• httpx: Version (output of pip show httpx)

Name: httpx
Version: 0.25.0
Summary: The next generation HTTP client.
Home-page: 
Author: 
Author-email: Tom Christie <[email protected]>
License: 
Location: /opt/homebrew/lib/python3.11/site-packages
Requires: certifi, httpcore, idna, sniffio
Required-by: anthropic, gotrue, openai, postgrest, qdrant-client, storage3, supabase, supafunc

• trio: Version (output of pip show trio)

Name: trio
Version: 0.23.1
Summary: A friendly Python library for async concurrency and I/O
Home-page: https://github.com/python-trio/trio
Author: Nathaniel J. Smith
Author-email: [email protected]
License: MIT OR Apache-2.0
Location: /opt/homebrew/lib/python3.11/site-packages
Requires: attrs, idna, outcome, sniffio, sortedcontainers
Required-by: selenium, trio-websocket

Let me know if any more details are necessary.

[codebanesr]

The error originates here, and it only happens when i try to run the app with gunicorn and eventlet

[CoolCat467]

This sounds incredibly similar to #2848.

EDIT:
In #2848, we found out that basically gevent monkey-patches some things in the select module and that breaks things for trio. Apparently this is intended according to gevent/gevent#2008.

According to gunicorn docs (https://docs.gunicorn.org/en/stable/settings.html#worker-class),
apparently you might want to create a subclass for gunicorn.workers.base.Worker that uses trio

EDIT II:
It might be worth looking into https://github.com/pgjones/hypercorn, which is also a wsgi server. I've personally had a lot of success working with it (my project sane scanner server uses it), and it has built-in support for a trio worker class.

[jakkdl]

The error originates here, and it only happens when i try to run the app with gunicorn and eventlet

right yeah, if we're getting here in the logic then we're neither on windows nor linux (true), and the select module does not appear to have kqueue (so apparently gevent.select does not have kqueue either. So indeed similar to #2848, but would not be resolved by #2928.

[A5rocks]

Yeah in #2928 I could make things work via gevent.monkey.get_original but I don't think that's really the right behavior. That will make trio.run block, I think.

Not entirely sure what is the best outcome to have here.

[A5rocks]

@codebanesr does this still fail? That PR was merged a while ago but I forgot to check back here.

[abhra-bv]

I am using mysql-connector in a flask env . I have gevent for sockets, and the trio still errors out on mac. Im just using mac for development purposes.any way we can get past this in a mac

[A5rocks]

I am using mysql-connector in a flask env . I have gevent for sockets, and the trio still errors out on mac. Im just using mac for development purposes.any way we can get past this in a mac

Sorry I forgot about one extra workaround I should have specified in your issue: httpx, or well really httpcore as of 1.0.6 handles trio raising this error as if trio doesn't exist. So you might be able to bump your dependencies and that will fix things.

[A5rocks]

I'm unable to reproduce this anymore with the following:

(repro-venv) # /t/tmp.IWl7yl0JXm> cat x.py
def app(environ, start_response):
  import trio
  start_response("200 OK", [])
  return [b"Hello, world!\n"]
(repro-venv) # /t/tmp.IWl7yl0JXm> python -m gunicorn -w 4 x:app &
(repro-venv) # /t/tmp.IWl7yl0JXm> [2025-03-20 23:14:40 +0000] [141363] [INFO] Starting gunicorn 23.0.0
[2025-03-20 23:14:40 +0000] [141363] [INFO] Listening at: http://127.0.0.1:8000 (141363)
[2025-03-20 23:14:40 +0000] [141363] [INFO] Using worker: sync
[2025-03-20 23:14:40 +0000] [141364] [INFO] Booting worker with pid: 141364
[2025-03-20 23:14:40 +0000] [141365] [INFO] Booting worker with pid: 141365
[2025-03-20 23:14:41 +0000] [141366] [INFO] Booting worker with pid: 141366
[2025-03-20 23:14:41 +0000] [141367] [INFO] Booting worker with pid: 141367
                                               curl http://127.0.0.1:8000
Hello, world!
(repro-venv) # /t/tmp.IWl7yl0JXm> fg
Send job 1 (python -m gunicorn -w 4 x:app &) to foreground
^C[2025-03-20 23:14:52 +0000] [141363] [INFO] Handling signal: int
[2025-03-20 23:14:53 +0000] [141365] [INFO] Worker exiting (pid: 141365)
[2025-03-20 23:14:53 +0000] [141367] [INFO] Worker exiting (pid: 141367)
[2025-03-20 23:14:53 +0000] [141364] [INFO] Worker exiting (pid: 141364)
[2025-03-20 23:14:53 +0000] [141366] [INFO] Worker exiting (pid: 141366)
[2025-03-20 23:14:53 +0000] [141363] [INFO] Shutting down: Master

I think we must have fixed this issue a few versions ago.