Uvicorn segfaults in a certain GitHub Action

[vytas7]

Recently we have a encountered a strange issue when running Falcon's E2E tests against Firefox. These tests spawn a Falcon (ASGI flavour) app, and interact with its WebSocket/SSE functionality from the browser.

I thought that it was a new version of Firefox, Selenium driver, or similar, and simply disabled the test, given it is not our top focus, and the Chrome E2E test is still working fine.

However, a member of our community @Cycloctane has now traced the problem to be caused by a segfault in Uvicorn, not sure what to make of it:

 Fatal Python error: Segmentation fault

Thread 0x00007ff61fe006c0 (most recent call first):
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/concurrent/futures/thread.py", line 90 in _worker
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/threading.py", line 1012 in run
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/threading.py", line 1075 in _bootstrap_inner
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/threading.py", line 1032 in _bootstrap

Current thread 0x00007ff624eaeb80 (most recent call first):
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/traceback.py", line 336 in walk_stack
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/traceback.py", line 392 in extended_frame_gen
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/traceback.py", line 418 in _extract_from_extended_frame_gen
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/traceback.py", line 395 in extract
  File "<shim>", line ??? in <interpreter trampoline>
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/middleware/proxy_headers.py", line 60 in __call__
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/protocols/http/httptools_impl.py", line 409 in run_asgi
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/asyncio/runners.py", line 118 in run
  File "/opt/hostedtoolcache/Python/3.12.8/x64/lib/python3.12/asyncio/runners.py", line 194 in run
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/server.py", line 66 in run
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/main.py", line 579 in run
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/main.py", line 412 in main
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/click/core.py", line 788 in invoke
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/click/core.py", line 1443 in invoke
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/click/core.py", line 1082 in main
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/click/core.py", line 1161 in __call__
  File "/home/runner/work/falcon/falcon/.tox/e2e_firefox/lib/python3.12/site-packages/uvicorn/__main__.py", line 4 in <module>
  File "<frozen runpy>", line 88 in _run_code
  File "<frozen runpy>", line 198 in _run_module_as_main

Extension modules: uvloop.loop, httptools.parser.parser, httptools.parser.url_parser, websockets.speedups (total: 4)

Can it be a genuine issue in one of Uvicorn's extension modules?
Or is it anything odd in that GitHub Actions environment, and not a real issue?
Or could it be a real issue but only when handling an edge case such as OoM, running out of file handles, etc.

It seems we cannot really reproduce the issue outside of the said GitHub Actions environment. The same test works fine against Chrome.
However, the Firefox CI gate seems to be failing in a 100% reproducible fashion. As I understand from Cycloctane's experiments, using Python 3.11 works, but trying to use other CPython 3.12 minor versions, or a different Ubuntu version, etc, makes no difference.

I skimmed through recent Uvicorn issues, and many seem to be complaining about workers getting killed (see, e.g., #2450, #2506, etc), but at first glance it doesn't really seem to be the same segfault.

[Kludex]

I've never seen this before, and I just found out that there was a bind in websockets. 😅

This doesn't look related to the other issues. It doesn't look like a Uvicorn issue, since it mentions all the cython related packages.