Monkey patching breaks usage of subprocess.TimeoutExpired in a virtual env

[dtantsur]

Reproducer (on Python 3.8 for me, also seems to reproduce on 3.6):

import eventlet
eventlet.monkey_patch()

import subprocess


try:
    subprocess.check_call('sleep 10', shell=True, timeout=1)
except subprocess.TimeoutExpired:
    print('okay')
except Exception as exc:
    assert exc.__class__ is subprocess.TimeoutExpired, exc.__class__

Expected: "okay" printed.
Actual:

Traceback (most recent call last):
  File "../_timeout.py", line 8, in <module>
    subprocess.check_call('sleep 10', shell=True, timeout=1)
  File "/usr/lib64/python3.8/subprocess.py", line 359, in check_call
    retcode = call(*popenargs, **kwargs)
  File "/usr/lib64/python3.8/subprocess.py", line 342, in call
    return p.wait(timeout=timeout)
  File "/home/dtantsur/Projects/oslo.concurrency/.tox/py3/lib/python3.8/site-packages/eventlet/green/subprocess.py", line 90, in wait
    raise TimeoutExpired(self.args, timeout)
subprocess.TimeoutExpired: Command 'sleep 10' timed out after 1 seconds

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "../_timeout.py", line 12, in <module>
    assert exc.__class__ is subprocess.TimeoutExpired, exc.__class__
AssertionError: <class 'subprocess.TimeoutExpired'>

Using

from eventlet.green import subprocess

fixes the problem but assumes modifying the code which is supposed to work after monkey patching.

[dtantsur]

This issue seems highly dependent the presence of a virtual env (and moon phases). I could not reproduce it with system python and eventlet on my Fedora, but can easily reproduce in a fresh venv.

I do wonder if we should avoid overriding TimeoutExpired for Python's that have it.

[temoto]

Maybe related to this branch https://github.com/eventlet/eventlet/commits/subprocess-413-more 18c889e though it doesn't apply to TimeoutExpired and maybe should.