Unless you already know, gevent is a great library built on top of greenlet module and libevent event loop, that allows for easy coroutine-based cooperative multitasking in networked applications. By the way, you know the C10K paper, don't you?
gevent's important features is monkey patching - ability to patch some standard Python modules to make them cooperative. There's no magic here, thanks to Python's dynamic nature
gevent simply replaces the standard APIs with its own cooperative equivalents. Existing blocking code can then be reused without any modifications. Cool, isn't it? Most notably,
gevent patches socket (replacing blocking I/O with libevent-based asynchronous one), thread and threading modules (replacing threads with greenlets).
There are some gotchas, however. Due to the way the patching work, it is important to patch all the necessary modules before they are imported by any 3rd party library you want to use. For example, this snippet works fine:
import gevent.monkey; gevent.monkey.patch_thread() import threading
It is enough, however, to swap the lines:
import threading import gevent.monkey; gevent.monkey.patch_thread()
Now this simple 2 lines snippet triggers an exception:
Exception KeyError: KeyError(3072570908L,) in <module 'threading' from '/usr/lib/python2.7/threading.pyc'> ignored
Why it happens? When
threading module is imported it uses main thread ID as a key in some module-level thread dictionary. At exit time it tries to obtain the thread instance from this dictionary (using current thread ID) to perform some clean up, but the ID of the main thread is no longer the same, thanks to patching. If patching is performed before importing the
threading module (as in the first example above), all is fine, as the thread ID stored at import time comes from patched threads implementation already. For more details, see my StackOverflow answer.
The above example was trivial, things get more nasty in real life scenarios. I came across this problem again recently when trying to use sniffer for testing. It's enough to patch threading in the test file to trigger the problem. Let's start with the following test file,
import gevent.monkey; gevent.monkey.patch_thread() import unittest class TestTesting(unittest.TestCase): def test_nose(self): self.assertTrue(True)
Sniffer works fine when started, nothing indicates the problems to come:
Did not find 'scent.py', running nose: . ---------------------------------------------------------------------- Ran 1 test in 0.009s OK In good standing
But it's enough to touch the test file to make it explode (there are more exceptions, but I copy the first one only for brevity):
Traceback (most recent call last): File "/home/czajnik/env/local/lib/python2.7/site-packages/sniffer/runner.py", line 95, in _run if self.run(): File "/home/czajnik/env/local/lib/python2.7/site-packages/sniffer/runner.py", line 177, in run return super(ScentSniffer, self).run() File "/home/czajnik/env/local/lib/python2.7/site-packages/sniffer/runner.py", line 115, in run import nose File "/home/czajnik/env/local/lib/python2.7/site-packages/nose/__init__.py", line 1, in
from nose.core import collector, main, run, run_exit, runmodule File "/home/czajnik/env/local/lib/python2.7/site-packages/nose/core.py", line 11, in from nose.config import Config, all_config_files File "/home/czajnik/env/local/lib/python2.7/site-packages/nose/config.py", line 8, in from nose.util import absdir, tolist File "/home/czajnik/env/local/lib/python2.7/site-packages/nose/util.py", line 19, in log = logging.getLogger('nose') File "/usr/lib/python2.7/logging/__init__.py", line 1555, in getLogger return Logger.manager.getLogger(name) File "/usr/lib/python2.7/logging/__init__.py", line 1022, in getLogger _acquireLock() File "/usr/lib/python2.7/logging/__init__.py", line 217, in _acquireLock _lock.acquire() File "/usr/lib/python2.7/threading.py", line 122, in acquire me = _get_ident() File "/home/czajnik/env/local/lib/python2.7/site-packages/gevent/thread.py", line 28, in get_ident return id(getcurrent()) TypeError: 'NoneType' object is not callable
Clearly, logging module uses threads (fair enough - it tries to make logging API thread-safe). Unfortunately, not much can be done here, all the imports are performed before our module even gets loaded. As a workaround I've hacked
nose/core.py, adding this line just before
import gevent.monkey; gevent.monkey.patch_thread()
sniffer works fine, yet it's a hack, obviously. For my project it's enough, as I use separate virtual environment and thus I can hack around without consequences. But this problem
definitely asks for some clean solution.