Alyssa Coghlan

I was able to reproduce it without Docker on Fedora 22 - same symptoms as @vpavlin, with only 4 entries in the cache rather than the expected 5. Dependencies were...

Running `py.test --pdb` rather than plain `py.test` so it drops into the debugger when the test fails, I can see that check.cached_total is 10, while check.add_cached is only 9. The...

Oh, now I understand that block comment in test_threads - I didn't realise what cached_property actually did :) In that case, I suppose it's possible that for some reason one...

I'm heading out, but I've kicked off a test run using this modified implementation of run_threads in the tests: ``` def run_threads(self, num_threads): launch = Event() def _replace_in_thread(started): return lambda:...

Even with that modified implementation, that still failed within 28 runs of the test suite. I'd suggest pinging the Fedora python-devel list at this point.

@vpavlin I was suggesting you do that. I'm now assuming there's something odd in the Fedora Python patches, and I'm the wrong person to ask about that.

I only discovered this via a "Prior art" PR on the GitHub "hub" tool (https://github.com/github/hub/pull/1345 ), so +1 for "git repo" being a bit hard to find. My suggestion would...

I was hitting problems along these lines with only a handful of target hosts, and managed to identify two plausible root causes: * concurrent Ansible invocations fighting over the ControlPersist...

It looks like this has been resolved in #291, so should this issue be closed as fixed in the v1.4.0 release?

Attempting to pass the traceback directly from the background process won't work, as tracebacks can't be serialised with the pickle module by default. If that wasn't the case, the multiprocessing...