Seems like worker threads are occasionally cleaned up before they're started

[initialed85]

If there are more calls than workers submitted in the moment before the cleanup runs, it seems that the freshly created threads (to meet the increase in demand) are immediately cleaned up.

This shouldn't happen as I'm checking for thread.ident (to show that thread was ever started) as well as not thread.isAlive() (to show that thread is now stopped).

Need to dig into it.

[lukecwik]

Since isAlive relies on __started being set, there is a race where ident can be set and isAlive returns false during the new threads bootstrapping process. This means that checking ident and then !isAlive can mark the worker as dead during the bootstrap process.

This seems to be an issue from Py 2.7:
https://hg.python.org/cpython/file/2.7/Lib/threading.py#l785
and still is an issue in the current version:
https://github.com/python/cpython/blob/60bd1f88f21073965a444c8b39c4202d015da5d6/Lib/threading.py#L929

Was there a point in time where the new worker was being added to the list of known workers before start() was being invoked?