From the linked issue:

The latest release, 2.2.2, has a problem with the bootstrap socket being double unregistered.
This only happens occasionally. See the example logs below.

I think the most obvious issue here is that socket node_id=bootstrap-1 is unregistered twice in a row and the second time, it throws. Also, it is related to #177. We suspect #156 has some problem. The PR description does not provide any test about how it solves the 100% CPU usage. Should we revert that pr?

I added logic to catch the KeyError, but also to skip the other logic in the loop if if the socket is closed.
Added tests as well for the _poll() function

Fixes #180