Comment 4 for bug 1229475

I think we're seeing this often enough that we should be concerned. It's in terminate_instance() every time, so it's not completely random. Putting it on the havana-rc1 milestone to raise its visibility.

Definitely looks like we're not cleaning up a closed fd somehow - e.g. a greenthread closes an fd but keeps the handle around and incorrectly read()/poll()s on it? Then rpc/neutronclient create a new connection, re-uses this fd and we see an error that suggests the issue is rpc/neutronclient rather than whatever other thing screwed up.