Can you provide a test case which actually illustrates that zombie socket connections present after the HTTPS Connection and Close?
I see the logic in your patch. But I don't see no left over unclosed connections upon ssl.close either.