The change made for bug 30378 caused a regression in our code by making
lookups for SysLogHandler addresses at init time, instead of making them more lazy. Example:

>>> import logging.handlers
>>> LOGGER = logging.getLogger("logger")
>>> LOGGER.addHandler(logging.handlers.SysLogHandler(address=('resolvessometimesbutnotthefirsttime.com', logging.handlers.SYSLOG_UDP_PORT)))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib64/python2.7/logging/handlers.py", line 767, in __init__
    ress = socket.getaddrinfo(host, port, 0, socktype)
socket.gaierror: [Errno -2] Name or service not known

This is exacerbated by the fact that a lot of code the organization I work for (and to be honest, I have written in the past) initializes global loggers once at import time. If the SysLogHandler is added to the global logger and the DNS resolution isn't possible (/etc/hosts is empty on Unix or does not contain the entry, and DNS out is to lunch), it will fall on its face at initialization time.

There needs to be a more graceful way of initializing loggers like this, or a way of delaying the host resolution, so it fails gracefully and doesn't crash the entire program (restoring the previous behavior).

Example: SMTPHandler doesn't do name resolution until it calls emit, which seems like a much more logical place to do the operation (especially since DNS records can change or become invalid).