fix: rest - set self.start from client.start Make elasped time include time since client was initialized. So elapsed is as close as we can get to an overall request service time. May need to add rest_elapsed or some other subsystem based timers as we try to track a possible performance regression in 2.3.0.

try:
    # Python 3+
    from ssl import SSLError
except (ImportError, AttributeError):
    # Python 2.5-2.7
    from socket import sslerror as SSLError    # noqa: F401