retry_async does not follow the timeout duration

[andrewsg]

(Imported from another source): rpc_call is wrapped by retry_async (_retry.py L26), but it seems that it does not follow the durations sleep_generator provided. By tracing the flow, this block (tasklets.py L260) might be the root cause of why there is almost no delay between retries.

The yielded tasklets.sleep is always suppressed because there is no assignment at the line self.generator.throw if the previous rpc call raises exception. I monkey-patched the block as

if error:
    try:
        traceback = error.__traceback__
    except AttributeError:  # pragma: NO PY3 COVER  # pragma: NO BRANCH  # noqa: E501
        traceback = None
    yielded = self.generator.throw(type(error), error, traceback)
else:
    # send_value will be None if this is the first time
    yielded = self.generator.send(send_value)

and the retry sleep would work.