def send_levels(self):

        self._send_sub()

    def send_levels_loop(self, delay=1000) -> None:

        now = time.time()

        def done(sec):

            reactor.callLater(max(0,

                                  time.time() - (now + delay)),

                              self.send_levels_loop)

        def err(e):

            log.warn('subclient loop: %r', e)

            reactor.callLater(2, self.send_levels_loop)

        d = self._send_sub()

        d.addCallbacks(done, err)

    def _send_sub(self) -> Deferred:

        try:

            with self.graph.currentState() as g:

                outputSettings = self.get_output_settings(_graph=g)

        except Exception:

            traceback.print_exc()

        return sendToCollector(

            'subclient',

            self.session,

            outputSettings,

            # when KC uses zmq, we get message

            # pileups and delays on collector (even

            # at 20fps). When sequencer uses zmp,

            # it runs great at 40fps. Not sure the

            # difference- maybe Tk main loop?

            useZmq=False)