def monitor_heartbeats(
coordinator: Coordinator, terminate_event: threading.Event
) -> None:
"""Monitors the heartbeat of participants.
If a heartbeat expires the participant is removed from the :class:`~.Participants`.
Note:
This is meant to be run inside a thread and expects an
:class:`~threading.Event`, to know when it should terminate.
Args:
coordinator (:class:`xain_fl.coordinator.coordinator.Coordinator`): The coordinator
to monitor for heartbeats.
terminate_event (:class:`~threading.Event`): A threading event to signal
that this method should terminate.
"""
logger.info("Heartbeat monitor starting...")
while not terminate_event.is_set():
participants_to_remove: List[str] = []
for participant in coordinator.participants.participants.values():
if participant.heartbeat_expires < time.time():
participants_to_remove.append(participant.participant_id)
for participant_id in participants_to_remove:
coordinator.remove_participant(participant_id)
next_expiration: float = coordinator.participants.next_expiration() - time.time()
logger.debug("Monitoring heartbeats", next_expiration=next_expiration)
time.sleep(next_expiration)
def test_remove_participant():
coordinator = Coordinator(minimum_participants_in_round=1,
fraction_of_participants=1.0)
coordinator.on_message(coordinator_pb2.RendezvousRequest(), "participant1")
assert coordinator.state == coordinator_pb2.State.ROUND
coordinator.remove_participant("participant1")
assert coordinator.participants.len() == 0
assert coordinator.state == coordinator_pb2.State.STANDBY
coordinator.on_message(coordinator_pb2.RendezvousRequest(), "participant1")
assert coordinator.state == coordinator_pb2.State.ROUND
