def _get_or_start_metric_monitor(self, gc_window_s):
"""Get the metric monitor belonging to this serve cluster.
If the metric monitor does not already exist, it will be started.
"""
try:
self.metric_monitor = ray.util.get_actor(SERVE_METRIC_MONITOR_NAME)
except ValueError:
logger.info("Starting metric monitor with name '{}'".format(
SERVE_METRIC_MONITOR_NAME))
self.metric_monitor = MetricMonitor.options(
detached=True,
name=SERVE_METRIC_MONITOR_NAME).remote(gc_window_s)
# TODO(edoakes): move these into the constructor.
start_metric_monitor_loop.remote(self.metric_monitor)
self.metric_monitor.add_target.remote(self.router)
def init_or_get_metric_monitor(self, gc_window_seconds=3600):
if "metric_monitor" not in self.actor_handle_cache:
[handle] = ray.get(
self.actor_nursery_handle.start_actor.remote(
MetricMonitor,
init_args=(gc_window_seconds, ),
tag="metric_monitor"))
start_metric_monitor_loop.remote(handle)
if "queue_actor" in self.actor_handle_cache:
handle.add_target.remote(
self.actor_handle_cache["queue_actor"])
self.refresh_actor_handle_cache()
return self.actor_handle_cache["metric_monitor"]
def start_metric_monitor(self, gc_window_seconds):
assert self.metric_monitor is None, "Metric monitor already started."
self.metric_monitor = MetricMonitor.remote(gc_window_seconds)
# TODO(edoakes): this should be an actor method, not a separate task.
start_metric_monitor_loop.remote(self.metric_monitor)
self.metric_monitor.add_target.remote(self.router)