remove expensive debug logs from controller (#57813)

logs in controller can be expensive for large cluster
logs in autoscaling state and just noisy

---------

Signed-off-by: abrar <abrar@anyscale.com>

Author: abrarsheikh

python/ray/serve/_private/autoscaling_state.py

@@ -932,21 +932,11 @@ def update_running_replica_ids(
         app_state = self._app_autoscaling_states.get(deployment_id.app_name)
         if app_state:
             app_state.update_running_replica_ids(deployment_id, running_replicas)
-        else:
-            logger.warning(
-                f"Cannot update running replica ids for deployment "
-                f"{deployment_id} because the application {deployment_id.app_name} is not registered"
-            )
 
     def on_replica_stopped(self, replica_id: ReplicaID):
         app_state = self._app_autoscaling_states.get(replica_id.deployment_id.app_name)
         if app_state:
             app_state.on_replica_stopped(replica_id)
-        else:
-            logger.warning(
-                f"Cannot invoke callback on replica stopped for replica "
-                f"{replica_id} because the application {replica_id.deployment_id.app_name} is not registered"
-            )
 
     def get_metrics_for_deployment(
         self, deployment_id: DeploymentID
@@ -956,10 +946,6 @@ def get_metrics_for_deployment(
                 deployment_id.app_name
             ].get_replica_metrics_by_deployment_id(deployment_id)
         else:
-            logger.warning(
-                f"Cannot get metrics for deployment "
-                f"{deployment_id} because the application {deployment_id.app_name} is not registered"
-            )
             return {}
 
     def get_total_num_requests_for_deployment(
@@ -970,10 +956,6 @@ def get_total_num_requests_for_deployment(
                 deployment_id.app_name
             ].get_total_num_requests_for_deployment(deployment_id)
         else:
-            logger.warning(
-                f"Cannot get total number of requests for deployment "
-                f"{deployment_id} because the application {deployment_id.app_name} is not registered"
-            )
             return 0
 
     def is_within_bounds(
@@ -992,11 +974,6 @@ def record_request_metrics_for_replica(
         )
         if app_state:
             app_state.record_request_metrics_for_replica(replica_metric_report)
-        else:
-            logger.warning(
-                f"Cannot record request metrics for replica "
-                f"{replica_metric_report.replica_id} because the application {replica_metric_report.replica_id.deployment_id.app_name} is not registered"
-            )
 
     def record_request_metrics_for_handle(
         self,
@@ -1008,11 +985,6 @@ def record_request_metrics_for_handle(
         )
         if app_state:
             app_state.record_request_metrics_for_handle(handle_metric_report)
-        else:
-            logger.warning(
-                f"Cannot record request metrics for handle "
-                f"{handle_metric_report.handle_id} because the application {handle_metric_report.deployment_id.app_name} is not registered"
-            )
 
     def drop_stale_handle_metrics(self, alive_serve_actor_ids: Set[str]) -> None:
         for app_state in self._app_autoscaling_states.values():

python/ray/serve/_private/controller.py

@@ -21,7 +21,6 @@
 from ray.serve._private.application_state import ApplicationStateManager, StatusOverview
 from ray.serve._private.autoscaling_state import AutoscalingStateManager
 from ray.serve._private.common import (
-    RUNNING_REQUESTS_KEY,
     DeploymentID,
     HandleMetricReport,
     NodeId,
@@ -275,9 +274,6 @@ def get_pid(self) -> int:
     def record_autoscaling_metrics_from_replica(
         self, replica_metric_report: ReplicaMetricReport
     ):
-        logger.debug(
-            f"Received metrics from replica {replica_metric_report.replica_id}: {replica_metric_report.aggregated_metrics.get(RUNNING_REQUESTS_KEY)} running requests"
-        )
         latency = time.time() - replica_metric_report.timestamp
         latency_ms = latency * 1000
         if latency_ms > RAY_SERVE_RPC_LATENCY_WARNING_THRESHOLD_MS:
@@ -294,10 +290,6 @@ def record_autoscaling_metrics_from_replica(
     def record_autoscaling_metrics_from_handle(
         self, handle_metric_report: HandleMetricReport
     ):
-        logger.debug(
-            f"Received metrics from handle {handle_metric_report.handle_id} for deployment {handle_metric_report.deployment_id}: "
-            f"{handle_metric_report.queued_requests} queued requests and {handle_metric_report.aggregated_metrics[RUNNING_REQUESTS_KEY]} running requests"
-        )
         latency = time.time() - handle_metric_report.timestamp
         latency_ms = latency * 1000
         if latency_ms > RAY_SERVE_RPC_LATENCY_WARNING_THRESHOLD_MS: