[Misc] [gpu_optimizer] add namespace info for log (vllm-project#1149)

Author: googs1025

python/aibrix/aibrix/gpu_optimizer/app.py

@@ -81,9 +81,12 @@ def new_deployment(deployment):
             min_replicas = int(label)
 
     return DeploymentStates(
-        deployment.metadata.name,
-        deployment.spec.replicas if deployment.spec.replicas is not None else 0,
-        min_replicas,
+        name=deployment.metadata.name,
+        namespace=deployment.metadate.namespace,
+        replicas=deployment.spec.replicas
+        if deployment.spec.replicas is not None
+        else 0,
+        min_replicas=min_replicas,
     )
 
 
@@ -101,7 +104,7 @@ def start_serving_thread(watch_ver, deployment, watch_event: bool) -> bool:
             watch_ver, deployment_name, namespace, lambda: new_deployment(deployment)
         )
         logger.info(
-            f'Deployment "{deployment_name}" found in watch version {watch_ver}, added to the model monitor for "{model_name}"'
+            f'Deployment "{namespace}/{deployment_name}" found in watch version {watch_ver}, added to the model monitor for "{model_name}"'
         )
         return False
 
@@ -120,11 +123,11 @@ def start_serving_thread(watch_ver, deployment, watch_event: bool) -> bool:
     model_monitors[model_name] = model_monitor
     if watch_event:
         logger.info(
-            f'New model monitor started for "{model_name}". Deployment "{deployment_name}" added.'
+            f'New model monitor started for "{model_name}". Deployment "{namespace}/{deployment_name}" added.'
         )
     else:
         logger.info(
-            f'Model monitor started for existed "{model_name}". Deployment "{deployment_name}" added.'
+            f'Model monitor started for existed "{model_name}". Deployment "{namespace}/{deployment_name}" added.'
         )
     return True
 
@@ -137,15 +140,15 @@ def update_deployment(watch_ver, deployment):
     namespace = deployment.metadata.namespace
     if model_monitor is None:
         logger.warning(
-            f'Updating "{deployment_name}" in the model monitor, but "{model_name}" has not monitored.'
+            f'Updating "{namespace}/{deployment_name}" in the model monitor, but "{model_name}" has not monitored.'
         )
         return
 
     if model_monitor.add_deployment(
         watch_ver, deployment_name, namespace, lambda: new_deployment(deployment)
     ):
         logger.info(
-            f'Updated "{deployment_name}" in the model monitor for "{model_name}".'
+            f'Updated "{namespace}/{deployment_name}" in the model monitor for "{model_name}".'
         )
 
 
@@ -157,20 +160,20 @@ def remove_deployment(deployment):
     namespace = deployment.metadata.namespace
     if model_monitor is None:
         logger.warning(
-            f'Removing "{deployment_name}" from the model monitor, but "{model_name}" has not monitored.'
+            f'Removing "{namespace}/{deployment_name}" from the model monitor, but "{model_name}" has not monitored.'
         )
         return
 
     if model_monitor.remove_deployment(deployment_name, namespace) == 0:
         model_monitor.stop()
         del model_monitors[model_name]
         logger.info(
-            f'Removing "{deployment_name}" from the model monitor, no deployment left in "{model_name}", stopping the model monitor.'
+            f'Removing "{namespace}/{deployment_name}" from the model monitor, no deployment left in "{model_name}", stopping the model monitor.'
         )
         return
 
     logger.info(
-        f'Removing "{deployment_name}" from the model monitor for "{model_name}".'
+        f'Removing "{namespace}/{deployment_name}" from the model monitor for "{model_name}".'
     )
 
 
@@ -316,7 +319,7 @@ def main(signal, timeout):
                     start_serving_thread(watch_version, deployment, False)
                 except Exception as e:
                     logger.warning(
-                        f"Error on handle existing deployment {deployment.metadata.name}: {e}"
+                        f"Error on handle existing deployment {deployment.metadata.namepsace}/{deployment.metadata.name}: {e} "
                     )
         except client.rest.ApiException as ae:
             logger.error(

python/aibrix/aibrix/gpu_optimizer/load_monitor/monitor.py

@@ -44,8 +44,11 @@
 class DeploymentStates:
     """States of a deployment with resource version."""
 
-    def __init__(self, name: str, replicas: int = 1, min_replicas: int = 0):
+    def __init__(
+        self, name: str, namespace: str = "", replicas: int = 1, min_replicas: int = 0
+    ):
         self.name = name
+        self.namespace = namespace
 
         # _replicas stores optimized value
         self._replicas = replicas
@@ -226,7 +229,7 @@ def read_deployment_num_replicas(self, deployment_name: str, namespace: str) ->
         key = self._deployment_entry_point(deployment_name, namespace)
         if key not in self.deployments:
             raise Exception(
-                f"Deployment {namespace}:{deployment_name} of model {self.model_name} is not monitored"
+                f"Deployment {namespace}/{deployment_name} of model {self.model_name} is not monitored"
             )
         return self.deployments[key].replicas
 
@@ -240,7 +243,7 @@ def update_deployment_num_replicas(
         key = self._deployment_entry_point(deployment_name, namespace)
         if key not in self.deployments:
             raise Exception(
-                f"Deployment {namespace}:{deployment_name} of model {self.model_name} is not monitored"
+                f"Deployment {namespace}/{deployment_name} of model {self.model_name} is not monitored"
             )
 
         if overriding:

python/aibrix/aibrix/gpu_optimizer/load_monitor/visualizer.py

@@ -122,7 +122,10 @@ def get_debug_model_montior(
         if profile_reader is not None:
             for _, profile in enumerate(profile_reader.read()):
                 debug_monitor.add_deployment(
-                    "0", profile.gpu, None, DeploymentStates(profile.gpu, 0)
+                    "0",
+                    profile.gpu,
+                    None,
+                    DeploymentStates(name=profile.gpu, replicas=0),
                 )
 
     return debug_monitor