Add cluster and job name labels to Shared Pathways Service metrics

pathwaysutils/experimental/shared_pathways_service/isc_pathways.py

@@ -278,14 +278,16 @@ def __init__(
     self.proxy_server_image = proxy_server_image
     self.proxy_options = proxy_options or ProxyOptions()
     self._old_jax_platforms = None
-    raw_collector = (
-        metrics_collector.MetricsCollector(self.project)
-        if collect_service_metrics
-        else None
-    )
-    self.metrics_collector = metrics_collector.SafeMetricsCollector(
-        raw_collector
-    )
+    if collect_service_metrics:
+      raw_collector = metrics_collector.MetricsCollector(
+          self.project, self.cluster, self._proxy_job_name
+      )
+      self.metrics_collector = metrics_collector.SafeMetricsCollector(
+          raw_collector
+      )
+    else:
+      self.metrics_collector = metrics_collector.SafeMetricsCollector(None)
+
     self.start_time = None
     self._old_jax_backend_target = None
     self._old_jax_platforms_config = None

pathwaysutils/experimental/shared_pathways_service/metrics_collector.py

@@ -75,8 +75,15 @@
 class MetricsCollector:
   """Collects usage metrics for Shared Pathways Service and reports to Cloud Monitoring."""
 
-  def __init__(self, project_id: str):
+  def __init__(
+      self,
+      project_id: str,
+      cluster_name: str,
+      job_name: str | None = None,
+  ):
     self.project_id = project_id
+    self.cluster_name = cluster_name
+    self.job_name = job_name
     self.client = monitoring_v3.MetricServiceClient()
     self.project_name = f"projects/{self.project_id}"
     self._lock = threading.Lock()
@@ -185,6 +192,9 @@ def _send_metric(
   ):
     """Queues a single metric in the buffer."""
     default_labels = {"client_instance_id": self._instance_id}
+    default_labels["cluster_name"] = self.cluster_name
+    if self.job_name:
+      default_labels["job_name"] = self.job_name
     if metric_labels:
       default_labels.update(metric_labels)
     _logger.info(
@@ -244,11 +254,23 @@ def _create_metric_descriptor(
               "description": description,
               "display_name": display_name,
               "unit": unit,
-              "labels": [{
-                  "key": "client_instance_id",
-                  "value_type": "STRING",
-                  "description": "Unique execution identifier",
-              }],
+              "labels": [
+                  {
+                      "key": "client_instance_id",
+                      "value_type": "STRING",
+                      "description": "Unique execution identifier",
+                  },
+                  {
+                      "key": "cluster_name",
+                      "value_type": "STRING",
+                      "description": "GKE cluster name",
+                  },
+                  {
+                      "key": "job_name",
+                      "value_type": "STRING",
+                      "description": "Pathways proxy job name",
+                  },
+              ],
           },
       )
       _logger.info("Created metric descriptor: %s", metric_type)