[core][dashboard] Adds Dashboard RSS metrics (ray-project#47035)

Dashboard has `ray_component_uss_mb` metrics but not `ray_component_rss_mb` but the latter is used in the Ray Grafana dash "Node Memory by Component". Adds it. Note gcs_server is still missing. Signed-off-by: Ruiyang Wang <[email protected]>
srinathk10 · Aug 10, 2024 · 2572bba · 2572bba
1 parent 6706311
commit 2572bba
Show file tree

Hide file tree

Showing 2 changed files with 28 additions and 22 deletions.
diff --git a/python/ray/dashboard/dashboard_metrics.py b/python/ray/dashboard/dashboard_metrics.py
@@ -91,14 +91,22 @@ def __init__(self, registry: Optional[CollectorRegistry] = None):
                 namespace="ray",
                 registry=self.registry,
             )
-            self.metrics_dashboard_mem = Gauge(
+            self.metrics_dashboard_mem_uss = Gauge(
                 "component_uss",
                 "USS usage of all components on the node.",
                 tuple(COMPONENT_METRICS_TAG_KEYS),
                 unit="mb",
                 namespace="ray",
                 registry=self.registry,
             )
+            self.metrics_dashboard_mem_rss = Gauge(
+                "component_rss",
+                "RSS usage of all components on the node.",
+                tuple(COMPONENT_METRICS_TAG_KEYS),
+                unit="mb",
+                namespace="ray",
+                registry=self.registry,
+            )
 
 except ImportError:
 

diff --git a/python/ray/dashboard/modules/metrics/metrics_head.py b/python/ray/dashboard/modules/metrics/metrics_head.py
@@ -306,30 +306,28 @@ def _create_default_prometheus_configs(self):
 
     @dashboard_utils.async_loop_forever(METRICS_RECORD_INTERVAL_S)
     async def record_dashboard_metrics(self):
-        self._dashboard_head.metrics.metrics_dashboard_cpu.labels(
-            ip=self._ip,
-            pid=self._pid,
-            Version=ray.__version__,
-            Component=self._component,
-            SessionName=self._session_name,
-        ).set(float(self._dashboard_proc.cpu_percent()))
-        self._dashboard_head.metrics.metrics_dashboard_mem.labels(
-            ip=self._ip,
-            pid=self._pid,
-            Version=ray.__version__,
-            Component=self._component,
-            SessionName=self._session_name,
-        ).set(float(self._dashboard_proc.memory_full_info().uss) / 1.0e6)
+        labels = {
+            "ip": self._ip,
+            "pid": self._pid,
+            "Version": ray.__version__,
+            "Component": self._component,
+            "SessionName": self._session_name,
+        }
+        self._dashboard_head.metrics.metrics_dashboard_cpu.labels(**labels).set(
+            float(self._dashboard_proc.cpu_percent())
+        )
+        self._dashboard_head.metrics.metrics_dashboard_mem_uss.labels(**labels).set(
+            float(self._dashboard_proc.memory_full_info().uss) / 1.0e6
+        )
+        self._dashboard_head.metrics.metrics_dashboard_mem_rss.labels(**labels).set(
+            float(self._dashboard_proc.memory_full_info().rss) / 1.0e6
+        )
 
         # Report the max lag since the last export, if any.
         if self._event_loop_lag_s_max is not None:
-            self._dashboard_head.metrics.metrics_event_loop_lag.labels(
-                ip=self._ip,
-                pid=self._pid,
-                Version=ray.__version__,
-                Component=self._component,
-                SessionName=self._session_name,
-            ).set(float(self._event_loop_lag_s_max))
+            self._dashboard_head.metrics.metrics_event_loop_lag.labels(**labels).set(
+                float(self._event_loop_lag_s_max)
+            )
             self._event_loop_lag_s_max = None
 
     async def run(self, server):