Revert "Revert "[autoscaler][observability] Autoscaler emit cluster/p…

…ending logical resource metrics"" (ray-project#30552) See ray-project#29820 for most of the details of the PR. The only difference is improving the NullMetric stub to not fail when tagging labesl for metrics. This reverts commit f9092dc. Co-authored-by: Alex <[email protected]> Signed-off-by: Capiru <[email protected]>
Capiru · Dec 22, 2022 · 2092abe · 2092abe
1 parent 60254d0
commit 2092abe
Showing 1 changed file with 21 additions and 0 deletions.
diff --git a/python/ray/autoscaler/_private/prom_metrics.py b/python/ray/autoscaler/_private/prom_metrics.py
@@ -248,6 +248,27 @@ def __init__(
                 namespace="autoscaler",
                 registry=self.registry,
             )
+            # This represents the autoscaler's view of essentially
+            # `ray.cluster_resources()`, it may be slightly different from the
+            # core metric from an eventual consistency perspective.
+            self.cluster_resources: Gauge = Gauge(
+                "cluster_resources",
+                "Total logical resources in the cluster.",
+                unit="resources",
+                namespace="autoscaler",
+                registry=self.registry,
+                labelnames=["resource"],
+            )
+            # This represents the pending launches + nodes being set up for the
+            # autoscaler.
+            self.pending_resources: Gauge = Gauge(
+                "pending_resources",
+                "Pending logical resources in the cluster.",
+                unit="resources",
+                namespace="autoscaler",
+                registry=self.registry,
+                labelnames=["resource"],
+            )
 
         @property
         def session_name(self):