Add application_utilization (cncf#62)

gRPC's client-side weighted round robin currently uses 'cpu_utilization' but the load balancing works with any utilization. Server applications that are not CPU bound can report an application specific utilization metric instead and balance the load better. This change introduces a new top-level field 'application_utilization' for this use. Signed-off-by: Yousuk Seung <[email protected]> Signed-off-by: Sergii Tkachenko <[email protected]>
sergiitk · Oct 6, 2023 · 8fb3f26 · 8fb3f26
1 parent 4003588
commit 8fb3f26
Show file tree

Hide file tree

Showing 3 changed files with 54 additions and 26 deletions.
diff --git a/go/xds/data/orca/v3/orca_load_report.pb.go b/go/xds/data/orca/v3/orca_load_report.pb.go
diff --git a/go/xds/data/orca/v3/orca_load_report.pb.validate.go b/go/xds/data/orca/v3/orca_load_report.pb.validate.go
diff --git a/xds/data/orca/v3/orca_load_report.proto b/xds/data/orca/v3/orca_load_report.proto
@@ -47,4 +47,12 @@ message OrcaLoadReport {
 
   // Application specific opaque metrics.
   map<string, double> named_metrics = 8;
+
+  // Application specific utilization expressed as a fraction of available
+  // resources. For example, an application may report the max of CPU and memory
+  // utilization for better load balancing if it is both CPU and memory bound.
+  // This should be derived from the latest sample or measurement.
+  // The value may be larger than 1.0 when the usage exceeds the reporter
+  // dependent notion of soft limits.
+  double application_utilization = 9 [(validate.rules).double.gte = 0];
 }