diff --git a/pilot/pkg/model/push_context.go b/pilot/pkg/model/push_context.go
index 45f92fab03f4..f5e92a08dee5 100644
--- a/pilot/pkg/model/push_context.go
+++ b/pilot/pkg/model/push_context.go
@@ -354,6 +354,7 @@ type PushRequest struct {
 
 type TriggerReason string
 
+// If adding a new reason, update xds/monitoring.go:triggerMetric
 const (
 	// EndpointUpdate describes a push triggered by an Endpoint change
 	EndpointUpdate TriggerReason = "endpoint"
diff --git a/pilot/pkg/xds/bench_test.go b/pilot/pkg/xds/bench_test.go
index be2f81a3dff6..22f280b11c11 100644
--- a/pilot/pkg/xds/bench_test.go
+++ b/pilot/pkg/xds/bench_test.go
@@ -502,3 +502,45 @@ func createEndpoints(numEndpoints, numServices, numNetworks int) []config.Config
 	}
 	return result
 }
+
+func BenchmarkPushRequest(b *testing.B) {
+	// allTriggers contains all triggers, so we can pick one at random.
+	// It is not a big issue if it falls out of sync, as we are just trying to generate test data
+	allTriggers := []model.TriggerReason{
+		model.EndpointUpdate,
+		model.ConfigUpdate,
+		model.ServiceUpdate,
+		model.ProxyUpdate,
+		model.GlobalUpdate,
+		model.UnknownTrigger,
+		model.DebugTrigger,
+		model.SecretTrigger,
+		model.NetworksTrigger,
+		model.ProxyRequest,
+		model.NamespaceUpdate,
+	}
+	// Number of (simulated) proxies
+	proxies := 500
+	// Number of (simulated) pushes merged
+	pushesMerged := 10
+	// Number of configs per push
+	configs := 1
+
+	for n := 0; n < b.N; n++ {
+		var req *model.PushRequest
+		for i := 0; i < pushesMerged; i++ {
+			trigger := allTriggers[i%len(allTriggers)]
+			nreq := &model.PushRequest{
+				ConfigsUpdated: map[model.ConfigKey]struct{}{},
+				Reason:         []model.TriggerReason{trigger},
+			}
+			for c := 0; c < configs; c++ {
+				nreq.ConfigsUpdated[model.ConfigKey{Kind: gvk.ServiceEntry, Name: fmt.Sprintf("%d", c), Namespace: "default"}] = struct{}{}
+			}
+			req = req.Merge(nreq)
+		}
+		for p := 0; p < proxies; p++ {
+			recordPushTriggers(req.Reason...)
+		}
+	}
+}
diff --git a/pilot/pkg/xds/monitoring.go b/pilot/pkg/xds/monitoring.go
index e7eeabfb0a6a..bbc572897349 100644
--- a/pilot/pkg/xds/monitoring.go
+++ b/pilot/pkg/xds/monitoring.go
@@ -202,9 +202,29 @@ func recordXDSClients(version string, delta float64) {
 	xdsClients.With(versionTag.Value(version)).Record(xdsClientTracker[version])
 }
 
+// triggerMetric is a precomputed monitoring.Metric for each trigger type. This saves on a lot of allocations
+var triggerMetric = map[model.TriggerReason]monitoring.Metric{
+	model.EndpointUpdate:  pushTriggers.With(typeTag.Value(string(model.EndpointUpdate))),
+	model.ConfigUpdate:    pushTriggers.With(typeTag.Value(string(model.ConfigUpdate))),
+	model.ServiceUpdate:   pushTriggers.With(typeTag.Value(string(model.ServiceUpdate))),
+	model.ProxyUpdate:     pushTriggers.With(typeTag.Value(string(model.ProxyUpdate))),
+	model.GlobalUpdate:    pushTriggers.With(typeTag.Value(string(model.GlobalUpdate))),
+	model.UnknownTrigger:  pushTriggers.With(typeTag.Value(string(model.UnknownTrigger))),
+	model.DebugTrigger:    pushTriggers.With(typeTag.Value(string(model.DebugTrigger))),
+	model.SecretTrigger:   pushTriggers.With(typeTag.Value(string(model.SecretTrigger))),
+	model.NetworksTrigger: pushTriggers.With(typeTag.Value(string(model.NetworksTrigger))),
+	model.ProxyRequest:    pushTriggers.With(typeTag.Value(string(model.ProxyRequest))),
+	model.NamespaceUpdate: pushTriggers.With(typeTag.Value(string(model.NamespaceUpdate))),
+}
+
 func recordPushTriggers(reasons ...model.TriggerReason) {
 	for _, r := range reasons {
-		pushTriggers.With(typeTag.Value(string(r))).Increment()
+		t, f := triggerMetric[r]
+		if f {
+			t.Increment()
+		} else {
+			pushTriggers.With(typeTag.Value(string(r))).Increment()
+		}
 	}
 }