Allow HPA to work when Inference is enabled (#4247) (#4248)

Problem: When the Inference Extension was enabled, the additional container was not given resource specifications, which prevented the HPA from working.

Solution: Add the resource specifications from the NGINX container to the inference container.

Author: sjberman

internal/controller/provisioner/objects.go

@@ -1019,6 +1019,7 @@ func (p *NginxProvisioner) buildNginxPodTemplateSpec(
 		},
 	}
 
+	var containerResources corev1.ResourceRequirements
 	if nProxyCfg != nil && nProxyCfg.Kubernetes != nil {
 		var podSpec *ngfAPIv1alpha2.PodSpec
 		var containerSpec *ngfAPIv1alpha2.ContainerSpec
@@ -1042,7 +1043,8 @@ func (p *NginxProvisioner) buildNginxPodTemplateSpec(
 		if containerSpec != nil {
 			container := spec.Spec.Containers[0]
 			if containerSpec.Resources != nil {
-				container.Resources = *containerSpec.Resources
+				containerResources = *containerSpec.Resources
+				container.Resources = containerResources
 			}
 			container.Lifecycle = containerSpec.Lifecycle
 			container.VolumeMounts = append(container.VolumeMounts, containerSpec.VolumeMounts...)
@@ -1174,6 +1176,7 @@ func (p *NginxProvisioner) buildNginxPodTemplateSpec(
 			Image:           p.cfg.GatewayPodConfig.Image,
 			ImagePullPolicy: pullPolicy,
 			Command:         command,
+			Resources:       containerResources,
 			SecurityContext: &corev1.SecurityContext{
 				AllowPrivilegeEscalation: helpers.GetPointer(false),
 				Capabilities: &corev1.Capabilities{

internal/controller/provisioner/objects_test.go

@@ -1883,7 +1883,20 @@ func TestBuildNginxResourceObjects_InferenceExtension(t *testing.T) {
 		},
 	}
 
-	objects, err := provisioner.buildNginxResourceObjects("gw-nginx", gateway, &graph.EffectiveNginxProxy{})
+	npCfg := &graph.EffectiveNginxProxy{
+		Kubernetes: &ngfAPIv1alpha2.KubernetesSpec{
+			Deployment: &ngfAPIv1alpha2.DeploymentSpec{
+				Container: ngfAPIv1alpha2.ContainerSpec{
+					Resources: &corev1.ResourceRequirements{
+						Limits: corev1.ResourceList{
+							corev1.ResourceCPU: resource.MustParse("500m"),
+						},
+					},
+				},
+			},
+		},
+	}
+	objects, err := provisioner.buildNginxResourceObjects("gw-nginx", gateway, npCfg)
 	g.Expect(err).ToNot(HaveOccurred())
 
 	// Find the deployment object
@@ -1907,4 +1920,5 @@ func TestBuildNginxResourceObjects_InferenceExtension(t *testing.T) {
 	g.Expect(containers).To(HaveLen(2))
 	g.Expect(containers[1].Name).To(Equal("endpoint-picker-shim"))
 	g.Expect(containers[1].Command).To(Equal(expectedCommands))
+	g.Expect(containers[1].Resources.Limits).To(HaveKeyWithValue(corev1.ResourceCPU, resource.MustParse("500m")))
 }