Refactor DCGM exporter constants and update metrics exporter labels

run-ai · Dec 2, 2024 · df53acc · df53acc
1 parent 67eb954
commit df53acc
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 7 deletions.
diff --git a/internal/common/constants/constants.go b/internal/common/constants/constants.go
@@ -17,6 +17,7 @@ const (
 	LabelTopologyCMNodeName         = "node-name"
 	LabelApp                        = "app"
 
+	DCGMExporterApp     = "nvidia-dcgm-exporter"
 	KwokDCGMExporterApp = "kwok-nvidia-dcgm-exporter"
 
 	ReservationNs = "runai-reservation"

diff --git a/internal/status-exporter/export/metrics/exporter.go b/internal/status-exporter/export/metrics/exporter.go
@@ -18,8 +18,12 @@ import (
 )
 
 const (
-	exporterPort          = 9400
-	exporterContainerName = "nvidia-dcgm-exporter"
+	exporterPort = 9400
+
+	exporterJobName        = constants.DCGMExporterApp
+	exporterServiceName    = constants.DCGMExporterApp
+	exporterContainerName  = constants.DCGMExporterApp
+	exporterHostnamePrefix = constants.DCGMExporterApp
 )
 
 type MetricsExporter struct {
@@ -116,7 +120,7 @@ func generateFakeHostname(nodeName string) string {
 	h := sha1.New()
 	h.Write([]byte(nodeName))
 	nodeNameSHA1 := h.Sum(nil)
-	nodeHostname := fmt.Sprintf("%s-%x", "nvidia-dcgm-exporter", nodeNameSHA1[:3])
+	nodeHostname := fmt.Sprintf("%s-%x", exporterHostnamePrefix, nodeNameSHA1[:3])
 	return nodeHostname
 }
 
@@ -132,8 +136,8 @@ func (e *MetricsExporter) enrichWithPrometheusLabels(labels prometheus.Labels) p
 	labels["pod"] = viper.GetString(constants.EnvImpersonatePodName)
 
 	labels["instance"] = fmt.Sprintf("%s:%d", viper.GetString(constants.EnvImpersonatePodIP), exporterPort)
-	labels["job"] = "nvidia-dcgm-exporter"
-	labels["service"] = "nvidia-dcgm-exporter"
+	labels["job"] = exporterJobName
+	labels["service"] = exporterServiceName
 
 	return labels
 }
diff --git a/internal/status-updater/handlers/node/fake_node_deployments.go b/internal/status-updater/handlers/node/fake_node_deployments.go
@@ -17,6 +17,10 @@ import (
 	"k8s.io/utils/ptr"
 )
 
+const (
+	dummyDcgmExporterPodTimeout = 5 * time.Minute
+)
+
 func (p *NodeHandler) applyFakeNodeDeployments(node *v1.Node) error {
 	if !isFakeNode(node) {
 		return nil
@@ -145,10 +149,10 @@ func (p *NodeHandler) generateFakeNodeDeploymentFromTemplate(template *appsv1.De
 }
 
 func (p *NodeHandler) getDummyDcgmExporterPod(nodeName string) (*v1.Pod, error) {
-	labelSelector := "app=nvidia-dcgm-exporter"
+	labelSelector := fmt.Sprintf("%s=%s", constants.LabelApp, constants.DCGMExporterApp)
 	fieldSelector := fields.OneTermEqualSelector("spec.nodeName", nodeName).String()
 
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+	ctx, cancel := context.WithTimeout(context.Background(), dummyDcgmExporterPodTimeout)
 	defer cancel()
 
 	watcher, err := p.kubeClient.CoreV1().Pods(v1.NamespaceAll).Watch(ctx, metav1.ListOptions{