fix: add remap for extra metrics label (#315)

* fix: add remap for extra metrics label

* fix: syntax issue

Author: Code2Life

internal/config/global_config.go

@@ -3,9 +3,9 @@ package config
 import "fmt"
 
 type GlobalConfig struct {
-	MetricsTTL            string   `yaml:"metricsTTL"`
-	MetricsFormat         string   `yaml:"metricsFormat"`
-	MetricsExtraPodLabels []string `yaml:"metricsExtraPodLabels"`
+	MetricsTTL            string            `yaml:"metricsTTL"`
+	MetricsFormat         string            `yaml:"metricsFormat"`
+	MetricsExtraPodLabels map[string]string `yaml:"metricsExtraPodLabels"`
 
 	AlertRules []AlertRule `yaml:"alertRules"`
 }
@@ -40,7 +40,7 @@ func MockGlobalConfig() *GlobalConfig {
 	return &GlobalConfig{
 		MetricsTTL:            "30d",
 		MetricsFormat:         "influx",
-		MetricsExtraPodLabels: []string{"kubernetes.io/app"},
+		MetricsExtraPodLabels: map[string]string{"kubernetes.io/app": "app"},
 		AlertRules: []AlertRule{
 			{
 				Name:               "mock",

internal/controller/gpunode_controller.go

@@ -21,7 +21,6 @@ import (
 	"encoding/json"
 	"fmt"
 	"maps"
-	"strings"
 
 	tfv1 "github.com/NexusGPU/tensor-fusion/api/v1"
 	"github.com/NexusGPU/tensor-fusion/internal/config"
@@ -384,6 +383,10 @@ func (r *GPUNodeReconciler) createHypervisorPod(ctx context.Context, key client.
 		if err := r.Get(ctx, client.ObjectKey{Name: *pool.Spec.SchedulingConfigTemplate}, schedulingConfigTemplate); err == nil {
 			if schedulingConfigTemplate.Spec.Hypervisor != nil {
 				if cfg, err := json.Marshal(schedulingConfigTemplate.Spec.Hypervisor); err == nil {
+					extraLabelsJson, err := json.Marshal(config.GetGlobalConfig().MetricsExtraPodLabels)
+					if err != nil {
+						return fmt.Errorf("invalid metricsExtraPodLabels config, not valid map: %w", err)
+					}
 					spec.Containers[0].Env = append(spec.Containers[0].Env, corev1.EnvVar{
 						Name:  constants.HypervisorSchedulingConfigEnv,
 						Value: string(cfg),
@@ -392,7 +395,7 @@ func (r *GPUNodeReconciler) createHypervisorPod(ctx context.Context, key client.
 						Value: config.GetGlobalConfig().MetricsFormat,
 					}, corev1.EnvVar{
 						Name:  constants.HypervisorMetricsExtraLabelsEnv,
-						Value: strings.Join(config.GetGlobalConfig().MetricsExtraPodLabels, ","),
+						Value: string(extraLabelsJson),
 					})
 				}
 			}

internal/metrics/recorder.go

@@ -270,8 +270,8 @@ func (mr *MetricsRecorder) RecordMetrics(writer io.Writer) {
 		enc.AddTag("workload", metrics.WorkloadName)
 
 		if config.GetGlobalConfig().MetricsExtraPodLabels != nil {
-			for _, label := range config.GetGlobalConfig().MetricsExtraPodLabels {
-				enc.AddTag(label, metrics.podLabels[label])
+			for k, v := range config.GetGlobalConfig().MetricsExtraPodLabels {
+				enc.AddTag(v, metrics.podLabels[k])
 			}
 		}