resolve comments

Author: spacewander

pkg/ext-proc/backend/pod_metrics_refresher.go

@@ -5,6 +5,7 @@ import (
 	"time"
 
 	klog "k8s.io/klog/v2"
+	logutil "sigs.k8s.io/gateway-api-inference-extension/pkg/ext-proc/util/logging"
 )
 
 type PodMetricsRefresher struct {
@@ -29,7 +30,7 @@ func NewPodMetricsRefresher(provider *Provider, pod Pod, interval, timeout time.
 
 func (r *PodMetricsRefresher) start() {
 	go func() {
-		klog.V(2).Infof("Starting refresher for pod %v", r.pod)
+		klog.V(logutil.DEFAULT).InfoS("Starting refresher", "pod", r.pod)
 		for {
 			select {
 			case <-r.done:
@@ -39,7 +40,7 @@ func (r *PodMetricsRefresher) start() {
 
 			err := r.refreshMetrics()
 			if err != nil {
-				klog.Errorf("Failed to refresh metrics for pod %s: %v", r.pod, err)
+				klog.ErrorS(err, "Failed to refresh metrics", "pod", r.pod)
 			}
 
 			time.Sleep(r.interval)
@@ -60,7 +61,7 @@ func (r *PodMetricsRefresher) refreshMetrics() error {
 		return nil
 	}
 
-	klog.V(4).Infof("Processing pod %v and metric %v", pod, existing.Metrics)
+	klog.V(logutil.DEBUG).InfoS("Refresh metrics", "pod", pod, "metrics", existing.Metrics)
 	updated, err := r.provider.pmc.FetchMetrics(ctx, r.pod, existing)
 	if err != nil {
 		return err
@@ -71,6 +72,6 @@ func (r *PodMetricsRefresher) refreshMetrics() error {
 }
 
 func (r *PodMetricsRefresher) stop() {
-	klog.V(2).Infof("Stopping refresher for pod %v", r.pod)
+	klog.V(logutil.DEFAULT).InfoS("Stopping refresher", "pod", r.pod)
 	close(r.done)
 }

pkg/ext-proc/backend/provider.go

@@ -11,9 +11,7 @@ import (
 )
 
 const (
-	// TODO: make it configurable. One idea is to provide a configuration singleton
-	// and put fields like refreshMetricsInterval in it. So far, we have to pass these
-	// fields across several layers.
+	// TODO: https://github.com/kubernetes-sigs/gateway-api-inference-extension/issues/336
 	metricsValidityPeriod = 5 * time.Second
 )
 
@@ -64,7 +62,7 @@ func (p *Provider) allPodMetrics(stale bool) []*PodMetrics {
 		if !stale {
 			if isPodMetricsStale(m) {
 				// exclude stale metrics for scheduler
-				klog.V(4).Infof("Pod metrics for %s is stale, skipping", m.Pod)
+				klog.V(logutil.DEBUG).InfoS("Pod metrics is stale, skipping", "pod", m.Pod)
 			} else {
 				res = append(res, m)
 			}
@@ -84,7 +82,7 @@ func (p *Provider) allPodMetrics(stale bool) []*PodMetrics {
 func (p *Provider) UpdatePodMetrics(pod Pod, pm *PodMetrics) {
 	pm.Metrics.UpdatedTime = time.Now()
 	p.podMetrics.Store(pod, pm)
-	klog.V(4).Infof("Updated metrics for pod %s: %v", pod, pm.Metrics)
+	klog.V(logutil.DEBUG).InfoS("Updated metrics", "pod", pod, "metrics", pm.Metrics)
 }
 
 func (p *Provider) GetPodMetrics(pod Pod) (*PodMetrics, bool) {
@@ -156,7 +154,7 @@ func (p *Provider) refreshPodsOnce(refreshMetricsInterval, refreshMetricsTimeout
 		pod := k.(Pod)
 		if _, ok := p.datastore.pods.Load(pod); !ok {
 			p.podMetrics.Delete(pod)
-			if v, ok := p.podMetrics.LoadAndDelete(pod); ok {
+			if v, ok := p.podMetricsRefresher.LoadAndDelete(pod); ok {
 				refresher := v.(*PodMetricsRefresher)
 				refresher.stop()
 			}

pkg/ext-proc/scheduling/scheduler.go

@@ -111,7 +111,6 @@ type PodMetricsProvider interface {
 
 // Schedule finds the target pod based on metrics and the requested lora adapter.
 func (s *Scheduler) Schedule(req *LLMRequest) (targetPod backend.Pod, err error) {
-	klog.V(logutil.VERBOSE).Infof("request: %v; metrics: %+v", req, s.podMetricsProvider.AllFreshPodMetrics())
 	pods, err := s.filter.Filter(req, s.podMetricsProvider.AllFreshPodMetrics())
 	if err != nil || len(pods) == 0 {
 		return backend.Pod{}, fmt.Errorf(