Merge pull request #51 from liu-cong/no-fatal-on-metric-error

Do not fail Init if fetch metrics fails. It can recover gracefully.

Author: k8s-ci-robot

pkg/ext-proc/backend/provider.go

@@ -59,11 +59,10 @@ func (p *Provider) GetPodMetrics(pod Pod) (*PodMetrics, bool) {
 
 func (p *Provider) Init(refreshPodsInterval, refreshMetricsInterval time.Duration) error {
 	if err := p.refreshPodsOnce(); err != nil {
-		return fmt.Errorf("failed to init pods: %v", err)
+		klog.Errorf("Failed to init pods: %v", err)
 	}
-	err := p.refreshMetricsOnce()
-	if err != nil {
-		return fmt.Errorf("failed to init metrics: %v", err)
+	if err := p.refreshMetricsOnce(); err != nil {
+		klog.Errorf("Failed to init metrics: %v", err)
 	}
 
 	klog.Infof("Initialized pods and metrics: %+v", p.AllPodMetrics())

pkg/ext-proc/backend/provider_test.go

@@ -71,11 +71,10 @@ func TestProvider(t *testing.T) {
 			datastore: &K8sDatastore{
 				Pods: populateMap(pod1.Pod, pod2.Pod),
 			},
-			initErr: true,
 			want: []*PodMetrics{
 				pod1,
 				// Failed to fetch pod2 metrics so it remains the default values.
-				&PodMetrics{
+				{
 					Pod: Pod{Name: "pod2"},
 					Metrics: Metrics{
 						WaitingQueueSize:    0,