Merge pull request #29125 from mwielgus/extra-debug-hpa

Automatic merge from submit-queue

Extra debug information in HPA events

Fixes: #29004

pkg/controller/podautoscaler/metrics/metrics_client.go

@@ -27,6 +27,7 @@ import (
 	"k8s.io/kubernetes/pkg/api/v1"
 	clientset "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	"k8s.io/kubernetes/pkg/labels"
+	"k8s.io/kubernetes/pkg/util/sets"
 
 	heapster "k8s.io/heapster/metrics/api/v1/types"
 	metrics_api "k8s.io/heapster/metrics/apis/metrics/v1alpha1"
@@ -172,7 +173,21 @@ func (h *HeapsterMetricsClient) getCpuUtilizationForPods(namespace string, selec
 	}
 
 	if len(metrics) != len(podNames) {
-		return 0, time.Time{}, fmt.Errorf("metrics obtained for %d/%d of pods", len(metrics), len(podNames))
+		present := sets.NewString()
+		for _, m := range metrics {
+			present.Insert(m.Name)
+		}
+		missing := make([]string, 0)
+		for expected := range podNames {
+			if !present.Has(expected) {
+				missing = append(missing, expected)
+			}
+		}
+		hint := ""
+		if len(missing) > 0 {
+			hint = fmt.Sprintf(" (sample missing pod: %s/%s)", namespace, missing[0])
+		}
+		return 0, time.Time{}, fmt.Errorf("metrics obtained for %d/%d of pods%s", len(metrics), len(podNames), hint)
 	}
 
 	sum := int64(0)
@@ -250,7 +265,17 @@ func (h *HeapsterMetricsClient) getCustomMetricForPods(metricSpec metricDefiniti
 
 	sum, count, timestamp := metricSpec.aggregator(metrics)
 	if count != len(podNames) {
-		return nil, time.Time{}, fmt.Errorf("metrics obtained for %d/%d of pods", count, len(podNames))
+		missing := make([]string, 0)
+		for i, expected := range podNames {
+			if len(metrics.Items) > i && len(metrics.Items[i].Metrics) == 0 {
+				missing = append(missing, expected)
+			}
+		}
+		hint := ""
+		if len(missing) > 0 {
+			hint = fmt.Sprintf(" (sample missing pod: %s/%s)", namespace, missing[0])
+		}
+		return nil, time.Time{}, fmt.Errorf("metrics obtained for %d/%d of pods%s", count, len(podNames), hint)
 	}
 
 	return &sum, timestamp, nil

pkg/controller/podautoscaler/metrics/metrics_client_test.go

@@ -184,10 +184,12 @@ func buildPod(namespace, podName string, podLabels map[string]string, phase api.
 }
 
 func (tc *testCase) verifyResults(t *testing.T, val *float64, timestamp time.Time, err error) {
-	assert.Equal(t, tc.desiredError, err)
 	if tc.desiredError != nil {
+		assert.Error(t, err)
+		assert.Contains(t, fmt.Sprintf("%v", err), fmt.Sprintf("%v", tc.desiredError))
 		return
 	}
+	assert.NoError(t, err)
 	assert.NotNil(t, val)
 	assert.True(t, tc.desiredValue-0.001 < *val)
 	assert.True(t, tc.desiredValue+0.001 > *val)
@@ -426,7 +428,7 @@ func TestCPUEmptyMetricsForOnePod(t *testing.T) {
 	tc := testCase{
 		replicas:           3,
 		targetResource:     "cpu-usage",
-		desiredError:       fmt.Errorf("metrics obtained for 2/3 of pods"),
+		desiredError:       fmt.Errorf("metrics obtained for 2/3 of pods (sample missing pod: test-namespace/test-pod-2)"),
 		reportedPodMetrics: [][]int64{{100}, {300, 400}},
 		useMetricsApi:      true,
 	}