Merge pull request #50231 from crassirostris/sd-logging-e2e-soak-fix

test/e2e/instrumentation/logging/stackdrvier/soak.go

@@ -84,8 +84,9 @@ var _ = instrumentation.SIGDescribe("Cluster level logging implemented by Stackd
 				for runIdx := 0; runIdx < podRunCount; runIdx++ {
 					// Starting one pod on each node.
 					for _, pod := range podsByRun[runIdx] {
-						err := pod.Start(f)
-						framework.Logf("Failed to start pod: %v", err)
+						if err := pod.Start(f); err != nil {
+							framework.Logf("Failed to start pod: %v", err)
+						}
 					}
 					<-t.C
 				}

test/e2e/instrumentation/logging/utils/logging_agent.go

@@ -65,7 +65,12 @@ func EnsureLoggingAgentRestartsCount(f *framework.Framework, appName string, max
 
 	maxRestartCount := 0
 	for _, pod := range agentPods.Items {
-		restartCount := int(pod.Status.ContainerStatuses[0].RestartCount)
+		contStatuses := pod.Status.ContainerStatuses
+		if len(contStatuses) == 0 {
+			framework.Logf("There are no container statuses for pod %s", pod.Name)
+			continue
+		}
+		restartCount := int(contStatuses[0].RestartCount)
 		maxRestartCount = integer.IntMax(maxRestartCount, restartCount)
 
 		framework.Logf("Logging agent %s on node %s was restarted %d times",

test/e2e/instrumentation/logging/utils/wait.go

@@ -150,9 +150,6 @@ func getFullIngestionPred(podsMap map[string]FiniteLoggingPod) NumberedIngestion
 	return func(name string, occ map[int]bool) (bool, error) {
 		p := podsMap[name]
 		ok := len(occ) == p.ExpectedLineCount()
-		if !ok {
-			framework.Logf("Pod %s is still missing %d lines", name, p.ExpectedLineCount()-len(occ))
-		}
 		return ok, nil
 	}
 }
@@ -160,24 +157,27 @@ func getFullIngestionPred(podsMap map[string]FiniteLoggingPod) NumberedIngestion
 func getFullIngestionTimeout(podsMap map[string]FiniteLoggingPod, slack float64) NumberedTimeoutFun {
 	return func(names []string, occs map[string]map[int]bool) error {
 		totalGot, totalWant := 0, 0
-		podsWithLosses := []string{}
+		lossMsgs := []string{}
 		for _, name := range names {
 			got := len(occs[name])
 			want := podsMap[name].ExpectedLineCount()
 			if got != want {
-				podsWithLosses = append(podsWithLosses, name)
+				lossMsg := fmt.Sprintf("%s: %d lines", name, want-got)
+				lossMsgs = append(lossMsgs, lossMsg)
 			}
 			totalGot += got
 			totalWant += want
 		}
-		if len(podsWithLosses) > 0 {
-			framework.Logf("Still missing logs from: %s", strings.Join(podsWithLosses, ", "))
+		if len(lossMsgs) > 0 {
+			framework.Logf("Still missing logs from:\n%s", strings.Join(lossMsgs, "\n"))
 		}
 		lostFrac := 1 - float64(totalGot)/float64(totalWant)
 		if lostFrac > slack {
 			return fmt.Errorf("still missing %.2f%% of logs, only %.2f%% is tolerable",
 				lostFrac*100, slack*100)
 		}
+		framework.Logf("Missing %.2f%% of logs, which is lower than the threshold %.2f%%",
+			lostFrac*100, slack*100)
 		return nil
 	}
 }