Wait for E2E Job Pod log stream to be over before finishing E2E tests (…

…#3377)

I think there are cases (eg. OCP) where streaming the E2E Pod logs is
slower than others. In such case, we may end up in a situation where, at
the end of the E2E tests, we haven't retrieved the full go test log.
Which then makes the job fail in CI since test logs cannot be parsed and
transformed to XML.

This commit solves it by waiting for the streamStatus chan to report
something (either err or nil), before returning from the
`monitorTestJob` function.

test/e2e/cmd/run/run.go

@@ -432,7 +432,12 @@ func (h *helper) monitorTestJob(client *kubernetes.Clientset) error {
 					}
 				}
 			case corev1.PodSucceeded:
-				log.Info("Tests completed successfully", "name", newPod.Name)
+				log.Info("Test Job succeeded, waiting for log stream to be over", "name", newPod.Name)
+				streamErr := <-streamStatus
+				if streamErr != nil {
+					log.Error(streamErr, "Stream failure")
+					err = streamErr
+				}
 				cancelFunc()
 			case corev1.PodFailed:
 				log.Info("Pod is in failed state", "name", newPod.Name)
@@ -484,14 +489,12 @@ func (h *helper) streamTestJobOutput(streamStatus chan<- error, client *kubernet
 
 	writer := io.MultiWriter(outputs...)
 	var buffer [logBufferSize]byte
-	if _, err := io.CopyBuffer(writer, stream, buffer[:]); err != nil {
-		if err == io.EOF {
-			log.Info("Log stream ended")
-			return
-		}
-
+	if _, err := io.CopyBuffer(writer, stream, buffer[:]); err != nil && err != io.EOF {
 		streamStatus <- err
+		return
 	}
+	log.Info("Log stream ended")
+	streamStatus <- nil
 }
 
 func (h *helper) kubectlApplyTemplate(templatePath string, templateParam interface{}) (string, error) {