Merge pull request kubernetes#77716 from wking/job-running-test-error-reporting

test/e2e/upgrades/apps/job: List Pods in failure message

Author: k8s-ci-robot

test/e2e/framework/job/wait.go

@@ -17,6 +17,8 @@ limitations under the License.
 package job
 
 import (
+	"fmt"
+	"strings"
 	"time"
 
 	batchv1 "k8s.io/api/batch/v1"
@@ -99,22 +101,28 @@ func WaitForJobGone(c clientset.Interface, ns, jobName string, timeout time.Dura
 	})
 }
 
-// CheckForAllJobPodsRunning uses c to check in the Job named jobName in ns is running. If the returned error is not
-// nil the returned bool is true if the Job is running.
-func CheckForAllJobPodsRunning(c clientset.Interface, ns, jobName string, parallelism int32) (bool, error) {
+// EnsureAllJobPodsRunning uses c to check in the Job named jobName in ns
+// is running, returning an error if the expected parallelism is not
+// satisfied.
+func EnsureAllJobPodsRunning(c clientset.Interface, ns, jobName string, parallelism int32) error {
 	label := labels.SelectorFromSet(labels.Set(map[string]string{JobSelectorKey: jobName}))
 	options := metav1.ListOptions{LabelSelector: label.String()}
 	pods, err := c.CoreV1().Pods(ns).List(options)
 	if err != nil {
-		return false, err
+		return err
 	}
+	podsSummary := make([]string, 0, parallelism)
 	count := int32(0)
 	for _, p := range pods.Items {
 		if p.Status.Phase == v1.PodRunning {
 			count++
 		}
+		podsSummary = append(podsSummary, fmt.Sprintf("%s (%s: %s)", p.ObjectMeta.Name, p.Status.Phase, p.Status.Message))
+	}
+	if count != parallelism {
+		return fmt.Errorf("job has %d of %d expected running pods: %s", count, parallelism, strings.Join(podsSummary, ", "))
 	}
-	return count == parallelism, nil
+	return nil
 }
 
 // WaitForAllJobPodsGone waits for all pods for the Job named jobName in namespace ns

test/e2e/upgrades/apps/job.go

@@ -55,9 +55,8 @@ func (t *JobUpgradeTest) Setup(f *framework.Framework) {
 func (t *JobUpgradeTest) Test(f *framework.Framework, done <-chan struct{}, upgrade upgrades.UpgradeType) {
 	<-done
 	ginkgo.By("Ensuring active pods == parallelism")
-	running, err := jobutil.CheckForAllJobPodsRunning(f.ClientSet, t.namespace, t.job.Name, 2)
+	err := jobutil.EnsureAllJobPodsRunning(f.ClientSet, t.namespace, t.job.Name, 2)
 	gomega.Expect(err).NotTo(gomega.HaveOccurred())
-	gomega.Expect(running).To(gomega.BeTrue())
 }
 
 // Teardown cleans up any remaining resources.