From b06675724d6fc1166f72adf7a97a88b15f1ee78e Mon Sep 17 00:00:00 2001
From: Michal Wozniak <michalwozniak@google.com>
Date: Thu, 19 Oct 2023 17:18:30 +0200
Subject: [PATCH] Add e2e test for backoffLimitPerIndex

---
 test/e2e/apps/job.go               | 30 ++++++++++++++++++++++++++++++
 test/e2e/framework/job/fixtures.go | 10 ++++++++++
 2 files changed, 40 insertions(+)

diff --git a/test/e2e/apps/job.go b/test/e2e/apps/job.go
index ca52e157213..12f7dd6ed2b 100644
--- a/test/e2e/apps/job.go
+++ b/test/e2e/apps/job.go
@@ -380,6 +380,36 @@ var _ = SIGDescribe("Job", func() {
 		gomega.Expect(gotIndexes).To(gomega.Equal(wantIndexes), "expected completed indexes %s, but got %s", wantIndexes, gotIndexes)
 	})
 
+	/*
+		Testcase: Ensure that all indexes are executed for an indexed job with backoffLimitPerIndex despite some failing
+		Description: Create an indexed job and ensure that all indexes are either failed or succeeded, depending
+		on the end state of the corresponding pods. Pods with odd indexes fail, while the pods with even indexes
+		succeeded. Also, verify that the number of failed pods doubles the number of failing indexes, as the
+		backoffLimitPerIndex=1, allowing for one pod recreation before marking that indexed failed.
+	*/
+	ginkgo.It("should execute all indexes despite some failing when using backoffLimitPerIndex", func(ctx context.Context) {
+		ginkgo.By("Creating an indexed job with backoffLimit per index and failing pods")
+		job := e2ejob.NewTestJob("failOddSucceedEven", "with-backoff-limit-per-index", v1.RestartPolicyNever, parallelism, completions, nil, backoffLimit)
+		job.Spec.BackoffLimit = nil
+		job.Spec.BackoffLimitPerIndex = ptr.To[int32](1)
+		mode := batchv1.IndexedCompletion
+		job.Spec.CompletionMode = &mode
+		job, err := e2ejob.CreateJob(ctx, f.ClientSet, f.Namespace.Name, job)
+		framework.ExpectNoError(err, "failed to create job in namespace: %s", f.Namespace.Name)
+
+		ginkgo.By("Awaiting for the job to fail as there are failed indexes")
+		err = e2ejob.WaitForJobFailed(f.ClientSet, f.Namespace.Name, job.Name)
+		framework.ExpectNoError(err, "failed to ensure job completion in namespace: %s", f.Namespace.Name)
+
+		ginkgo.By("Verifying the Job status fields to ensure all indexes were executed")
+		job, err = e2ejob.GetJob(ctx, f.ClientSet, f.Namespace.Name, job.Name)
+		framework.ExpectNoError(err, "failed to retrieve latest job object")
+		gomega.Expect(job.Status.FailedIndexes).Should(gomega.HaveValue(gomega.Equal("1,3")))
+		gomega.Expect(job.Status.CompletedIndexes).Should(gomega.Equal("0,2"))
+		gomega.Expect(job.Status.Failed).Should(gomega.Equal(int32(4)))
+		gomega.Expect(job.Status.Succeeded).Should(gomega.Equal(int32(2)))
+	})
+
 	/*
 		Testcase: Ensure that the pods associated with the job are removed once the job is deleted
 		Description: Create a job and ensure the associated pod count is equal to parallelism count. Delete the
diff --git a/test/e2e/framework/job/fixtures.go b/test/e2e/framework/job/fixtures.go
index ca0f6af8b28..2b392d7ba42 100644
--- a/test/e2e/framework/job/fixtures.go
+++ b/test/e2e/framework/job/fixtures.go
@@ -90,6 +90,16 @@ func NewTestJobOnNode(behavior, name string, rPol v1.RestartPolicy, parallelism,
 		job.Spec.Template.Spec.Containers[0].Command = []string{"sleep", "1000000"}
 	case "fail":
 		job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "exit 1"}
+	case "failOddSucceedEven":
+		job.Spec.Template.Spec.Containers[0].Command = []string{"sh", "-c"}
+		job.Spec.Template.Spec.Containers[0].Args = []string{`
+			if [ $(expr ${JOB_COMPLETION_INDEX} % 2) -ne 0 ]; then
+				exit 1
+			else
+				exit 0
+			fi
+			`,
+		}
 	case "succeed":
 		job.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "exit 0"}
 	case "randomlySucceedOrFail":