Terminate TaskRun when Pod fails due to ImagePullBackOff.

Prior to this, if the Pod was in ImagePullBackOff state,
the TaskRun would remain `Running` with the message `Pending` until it eventually timed out.
This led to lots of delays. The expected behavior should have been to
terminate the TaskRun and set it to `fail`. This PR addresses issue
#4895.

docs/taskruns.md

@@ -560,6 +560,7 @@ False|\[Error message\]|No|The TaskRun encountered a non-permanent error, and it
 False|\[Error message\]|Yes|The TaskRun failed with a permanent error (usually validation).
 False|TaskRunCancelled|Yes|The TaskRun was cancelled successfully.
 False|TaskRunTimeout|Yes|The TaskRun timed out.
+False|TaskRunImagePullFailed|Yes|The TaskRun failed due to one of its steps not being able to pull the image. 
 
 When a `TaskRun` changes status, [events](events.md#taskruns) are triggered accordingly.
 

pkg/apis/pipeline/v1beta1/taskrun_types.go

@@ -143,6 +143,8 @@ const (
 	// TaskRunReasonResolvingTaskRef indicates that the TaskRun is waiting for
 	// its taskRef to be asynchronously resolved.
 	TaskRunReasonResolvingTaskRef = "ResolvingTaskRef"
+	// TaskRunReasonImagePullFailed is the reason set when the step of a task fails due to image not being pulled
+	TaskRunReasonImagePullFailed TaskRunReason = "TaskRunImagePullFailed"
 )
 
 func (t TaskRunReason) String() string {

pkg/reconciler/taskrun/taskrun.go

@@ -155,6 +155,17 @@ func (c *Reconciler) ReconcileKind(ctx context.Context, tr *v1beta1.TaskRun) pkg
 		return c.finishReconcileUpdateEmitEvents(ctx, tr, before, err)
 	}
 
+	// Check for Pod Failures: Image Pull Failed
+	// TODO: Might want to wrap this in its own function
+	// to capture different types of pod failures that should terminate the TaskRun.
+	for _, step := range tr.Status.Steps {
+		if step.Waiting != nil && step.Waiting.Reason == "ImagePullBackOff" {
+			message := fmt.Sprintf(`A step in TaskRun %q failed to pull the image. The pod errored with the message: "%s."`, tr.Name, step.Waiting.Message)
+			err := c.failTaskRun(ctx, tr, v1beta1.TaskRunReasonImagePullFailed, message)
+			return c.finishReconcileUpdateEmitEvents(ctx, tr, before, err)
+		}
+	}
+
 	// prepare fetches all required resources, validates them together with the
 	// taskrun, runs API conversions. In case of error we update, emit events and return.
 	_, rtr, err := c.prepare(ctx, tr)

pkg/reconciler/taskrun/taskrun_test.go

@@ -1884,6 +1884,86 @@ status:
 	}
 }
 
+func TestReconcilePodFailures(t *testing.T) {
+	type testCase struct {
+		name           string
+		taskRun        *v1beta1.TaskRun
+		expectedStatus *apis.Condition
+		wantEvents     []string
+	}
+
+	testcases := []testCase{
+		{
+			name: "taskrun with pod image pull failure",
+			taskRun: parse.MustParseTaskRun(t, `
+metadata:
+  name: test-imagepull-fail
+  namespace: foo
+spec:
+  taskSpec:
+    steps:
+    - image: whatever
+      name: ""
+      resources: {}
+      script: "true"
+  timeout: 1h0m0s
+status:
+  steps:
+  - container: step-unnamed-0
+    name: unnamed-0
+    waiting:
+      message: Back-off pulling image "whatever"
+      reason: ImagePullBackOff
+  taskSpec:
+    steps:
+    - image: whatever
+      name: ""
+      resources: {}
+      script: "true"
+`),
+
+			expectedStatus: &apis.Condition{
+				Type:    apis.ConditionSucceeded,
+				Status:  corev1.ConditionFalse,
+				Reason:  "TaskRunImagePullFailed",
+				Message: `A step in TaskRun "test-imagepull-fail" failed to pull the image. The pod errored with the message: "Back-off pulling image "whatever"."`,
+			},
+			wantEvents: []string{
+				"Normal Started ",
+				`Warning Failed A step in TaskRun "test-imagepull-fail" failed to pull the image. The pod errored with the message: "Back-off pulling image "whatever".`,
+			},
+		}}
+
+	for _, tc := range testcases {
+		t.Run(tc.name, func(t *testing.T) {
+			d := test.Data{
+				TaskRuns: []*v1beta1.TaskRun{tc.taskRun},
+				// Tasks:    []*v1beta1.Task{simpleTask, saTask},
+			}
+			testAssets, cancel := getTaskRunController(t, d)
+			defer cancel()
+			c := testAssets.Controller
+			clients := testAssets.Clients
+
+			if err := c.Reconciler.Reconcile(testAssets.Ctx, getRunName(tc.taskRun)); err != nil {
+				t.Fatalf("Unexpected error when reconciling completed TaskRun : %v", err)
+			}
+			newTr, err := clients.Pipeline.TektonV1beta1().TaskRuns(tc.taskRun.Namespace).Get(testAssets.Ctx, tc.taskRun.Name, metav1.GetOptions{})
+			if err != nil {
+				t.Fatalf("Expected completed TaskRun %s to exist but instead got error when getting it: %v", tc.taskRun.Name, err)
+			}
+			condition := newTr.Status.GetCondition(apis.ConditionSucceeded)
+			if d := cmp.Diff(tc.expectedStatus, condition, ignoreLastTransitionTime); d != "" {
+				t.Fatalf("Did not get expected condition %s", diff.PrintWantGot(d))
+			}
+			err = eventstest.CheckEventsOrdered(t, testAssets.Recorder.Events, tc.taskRun.Name, tc.wantEvents)
+			if !(err == nil) {
+				t.Errorf(err.Error())
+			}
+		})
+	}
+}
+
 func TestReconcileTimeouts(t *testing.T) {
 	type testCase struct {
 		name           string