Switch PipelineRun timeout -> TaskRun logic to instead signal the Tas…

…kRuns to stop

Fixes #5127

As noted in #5127, the logic around calculating a timeout for a `PipelineRun`'s `TaskRun` to make sure that the `TaskRun`'s timeout is always going to end before the `PipelineRun`'s timeout ends is problematic. It can result in race conditions where a `TaskRun` gets timed out, immediately followed by the `PipelineRun` being reconciled while not yet having hit the end of its own timeout. This change gets rid of that behavior, and instead sets the `TaskRun.Spec.Status` to a new value, `TaskRunTimedOut`, with the `TaskRun` reconciler handling that in the same way it does setting `TaskRun.Spec.Status` to `TaskRunCancelled`.

By doing this, we can unify the behavior for both `TaskRun`s and `Run`s upon `PipelineRun`s timing out, given that we already cancel `Run`s upon `PipelineRun` timeout, and we can kill off a bunch of flaky outcomes for `PipelineRun`s.

Huge thanks to @jerop for suggesting this approach!

Signed-off-by: Andrew Bayer <andrew.bayer@gmail.com>

pkg/apis/pipeline/v1beta1/taskrun_types.go

@@ -88,6 +88,9 @@ const (
 	// TaskRunSpecStatusCancelled indicates that the user wants to cancel the task,
 	// if not already cancelled or terminated
 	TaskRunSpecStatusCancelled = "TaskRunCancelled"
+	// TaskRunSpecStatusTimedOut indicates that the PipelineRun owning this task wants to mark it as timed out,
+	// if not already cancelled or terminated
+	TaskRunSpecStatusTimedOut = "TaskRunTimedOut"
 )
 
 // TaskRunDebug defines the breakpoint config for a particular TaskRun
@@ -424,6 +427,11 @@ func (tr *TaskRun) IsCancelled() bool {
 	return tr.Spec.Status == TaskRunSpecStatusCancelled
 }
 
+// IsSpecTimedOut returns true if the TaskRun's spec status is set to TimedOut state
+func (tr *TaskRun) IsSpecTimedOut() bool {
+	return tr.Spec.Status == TaskRunSpecStatusTimedOut
+}
+
 // HasTimedOut returns true if the TaskRun runtime is beyond the allowed timeout
 func (tr *TaskRun) HasTimedOut(ctx context.Context, c clock.PassiveClock) bool {
 	if tr.Status.StartTime.IsZero() {

pkg/apis/pipeline/v1beta1/taskrun_validation.go

@@ -79,8 +79,8 @@ func (ts *TaskRunSpec) Validate(ctx context.Context) (errs *apis.FieldError) {
 	}
 
 	if ts.Status != "" {
-		if ts.Status != TaskRunSpecStatusCancelled {
-			errs = errs.Also(apis.ErrInvalidValue(fmt.Sprintf("%s should be %s", ts.Status, TaskRunSpecStatusCancelled), "status"))
+		if ts.Status != TaskRunSpecStatusCancelled && ts.Status != TaskRunSpecStatusTimedOut {
+			errs = errs.Also(apis.ErrInvalidValue(fmt.Sprintf("%s should be %s or %s", ts.Status, TaskRunSpecStatusCancelled, TaskRunSpecStatusTimedOut), "status"))
 		}
 	}
 	if ts.Timeout != nil {

pkg/apis/pipeline/v1beta1/taskrun_validation_test.go

@@ -188,7 +188,7 @@ func TestTaskRunSpec_Invalidate(t *testing.T) {
 			},
 			Status: "TaskRunCancell",
 		},
-		wantErr: apis.ErrInvalidValue("TaskRunCancell should be TaskRunCancelled", "status"),
+		wantErr: apis.ErrInvalidValue("TaskRunCancell should be TaskRunCancelled or TaskRunTimedOut", "status"),
 	}, {
 		name: "invalid taskspec",
 		spec: v1beta1.TaskRunSpec{

pkg/reconciler/pipelinerun/pipelinerun.go

@@ -110,6 +110,9 @@ const (
 	// ReasonCouldntCancel indicates that a PipelineRun was cancelled but attempting to update
 	// all of the running TaskRuns as cancelled failed.
 	ReasonCouldntCancel = "PipelineRunCouldntCancel"
+	// ReasonCouldntTimeOut indicates that a PipelineRun was timed out but attempting to update
+	// all of the running TaskRuns as timed out failed.
+	ReasonCouldntTimeOut = "PipelineRunCouldntTimeOut"
 	// ReasonInvalidTaskResultReference indicates a task result was declared
 	// but was not initialized by that task
 	ReasonInvalidTaskResultReference = "InvalidTaskResultReference"
@@ -579,6 +582,13 @@ func (c *Reconciler) reconcile(ctx context.Context, pr *v1beta1.PipelineRun, get
 	// Reset the skipped status to trigger recalculation
 	pipelineRunFacts.ResetSkippedCache()
 
+	// If the pipelinerun has timed out, mark tasks as timed out and update status
+	if pr.HasTimedOut(ctx, c.Clock) {
+		if err := timeoutPipelineRun(ctx, logger, pr, c.PipelineClientSet); err != nil {
+			return err
+		}
+	}
+
 	after := pipelineRunFacts.GetPipelineConditionStatus(ctx, pr, logger, c.Clock)
 	switch after.Status {
 	case corev1.ConditionTrue:
@@ -621,7 +631,7 @@ func (c *Reconciler) processRunTimeouts(ctx context.Context, pr *v1beta1.Pipelin
 	}
 	for _, rpt := range pipelineState {
 		if rpt.IsCustomTask() {
-			if rpt.Run != nil && !rpt.Run.IsCancelled() && (pr.HasTimedOut(ctx, c.Clock) || (rpt.Run.HasTimedOut(c.Clock) && !rpt.Run.IsDone())) {
+			if rpt.Run != nil && !rpt.Run.IsCancelled() && rpt.Run.HasTimedOut(c.Clock) && !rpt.Run.IsDone() {
 				logger.Infof("Cancelling run task: %s due to timeout.", rpt.RunName)
 				err := cancelRun(ctx, rpt.RunName, pr.Namespace, c.PipelineClientSet)
 				if err != nil {
@@ -1141,12 +1151,10 @@ func calculateTaskRunTimeout(timeout time.Duration, pr *v1beta1.PipelineRun, rpt
 		if pElapsedTime > timeout {
 			return &metav1.Duration{Duration: 1 * time.Second}
 		}
-		timeRemaining := (timeout - pElapsedTime)
-		// Return the smaller of timeRemaining and rpt.pipelineTask.timeout
-		if rpt.PipelineTask.Timeout != nil && rpt.PipelineTask.Timeout.Duration < timeRemaining {
+		if rpt.PipelineTask.Timeout != nil && rpt.PipelineTask.Timeout.Duration < timeout {
 			return &metav1.Duration{Duration: rpt.PipelineTask.Timeout.Duration}
 		}
-		return &metav1.Duration{Duration: timeRemaining}
+		return &metav1.Duration{Duration: timeout}
 	}
 
 	if timeout == apisconfig.NoTimeoutDuration && rpt.PipelineTask.Timeout != nil {

pkg/reconciler/pipelinerun/pipelinerun_test.go

@@ -1736,16 +1736,44 @@ spec:
     name: test-pipeline
   serviceAccountName: test-sa
 status:
+  conditions:
+  - message: running...
+    reason: Running
+    status: Unknown
+    type: Succeeded
   startTime: "2021-12-31T00:00:00Z"
+  runs:
+    test-pipeline-run-custom-task-hello-world-1:
+      pipelineTaskName: hello-world-1
+      status:
+        conditions:
+        - status: Unknown
+          type: Succeeded
 `)}
 			prs[0].Spec.Timeout = tc.timeout
 			prs[0].Spec.Timeouts = tc.timeouts
 
+			runs := []*v1alpha1.Run{mustParseRunWithObjectMeta(t,
+				taskRunObjectMeta("test-pipeline-run-custom-task-hello-world-1", "test", "test-pipeline-run-custom-task",
+					"test-pipeline", "hello-world-1", true),
+				`
+spec:
+  ref:
+    apiVersion: example.dev/v0
+    kind: Example
+status:
+  conditions:
+  - status: Unknown
+    type: Succeeded
+  startTime: "2021-12-31T11:58:59Z"
+`)}
+
 			cms := []*corev1.ConfigMap{withCustomTasks(newFeatureFlagsConfigMap())}
 			d := test.Data{
 				PipelineRuns: prs,
 				Pipelines:    ps,
 				ConfigMaps:   cms,
+				Runs:         runs,
 			}
 			prt := newPipelineRunTest(d, t)
 			defer prt.Cancel()
@@ -1763,9 +1791,9 @@ status:
 			}
 
 			gotTimeoutValue := postReconcileRun.GetTimeout()
-			expectedTimeoutValue := time.Second
+			expectedTimeoutValue := time.Hour
 
-			if d := cmp.Diff(gotTimeoutValue, expectedTimeoutValue); d != "" {
+			if d := cmp.Diff(expectedTimeoutValue, gotTimeoutValue); d != "" {
 				t.Fatalf("Expected timeout for created Run, but got a mismatch %s", diff.PrintWantGot(d))
 			}
 
@@ -2755,6 +2783,120 @@ spec:
 	}
 }
 
+func TestReconcileFailsTaskRunTimeOut(t *testing.T) {
+	prName := "test-pipeline-fails-to-timeout"
+
+	// TestReconcileFailsTaskRunTimeOut runs "Reconcile" on a PipelineRun with a single TaskRun.
+	// The TaskRun cannot be timed out. Check that the pipelinerun timeout fails, that reconcile fails and
+	// an event is generated
+	names.TestingSeed()
+	prs := []*v1beta1.PipelineRun{parse.MustParsePipelineRun(t, `
+metadata:
+  name: test-pipeline-fails-to-timeout
+  namespace: foo
+spec:
+  pipelineRef:
+    name: test-pipeline
+  timeout: 1h0m0s
+status:
+  conditions:
+  - message: running...
+    reason: Running
+    status: Unknown
+    type: Succeeded
+  startTime: "2021-12-31T22:59:00Z"
+  taskRuns:
+    test-pipeline-fails-to-timeouthello-world-1:
+      pipelineTaskName: hello-world-1
+`)}
+	ps := []*v1beta1.Pipeline{parse.MustParsePipeline(t, `
+metadata:
+  name: test-pipeline
+  namespace: foo
+spec:
+  tasks:
+  - name: hello-world-1
+    taskRef:
+      name: hello-world
+  - name: hello-world-2
+    taskRef:
+      name: hello-world
+`)}
+	tasks := []*v1beta1.Task{simpleHelloWorldTask}
+	taskRuns := []*v1beta1.TaskRun{
+		getTaskRun(
+			t,
+			"test-pipeline-fails-to-timeouthello-world-1",
+			prName,
+			"test-pipeline",
+			"hello-world",
+			corev1.ConditionUnknown,
+		),
+	}
+
+	cms := []*corev1.ConfigMap{withEnabledAlphaAPIFields(newFeatureFlagsConfigMap())}
+
+	d := test.Data{
+		PipelineRuns: prs,
+		Pipelines:    ps,
+		Tasks:        tasks,
+		TaskRuns:     taskRuns,
+		ConfigMaps:   cms,
+	}
+
+	testAssets, cancel := getPipelineRunController(t, d)
+	defer cancel()
+	c := testAssets.Controller
+	clients := testAssets.Clients
+	failingReactorActivated := true
+
+	// Make the patch call fail, i.e. make it so that the controller fails to cancel the TaskRun
+	clients.Pipeline.PrependReactor("patch", "taskruns", func(action ktesting.Action) (bool, runtime.Object, error) {
+		return failingReactorActivated, nil, fmt.Errorf("i'm sorry Dave, i'm afraid i can't do that")
+	})
+
+	err := c.Reconciler.Reconcile(testAssets.Ctx, "foo/test-pipeline-fails-to-timeout")
+	if err == nil {
+		t.Errorf("Expected to see error returned from reconcile after failing to timeout TaskRun but saw none!")
+	}
+
+	// Check that the PipelineRun is still running with correct error message
+	reconciledRun, err := clients.Pipeline.TektonV1beta1().PipelineRuns("foo").Get(testAssets.Ctx, "test-pipeline-fails-to-timeout", metav1.GetOptions{})
+	if err != nil {
+		t.Fatalf("Somehow had error getting reconciled run out of fake client: %s", err)
+	}
+
+	if val, ok := reconciledRun.GetLabels()[pipeline.PipelineLabelKey]; !ok {
+		t.Fatalf("expected pipeline label")
+	} else if d := cmp.Diff("test-pipeline", val); d != "" {
+		t.Errorf("expected to see pipeline label. Diff %s", diff.PrintWantGot(d))
+	}
+
+	// The PipelineRun should not be timed out b/c we couldn't timeout the TaskRun
+	checkPipelineRunConditionStatusAndReason(t, reconciledRun, corev1.ConditionUnknown, ReasonCouldntTimeOut)
+	// The event here is "Normal" because in case we fail to timeout we leave the condition to unknown
+	// Further reconcile might converge then the status of the pipeline.
+	// See https://github.com/tektoncd/pipeline/issues/2647 for further details.
+	wantEvents := []string{
+		"Normal PipelineRunCouldntTimeOut PipelineRun \"test-pipeline-fails-to-timeout\" was timed out but had errors trying to time out TaskRuns and/or Runs",
+		"Warning InternalError 1 error occurred",
+	}
+	err = eventstest.CheckEventsOrdered(t, testAssets.Recorder.Events, prName, wantEvents)
+	if err != nil {
+		t.Errorf(err.Error())
+	}
+
+	// Turn off failing reactor and retry reconciliation
+	failingReactorActivated = false
+
+	err = c.Reconciler.Reconcile(testAssets.Ctx, "foo/test-pipeline-fails-to-timeout")
+	if err == nil {
+		// No error is ok
+	} else if ok, _ := controller.IsRequeueKey(err); !ok { // Requeue is also fine.
+		t.Errorf("Expected to timeout TaskRun successfully!")
+	}
+}
+
 func TestReconcilePropagateLabelsAndAnnotations(t *testing.T) {
 	names.TestingSeed()
 
@@ -3094,7 +3236,7 @@ status:
 			reconciledRun, _ := prt.reconcileRun("foo", "test-pipeline-retry-run-with-timeout", []string{}, false)
 
 			if len(reconciledRun.Status.TaskRuns["hello-world-1"].Status.RetriesStatus) != tc.retries {
-				t.Fatalf(" %d retry expected but %d ", tc.retries, len(reconciledRun.Status.TaskRuns["hello-world-1"].Status.RetriesStatus))
+				t.Fatalf(" %d retries expected but got %d ", tc.retries, len(reconciledRun.Status.TaskRuns["hello-world-1"].Status.RetriesStatus))
 			}
 
 			if status := reconciledRun.Status.TaskRuns["hello-world-1"].Status.GetCondition(apis.ConditionSucceeded).Status; status != tc.conditionSucceeded {
@@ -3240,7 +3382,7 @@ func TestGetTaskRunTimeout(t *testing.T) {
 				},
 			},
 		},
-		expected: &metav1.Duration{Duration: 10 * time.Minute},
+		expected: &metav1.Duration{Duration: 20 * time.Minute},
 	}, {
 		name: "taskrun with elapsed time; task.timeout applies",
 		timeoutFields: &v1beta1.TimeoutFields{
@@ -3259,7 +3401,7 @@ func TestGetTaskRunTimeout(t *testing.T) {
 				},
 			},
 		},
-		expected: &metav1.Duration{Duration: 10 * time.Minute},
+		expected: &metav1.Duration{Duration: 15 * time.Minute},
 	}, {
 		name: "taskrun with elapsed time; timeouts.pipeline applies",
 		timeoutFields: &v1beta1.TimeoutFields{
@@ -3278,7 +3420,7 @@ func TestGetTaskRunTimeout(t *testing.T) {
 				},
 			},
 		},
-		expected: &metav1.Duration{Duration: 10 * time.Minute},
+		expected: &metav1.Duration{Duration: 15 * time.Minute},
 	}}
 
 	for _, tc := range tcs {
@@ -3296,7 +3438,7 @@ func TestGetTaskRunTimeout(t *testing.T) {
 					},
 				},
 			}
-			if d := cmp.Diff(getTaskRunTimeout(context.TODO(), pr, tc.rpt, testClock), tc.expected); d != "" {
+			if d := cmp.Diff(tc.expected, getTaskRunTimeout(context.TODO(), pr, tc.rpt, testClock)); d != "" {
 				t.Errorf("Unexpected task run timeout. Diff %s", diff.PrintWantGot(d))
 			}
 		})
@@ -3449,7 +3591,7 @@ func TestGetFinallyTaskRunTimeout(t *testing.T) {
 				},
 			},
 		},
-		expected: &metav1.Duration{Duration: 11 * time.Minute},
+		expected: &metav1.Duration{Duration: 15 * time.Minute},
 	}}
 
 	for _, tc := range tcs {